Backprojection (BP) algorithm is successfully used to focus the layered scenes using ground-penetrating radar data. However, the BP algorithm with the time-consuming refraction point computation is impractical to autofocus in the layered scenes lacking medium information, where iterative focusing for estimating information continuously accumulates the calculation time of refraction points. Therefore, we propose a refraction point estimation (RPE)-based BP algorithm for autofocusing technique in unknown layered scenes. Our RPE algorithm, including two steps, directly expresses the refraction point. The first step coarsely estimates an initial position using an adjustment factor. Subsequently, the second step precisely converts the quartic equation of the refraction point into a linear form subject to the coarse estimation. Following this, we propose an RPE-based BP autofocusing technique to efficiently complete information estimation and refocusing. Compared with the existing algorithms, the proposed RPE algorithm directly estimates the refraction point and avoids iteration. The results indicate that the RPE-based BP algorithm has significant advantages in efficiency and accuracy, prompting the practical application of the BP autofocusing in unknown layered scenes.

Hyperspectral unmixing addressing spectral variability remains an important challenge. In this field, unmixing methods do not exploit the possible availability of some spectral information that corresponds to known spectra of some pure materials present in an acquired scene. In this work, a hyperspectral unmixing method, which considers not only the spectral variability phenomenon but also exploits one or more available known pure material spectra, is proposed. Such a combination, initially proposed here, constitutes the originality of the conducted work that distinguishes it from other investigations in the hyperspectral unmixing topic. The proposed method, based on an informed nonnegative matrix factorization technique, employs a partial structured additively-tuned linear mixing model that deals with spectral variability. Experimental results, based on real data, show that the designed informed algorithm, which addresses spectral variability, yields very satisfactory results and outperforms tested literature approaches. Thus, such an unmixing algorithm may be used for automatically detecting and mapping, using hyperspectral data, materials of interest whose spectra are known while dealing with their spectral variability.

Advanced Spaceborne Thermal Emission and Reflection Radiometer L1T data were used for the derivation of the new indices for the identification and mapping of the silica-rich and mafic-ultramafic rocks of the Udaipur-Rakhabdev region. The newly derived silica-rich rock index Ronak Jain (QI-RJ) was used in four bands except band 10 as [((band 13 – sqrt(band 11))/band 14)/band 12]. The mafic-rich rocks were also defined by a new index (MI-RJ), as (band 10 + band 14) / band 13. The new indices had advantages in accurate mapping compared with those given by Ninomiya (QI-N), Rockwell and Hofstra (QI-RH), and Guha and Vinod Kumar (QI-AV) for mapping of silica-rich rocks by removing striping noise. Similarly, mafic minerals were mapped by indices given by Ninomiya (MI-N) and Guha and Vinod Kumar (MI-AV) for ultramafics, but they have not mapped the discontinuous outcrops of serpentinites. The statistical regression analysis approach was used for the prediction of the goodness of fit between the new indices and already existing indices. The silica index (QI-RJ) is highly correlated with the QI-RH and QI-AV and represents low values for QI-N. In a similar way, MI-RJ shows a high correlation with the MI-N and MI-AV indices. The regression analyses for the QI-RJ with MI-RJ, MI-N, and MI-AV, show a negative trend but have high values of regression correlation. The generated maps from the newly defined index were also compared with the indices maps generated from the existing indices and were found to have better maps of litho-units. Field collected data and petrological studies also validated the presence of silica-rich rocks and ultramafics of the region mapped with the help of newly defined indices, and a better spatial accuracy of maps was obtained. The new mafic index was more suitable for mapping serpentinites and their alteration products exposed in the area.

Raghunandan Hills Reserve is an important protected area in Bangladesh that supports some remnant patches of natural forest and is the habitat of several globally threatened primates including Western Hoolock Gibbon, Northern Pig-tailed Macaque, and Capped Langur. However, deforestation and forest degradation due to anthropogenic factors, such as illegal logging and fuelwood collection are age-old problems at Raghunandan. The areas of the reserve vulnerable to future conversions due to the possible proximate or underlaying causes were unknown. This study analyzed the historical trend of forest and land-use/landcover transitions at Raghunandan Hills Reserve from 1995 until 2015 at a 10-year interval using Monte Carlo spectral unmixing and knowledge-based classification approaches to Landsat satellite images in Claslite and ArcGIS software. Based on the past trend, it then predicted the future trend of forest land-use/landcover transitions for 2025 and 2035 using an artificial multi-layer perceptron neural network with Markov Chain machine learning algorithm integrated into the land change modeler module of IDRISI/TerrSet software. Results indicated that ∼30  %   to 35% of the total area of the reserve was covered by forest, which included patches of natural forest and plantations, whereas the remaining area was occupied by non-forest categories like scattered degraded forests, grasses, and shrubs. Forest cover declined during 1995–2005, and then increased slightly during 2005–2015 due to afforestation activities. This trend is likely to continue in the future with forest cover occupying nearly 40% of the reserve by 2025 and 2035. Along with identifying the areas where the forest is likely to be expanded, the areas of the reserve vulnerable to deforestation (hotspots) were also highlighted and quantified in the form of maps and statistics. The findings have useful implications for any forest conservation initiatives including the global climate change mitigation program reducing emissions from deforestation and forest degradation+, which requires identifying at-risk areas of planned and unplanned deforestation.

To quantify the variation in the ecosystem service values (ESV) in response to land use and land cover changes in the Lake Pátzcuaro watershed, the third largest lake in Mexico, Landsat images from the period 1986–2022 were analyzed and a benefits transfer method was used. Forest class and agricultural soils were the dominant covers occupying an average of 42% and 25% of the study area, respectively. The most evident change was a reduction of 2550 ha of the Lake Pátzcuaro surface. Further, with the natural vegetation increasing gradually due to the abandonment of agricultural lands, the ESV also increased from 279.1 to 302.0 million dollars from 1986–2022. These results suggest a positive influence on the protection of natural environments, and although it was not evaluated here but given the lake area reduction, they highlight the need to avoid detrimental actions that affect the biota inhabiting it.

Hyperspectral image (HSI) classification is a procedure of interest in remote sensing. HSIs contain complex spectral and spatial information, so classification tasks remain difficult. Although current deep-learning models have made significant progress in HSI classification, dealing with spectral and spatial information still requires careful investigation. To better manage spectral and spatial information and improve classification accuracy, we introduce a multiscale residual weakly dense network with an attention mechanism. First, we designed two residual weakly dense (Res-WDens) branches to extract spectral and spatial feature information and then applied the Concat method to fuse the two kinds of information. We also designed a plug-and-play hybrid attention module to refine the fused information so the network could focus on the essential spectral and spatial features. Finally, considering the relevance of spectral and spatial information, a dual-channel multiscale feature extraction module was used to extract the spectral–spatial multiscale information of HSIs. The overall accuracies of our proposed method reached 99.76%, 99.97%, and 100% on three publicly available datasets. A series of experiments demonstrated that our method is comparable to current state-of-the-art methods.

Ground-based and spaceborne interferometric synthetic aperture radar (InSAR) were combined to monitor slope deformation and analyze the main influencing factors of an unstable slope on the right bank of the Dajinchuan River in Danba County, Sichuan Province, China. We applied the small baseline subset time series strategy with 656 ground-based radar scenes acquired in September 13 to 17, 2019, and 62 Sentinel-1 InSAR scenes from July, 2018, to October, 2020. The qualitative and quantitative deformation features of the unstable slope were studied, and the possible influencing factors contributing to slope instability, including the geological structure, topography, external environment, and human activities, were discussed. The area of maximum deformation detected by the ground-based radar is located in the bedrock above the target area with a maximum cumulative deformation of more than 30 mm during the observation period, whereas the maximum deformation rate detected by spaceborne InSAR exceeds 40 mm/year. The differences between these results were analyzed, and the reasons were discussed. We provide references and suggestions for investigating potential landslide risks by combining ground-based with spaceborne InSAR.

We aim to investigate the relation between advanced spaceborne thermal emission and reflection radiometer (ASTER) and thermal infrared (TIR) data with the hydrothermal alteration zones within the hydrothermal field of the caldera of Nisyros volcano in Greece. This is achieved by utilizing clustering on a dataset of five ASTER L1T TIR images spanning the period 2003 to 2007. Each one of these images is converted to temperature images, coregistered, and vegetation masked, creating a five-band ASTER temperature image. In the sequel, probabilistic clustering hypothesis is applied to the five-band vegetation masked ASTER temperature image, subset to the area of the hydrothermal field. The probabilistic clustering algorithm identifies three “thermal” clusters that exhibit a satisfactory degree of agreement with the active hydrothermal alteration field of Nisyros. Each cluster corresponds to a different degree of hydrothermal alteration zone. We show the existence of a middle temperature zone (cluster), which exhibits significant overlap with the middle hydrothermal alteration zone, along with a high-temperature zone and a low-temperature zone. This temperature transition from high-temperature cluster to the middle and gradually to low-temperature cluster for the time-spanning of the data is significant as the hydrothermal field of Nisyros was very active during this period. This finding coincides temporally and spatially with the continuous N–S trending (Lakki rupture).

The use of optical remote sensing to monitor short-term postfire dynamics in rapidly recovering ecosystems is a complex task, requiring strong spectral separability and high temporal resolution. Thus, characterizing the effects of temporal lag in image acquisitions in relation to fire occurrence is of fundamental importance to improve our understanding of how fire-related environmental and ecological processes are connected to multitemporal spectral information. We used a harmonized series of Sentinel-2A/2B and Landsat-7/8 images paired with a fire experiment specifically designed for remote sensing assessment to evaluate short-term spectral responses of fire in tropical savannas. Our experimental design included strategic geolocation of experimental plots and separate prescribed burnings during the early- and mid-dry seasons. The M-statistic was used to assess spectral separability between burned and control treatments, and the temporal stability was assessed using fire scar latency, a custom statistic adapted to capture scar spectral responses over time. Finally, linear regressions were fitted to explore the association between consumed fuel load and spectral indices. The variation of normalized burn ratio and mid-infrared bispectral index indices stood out with the highest separability (M-statistic > 3) in both treatments and moderate-high association with consumed fuel load (R² > 0.54). We observed that burned areas spectral sensitivity depends not only on timing of image acquisition but also on fire season and their interactions with land surface phenology. Our characterization provides a solid benchmark for future experimental assessments of fire spectral responses and provides new insights for improving fire monitoring algorithms.

Coastal lagoons are highly productive ecosystems that offer a range of natural services for the environment and society. However, anthropic pressure in their basins cause water quality loss and endangers their ecosystem. Monitoring water quality constituent by remote sensing can provide spatial and high temporal resolution analysis. Not recognizing how depth influences on the signal received by a sensor causes on significant errors in estimating water constituent’s concentrations. Using a bathymetry survey and Landsat-5 thematic mapper (TM) images, different patterns of depth influence on reflectance were recognized. By means of a statistical analysis, it was able to classify and analyze three different regions of homogeneous spectral behavior for the closest image from the bathymetric survey and for a median image from 2001 to 2011. Nonparametric Kruskal–Wallis test and Dunn’s post hoc test provided great values in differentiating all groups generated for the coastal lagoon to both closest and median image. The methodology proposed here, being based on statistical methods, can be applied with the use of other images to assist studies in coastal lagoons across the globe.

Coastal areas are constantly changing and are influenced by a variety of variables, including environmental, ecological, geological, and anthropological factors. The plethora of coastal change has created substantial challenges for coastal land infrastructures, animal habitat, vegetation conservation, and monitoring. It is important to track these changes, monitor, predict, and preserve the future of coastal areas. Shoreline mapping has mostly concentrated on two-dimensional (2D) distance changes along the coast due to the lack of coastal elevational data and the precise information required to construct the three-dimensional (3D) surface. Modern technology has covered the gaps by developing a range of sensors and algorithms for data analysis on 3D coastal erosion and accretion changes throughout the beach topography, which has become crucial. The ultimate goal is to develop a cost-effective mapping procedure for an area that includes a high-resolution mapping system that abstracts evidence of coastal areas for long-term 2D change rates from 1953 to 2020 and short-term 3D volumetric changes of sand erosion and accretion rates from 2018 to 2020. This would be accomplished through Unmanned Aircraft Systems Mapping, which employs digital photogrammetry principles including comprehensive geospatial information systems spatial data and linear regression analysis to investigate massive quantities of data. The feasibility of this methodology was applied to the Jupiter Inlet Lighthouse Outstanding Natural Area in Jupiter, Florida, a study coastal area that had receded almost 27 m in the last 67 years. The outcomes of a short-term coastal analysis quantify the amount of soil erosion and accretion that occurred along the shore over two years as a result of rising sea levels. The long-term change analysis indicates how much sand was eroded each year as a result of several variables affecting shoreline changes, which will help the land management team in developing new sustainable policies for future coastal environmental conservation.

Oriented object detection is a challenging task in aerial images since the objects in aerial images are displayed in arbitrary directions and are frequently densely packed. The mainstream detectors describe rotating objects using a five-parameter or eight-parameter representations, which have representation ambiguity for orientated object definition and border loss discontinuity in the regression process. We proposed an innovative representation method based on area ratio of parallelogram, called area ratio of parallelogram (ARP). Specifically, ARP regresses the minimum bounding rectangle of the oriented object and three area ratios. Three area ratios include the area ratio of a directed object to the smallest circumscribed rectangle and two parallelograms to the minimum circumscribed rectangle. It simplifies offset learning and eliminates the issue of angular periodicity or label point sequences for oriented objects. To further remedy the confusion issue of nearly horizontal objects, the area ratio between the object and its minimal circumscribed rectangle is employed to guide the selection of horizontal or oriented detection for each object. Moreover, the rotated efficient intersection over union loss with horizontal bounding box and three area ratios are designed to optimize the bounding box regression for rotating objects. Experimental results on remote sensing datasets, including HRSC2016, DOTA, and UCAS-AOD, show that the proposed method achieves superior detection performance than many state-of-the-art approaches.

Accurate farmland segmentation is essential for modern agriculture and automated navigation. We propose an improved U-Net for farmland area segmentation. The wheat farmland data images were collected at the winter wheat experimental base of the Institute of Agricultural Economics and Information, Henan Academy of Agricultural Sciences. U-Net adopts the encoder–decoder structure and skips connection to achieve segmentation. The downsampling operation in the encoder stage weakens the detailed features. The semantic gap between the decoder and the encoder will cause the sparse wheat seedlings in the farmland cannot be captured. Based on the above problems, the improved U-Net uses a multiscale global attention module (MGA) in the bottleneck layer. MGA forms enhanced features by aggregating multiscale global context information and using an improved attention mechanism. An interaction mechanism (IM) is added between the decoder and the encoder. The encoder–decoder IM concatenates multiple attention units and fuses them with the original features on the encoder side to update the input features to the encoder. To lighten the model, we also define two multiplexed convolution kernel sequences in the code, which are shared by all encoders or decoders. The method proposed in this paper is evaluated on the farmland segmentation dataset. Significantly better segmentation results are achieved compared to classical models (U-Net, U-Net++, PSPNet, FPN, and DeepLabV3). In the case of obtaining similar segmentation results, with a smaller amount of parameters compared with State Of The Art (U-Net3+, ACSNet, PraNet, and CCBANet). We also use the farmland data provided by Sichuan Agricultural University for testing, the Dice is 93.88%, which has good generalization performance.

Power line extraction is useful for low aircraft altitude obstacle avoidance, and it also plays an important role in unmanned aerial vehicle (UAV)-based power autonomous inspection. However, the current power line detection algorithms have the disadvantages of slow detection speed and poor detection performance because the scale of objects varies greatly in UAV images. We proposed an efficient power line extraction method named power line extraction network (PLENet) to achieve high accuracy (ACC) and speed of power line detection in complex scenes. The proposed method overcomes the imbalance of the positive and negative samples, detects the power lines in images with different scales, and achieves good ACC and speed performance at the same time. For this method, we introduced convolution kernels of various sizes in the input layer and concatenated the extracted features to obtain more robust feature maps. Then, we implemented the class-balanced cross-entropy loss function to overcome the problem of the imbalance between positive and negative samples. Finally, we proposed a bilinear interpolation upsampling module in the output layer to replace the deconvolution to solve the gridding effect. To evaluate the effectiveness of the proposed method, we conducted a series of experiments on our collected aerial imagery dataset with complex scenes. The ACC and intersection over union of the proposed method are up to 86.22% and 71.7% on the server, respectively, whereas 86.61% and 70.58% on the mobile terminals, respectively. The experimental results show that our PLENet has apparent advantages in multiple indicators compared with other networks. The multiscene experiment results also demonstrate that PLENet has strong robustness, overcoming the influence of complex scenes and accurately extracting power lines.

Coastal dunes are fragile ecosystems emerging at the interface between marine and continental environments. They provide multiple services, among which are the protection against the impact of storms and the hosting of diverse and unique species of fauna and flora. However, changes in the topography or biological component of these systems may endanger the perpetuation of service provision. Topographic changes within dunes can significantly differ in magnitude depending on the type of process (i.e., marine or aeolian) and the temporal scale of analysis (event to annual scale), making their monitoring a challenging task. In recent years, unmanned aerial vehicles (UAVs) have been increasingly used to monitor coastal dunes, proving to be a cost-efficient methodology for the collection of topographic data. Yet, the application of UAVs in combination with the structure from motion approach to obtain digital surface models (DSMs) presents some limitations related to the level of accuracy provided for the evaluation of topographical changes in dunes with low sedimentation rates. This work explores different survey configurations using UAVs flying at low altitudes with the aim of obtaining high-quality DSMs with vertical accuracies preferably around or lower than 0.04 m. Several tests were performed to evaluate the influence of different parameters on the accuracy of the DSM, including flight altitude and orientation, density and spatial distribution of ground control points (GCPs), terrain slope, vegetation cover, and sun-related parameters. The results indicate that the intended accuracies can be obtained by combining overlapped perpendicular flights, GCPs distributed regularly following a diamond grid, with densities of at least 6 GCPs per hectare, sun altitudes between 30 and 40 deg, and a total solar radiation per hour between 1750 and 2250 KJ / m². In addition, better results were obtained across gentle slope areas, suggesting the eventual need to adapt to the particularities of each site to ensure the accuracy.

New radar altimeters have supported elevation modeling with greater accuracy and extended periods in recent years. Sentinel-3 has been rarely used in previous studies for monitoring the Antarctic ice sheet. The CryoSat-2 and Sentinel-3 satellite radar altimeter data were combined to extract the elevation changes in the Antarctic ice sheet. First, a subregional data filtering method based on a clustering algorithm was proposed to improve the accuracy of the elevation change results for the outliers in the original measurements. Next, an improved model-fitting model that incorporated multiple parameters was used to extract elevation changes, and the results were validated using airborne laser altimetry data and ICESat-2 data. The average changes in the Antarctic ice sheet elevation from 2016 to 2019 were −4.3 ± 0.9 cm / y. The elevation changes of the ice shelf edges with large slopes and complex topography were much larger than the interior. Some specific areas showed substantial elevation changes. The most significant surface decreases occurred at the Pine Island Glacier and the Totten Glacier, while a pronounced surface heightening existed at the Kamb Ice Stream. We reveal an effective method for combining Antarctic ice sheet data from new radar altimeters. This method supports the long-term monitoring of the Antarctic ice sheet and global climate change in the future.

To obtain high-spatial-resolution land surface temperature (LST) in karst areas, it is necessary to select a downscale regression model with a better simulation effect and the scale factors that can best represent the topographic characteristics of karst mountainous areas. In Guiyang, a typical karst mountain city, two areas are selected as the study area, which is dominated by natural surface and construction land. Based on the data of Landsat-8 Thermal Infrared Sensor (TIRS), Sentinel-2, Advanced Land Observing Satellite Digital Elevation Model (ALOS DEM), and meteorological stations, the scale factors representing bare land: bare soil index and topographic relief: mountain shadow (hillshade), relief degree of land surface (RDLS), solar incident angle, and sky view factor (SVF) are added on the basis of the conventional factors. At the same time, random forest (RF) and extreme gradient boosting (XGBoost) models are used to construct an LST downscaling method that is more suitable for karst mountain cities. After the above steps, the LST product with a spatial resolution of 10 m is finally estimated. The results show that, due to the characteristics of large elevation variation, fragmentation, and high heterogeneity of surface landscape in karst areas, digital elevation model (DEM), RDLS, and SVF factors need to be considered in the downscaling of surface temperature, and the contribution rates of these factors are all more than 6% in the model. In terms of accuracy evaluation of ground temperature, XGBoost model has the highest accuracy with an average absolute error of 1.67K, RF model has an average error of 1.90K, and thermal image sharpening has the worst accuracy with an average error of 2.41K. In terms of accuracy evaluation of ascending scale, the XGBoost model also shows higher accuracy and richer texture details. The research results can provide basic data for the acquisition of high-resolution LST and its intermediate parameters in this area and also provide a method reference for the reduction of high-resolution LST in similar areas.

Facing the climate change and anthropogenic activities that have been discharging a large proportion of carbon dioxide CO₂) into the atmosphere, wetlands stand out as an important sink for CO₂ that is fixed in plant biomass and peatlands. Therefore, quantifying and monitoring wetland biomass is of great importance to preserve carbon stocks. This study aims to explore the potential of multispectral bands and vegetation indices (VIs) derived from PlanetScope and Sentinel-2A sensors to estimate of aboveground biomass (AGB) and organic carbon in AGB (Corg) in the emergent vegetation of a palustrine wetland. We use correlation analysis and linear regression models to examine the relationships between spectral and biophysical variables and verify the best predictor spectral variables for AGB and Corg. Scirpus giganteus vegetation was sampled in the Banhado Grande wetland, in southern Brazil. The VIs were best correlated and preferred as predictor variables. The most accurate model used data from the PlanetScope sensor and VI of photochemical reflectance. Both sensors showed potential for pixel-based estimates of AGB and Corg due to their low RMSE values and their contribution as predictors of biophysical variables, which can contribute to opening new avenues in scientific research focusing on the management, monitoring, and conservation of marshes and your ecosystem service of carbon sink.

Drone object detection in low-altitude airspace plays an essential role in many practical applications, such as security and airspace monitoring. Despite the remarkable progress made by many methods, drone object detection still remains challenging due to the complex background and huge differences in scales of drones. To address the above issues, an improved fully convolutional one-stage object detection (FCOS) model based on adaptive weighted feature fusion (AWFF) module is proposed for multiscale drone object detection in complex background. By learning the spatial relevance of feature maps at each scale and improving the scale invariance of features based on the channel attention mechanism, AWFF module could adaptively fuse the features of adjacent scale. In addition, a receptive field enhancement module is designed to reduce the information loss in the feature fusion process. Extensive experiments are conducted to evaluate the effectiveness of the proposed module and method on the constructed low-altitude drone dataset, which concludes that the mean average precision of the AWFF-FCOS is increased by 2.1% compared with the baseline method. And extensive ablation experiments further demonstrate that the proposed AWFF module and REF module could be integrated into the state-of-the-art method to improve the performance of drone object detection.

Landslides are a recurring phenomenon in Brazil and have caused many socioeconomic losses and casualties. To monitor them, land use and land cover (LULC) and landslide inventory maps are essential to identifying high susceptibility areas. In this sense, the main aim of this study is to produce LULC classification focused on landslide detection via semi-automatic methods, using data mining techniques with remote sensing time-series imagery. For that, different indices, such as the normalized difference vegetation index, the normalized difference built-up index (NDBI), and the soil adjusted vegetation index were extracted from Sentinel-2 imagery. Basic, polar, and fractal metrics were extracted from the time series. From the Shuttle Radar Topography Mission digital elevation model, six geomorphometric features were extracted. Then, classification was performed with random forest with four different approaches: mono-temporal, bi-temporal, metrical, and all. In every approach, the NDBI index or metric derived from it presented the highest importance, and the slope was ranked among the six first predictors. The all approach showed the highest overall accuracy (OA) (88.96%), followed by metrical (87.90%), bi-temporal (82.59%), and mono-temporal (74.95%). Briefly, the metrical approach presented the most beneficial result, presenting high OA and low levels of commission and omission errors.

Grain agriculture is one of the main economic activities in Thunder Bay. However, farming faces many challenges: shorter growing seasons, severe weather conditions, climate change, and limited services and economic support. Lakehead University Agricultural Research Station (LUARS) is continuously working toward improving farming in Northern Ontario and testing different agricultural initiatives. Precision agriculture (PA) is one of the strategic means to improve Northern Ontario farming and sustain agricultural operations. Hence, PA requires accurate and up-to-date baseline information; the aim of this study is to develop baseline spatial information (land cover and soil pH variability maps) for the LUARS using remote sensing technology. A set of aerial images were acquired from the remotely piloted aircraft system (RPAS) and created an orthophoto (7-cm spatial resolution). A detailed land cover map was produced using a deep learning algorithm. We collected soil samples from three different sections (plots) of the area to analyze soil pH variations. Different geostatistical methods [ordinary kriging, Simple kriging, and empirical Bayesian kriging (EBK)] and deterministic interpolation techniques: local polynomial interpolation (LPI) and inverse distance weighted (IDW) were tested to map the spatial distribution of soil pH. A land cover classification achieved an overall accuracy of 73% and a Kappa coefficient of 0.70. Most of the homogeneous agricultural plots indicated higher producer’s and user’s classification accuracies. There was no linear correlation between soil sample data and vegetation indices derived from RPAS images to test the linear regression for soil pH mapping. EBK, LPI, and IDW produced soil pH maps with higher accuracies for Secs. 1–3, respectively.

Optical remote sensing images play an important role in many fields due to their natural visual representations. However, clouds obstruct a large proportion of recorded spectral signals and greatly affect the availability of high-quality optical images. In addition, the progress of cloud removal methods has long been restricted by the deficiency in theoretical basis and low computing power. With the advent of deep learning technology, this problem can be well addressed. We propose a network based on style transfer theory to remove thin and thick clouds in optical remote sensing images. An input data concatenation procedure with synthetic aperture radar (SAR) images is used to provide auxiliary information for the recovery of missing data caused by clouds. Moreover, the spatial attention mechanism adept at generating saliency information of image objects is also used to locate the clouds and corresponding shadows to make the network concentrate on the cloud region reconstruction task. Furthermore, to make the spatial attention mechanism self-adapted to the practical characteristics of clouded images, a loss function including a cloud mask is designed to assist the training step of the network. The proposed network is trained with a global open-source dataset and a region-specified dataset. The results of ablation experiments and comparison with baseline methods demonstrate that a high accuracy is guaranteed by the proposed model, even under high cloud coverage conditions [mean absolute error (MAE) = 0.0883, root-mean-square error (RMSE) = 0.1775, peak signal-to-noise ratio (PSNR) = 23.8, and structural similarity (SSIM) = 0.742). Using optical images independently to remove clouds results in poor results (MAE = 0.1156, RMSE = 0.2825, PSNR = 21.6, and SSIM = 0.596), which need to be addressed with data concatenation, such as SAR context information. Spatial attention mechanism plays an important role in improving the accuracy of cloud removal (MAE = 0.0251, RMSE = 0.0363, PSNR = 27.1, and SSIM = 0.876). For the missing values of optical images caused by cloud coverage, the Spatial-attention-CycleGAN proposed provides a potential solution for information recovery.

Information about species mapping is an essential approach for the management of forests and sustainable practices of conservation. Remote sensing data have proven to be an asset for the assessment of the spatial distribution of species over time. We used time-series data from a single sensor (PlanetScope) and dual-sensor (PlanetScope with Sentinel-2) to map Dalbergia sissoo, a timber species, found on both sides of the Jakhan river in Dehradun district of Lesser Himalaya. The dimensionality of the temporal images was reduced by normalized difference vegetation index (NDVI) and class-based sensor independent NDVI (CBSI-NDVI). Separability analysis was conducted to find optimal dates of the data set using three measures of separability, and Euclidean distance was found to be the best measure of separability for both the indices. Due to limitations of the classifiers in handling mixed pixels, a fuzzy-based modified possibilistic c-means (MPCM) algorithm was tested to extract a single class, i.e., Shisham (Dalbergia sissoo) tree. We used the conventional mean and individual sample as mean (ISM) as training parameter concepts in the MPCM supervised classification approach. We found that the ISM approach was able to handle the heterogeneity within the class for both vegetation indices. It was seen that PlanetScope data were able to spatially map the target class in a better way, whereas combined data of PlanetScope and Sentinel-2 helped to reduce the spectral overlap between target and nontarget classes. An accuracy assessment was performed using mean membership difference, variance, and entropy where variance and entropy showed that NDVI-based ISM approach outperformed the CBSI-NDVI-based approach. Both single-and dual-sensor time-series data showed good classification results for single species extraction.

Ultra-spectrally resolved infrared measurements from aircraft and space-based observations contain information about tropospheric carbon monoxide (CO) and ozone (O₃), as well as other trace species. A methodology for retrieving these tropospheric trace species from such remotely sensed spectral data has been developed and validated for the National Airborne Sounder Testbed-Interferometer (NAST–I). The Fire Influence on Regional to Global Environments and Air Quality (FIREX-AQ) field campaign was conducted during August 2019 to investigate the impact of wildfire and biomass smoke on air quality and weather in the continental United States. NAST–I CO and O₃ measurements from the recent FIREX-AQ field campaign are presented and used to estimate wildfire plume age. Results show enhanced levels of CO in the evolving plume as it is transported away from the fire ground site, and its plume age is associated with the plume distance in both the vertical and horizontal directions from the wildfire location. These results are enabled by the moderate-vertical and high-horizontal resolution obtained from the NAST–I IR spectrometer onboard the NASA ER-2 aircraft. This study advances our knowledge of fire-induced plumes with their evolution and age characterized in three-dimensional space using information from NAST–I retrieved CO and O₃ and relative changes in their concentrations.

Fusion of complementary information from multisensor data is of great importance for identifying the land covers. However, integration of multisource information is a challenging task. A framework is developed to integrate hyperspectral and LiDAR data for land cover classification. In the proposed method, sparse stacked autoencoders are used to represent the spectral and spatial information in a compact form. The spatial information is extracted both from hyperspectral and LiDAR data using morphological operators. The encoded spectral and spatial features are combined with elevation information to form a joint feature vector. The joint features are fed to a convolutional neural network (CNN) classifier to classify the land covers. The CNN classifier is a hybrid three- and two-dimensional (3D)–(2D) model having three 3D convolutional layers and one 2D convolutional layer. The experiments are carried out on two datasets Houston and Samford to evaluate the performance of the proposed method. The results have demonstrated the effectiveness of the method with global κ = 0.9285 and global naive accuracy (OA) of 93.44% for Houston data. For Samford data, it achieves κ = 0.9811 and OA = 98.93 % .

Accurate assessment of woody species diversity using remote sensing can assist ecologists by providing timely information for ecosystem management. The increasing availability of remotely sensed data necessitates the investigation of accuracies of different sensors in classifying plant species, especially during the dry season when foliage amount is low. WorldView-2, SPOT-6, and Sentinel-2 images were compared in detecting woody species (n = 27) and three coexisting land cover types in a savanna environment during a dry period. Random Forest (RF) and Support Vector Machine (SVM) classifiers were applied to each imagery to make a strong case for the comparison. The overall classification accuracies ranged between 52% and 65% for all images, with the WorldView-2 image performing the best followed by Sentinel-2 and SPOT-6 images. These accuracy rankings were similar for both the RF and SVM classifiers, with the former faring better. Pairwise comparison of the images using McNemar’s test showed significant differences between images in their ability to correctly identify woody species. Analysis of band importance revealed better contributions to the classifications of infrared bands for all images. Overall, the findings showed the potential of optical imagery in classifying and monitoring woody species hotspots in savanna environments even during a low photosynthesis season.

Mangroves play an extremely important role in purifying the atmosphere and responding to global temperature changes. The analysis of chemical elements (carbon, nitrogen, phosphorus, etc.) in mangroves is an effective way to investigate physiological activities, such as vegetation growth, development, and material metabolism. Therefore, the monitoring of mangrove stoichiometry is extremely important for mangrove restoration. Here, two mangrove species, Kandelia candel (KC) and Aegiceras corniculatum (AC), were studied in the Quanzhou Bay Estuary Wetland Nature Reserve. Two machine learning models [random forest (RF) and back propagation neural network (BPNN)] and partial least squares (PLS) were established with the original spectral data as independent variables, and the optimal model was selected by comparing the simple cross-validation VE_cv, the ratio of performance to deviation, and the root mean square error (RMSE). The results showed that: (1) the contents of total phosphorous and total nitrogen decreased gradually and the content of total carbon of mangroves increased gradually with an increase in age at restoration; (2) hyperspectral modeling can invert the ecological stoichiometries of KC and AC, and it can be used to effectively monitor the growth status of the species studied; and (3) model performance ranking, PLS > RF > BPNN, where PLS (VE_cv ≥ 0.60 and RMSE < 40) was significantly better than the other two models, and BPNN was the least effective and not suitable for hyperspectral inversion modeling of KC and AC ecological stoichiometries. This study provides a methodological basis for long-term and large-scale dynamic monitoring of mangrove ecological stoichiometries and mangrove restoration quality based on hyperspectral data.

Long time series urban mapping using Landsat images remains challenging due to the difficulty of collecting representative ground samples. We proposed a mapping approach integrating a sample-transferring scheme, multiple-feature fusion classification, and temporal filtering to generate annual urban land maps in Zhengzhou during 1986 to 2021 with the aid of the Google Earth Engine (GEE). The training and validation samples for each historical year were obtained by a proposed sample transfer scheme based on stable areas derived from the GlobeLand30 datasets. Thirteen variables, such as spectral indices, spectral bands, and terrain factors, were combined to form the feature vector. Annual classification was conducted yearly using the random forest algorithm on GEE to obtain annual impervious surface (ISs) maps. Finally, the IS classification maps were postprocessed to obtain the final urban land maps. Zhengzhou was selected as the study area due to its dramatic urbanization in recent decades. The results showed that the overall accuracy and Kappa coefficient of the IS classification maps ranged from 0.8888 to 0.9578 and 0.6548 to 0.8331, respectively. Compared with other 30-m land-cover products, our urban land maps showed higher accuracy and more reliable spatial details. The proposed approach can generate massive ground samples for long time series urban mapping and is helpful for updating regional and global land-cover products. Zhengzhou’s urban area increased by 1423.24 km² with an average expansion area of 40.66 km² / year from 1986 to 2021. The urban expansion in Zhengzhou showed significant stage characteristics and spatial variations. The expansion intensity indices were 0.204, 0.404, 0.736, and 2.018, respectively, at four time periods and varied across the 12 regions in Zhengzhou.

An evaluation of land use and cover change is a vital component of any study into climate change, ecological evolution, and human civilization’s long-term growth. Remote sensing image data-based land use and cover change (LUCC) research has become an essential and frequently utilized approach. Given the scarcity of high spatial resolution imagery in urban remote sensing, as well as the low accuracy and efficiency of urban land use classification, a new satellite image fusion methodology defined as nonshear wave transformation, a pulse linked neural network, and intensity–hue–saturation theory are suggested. From 2000 to 2020, the upgraded convolutional neural network approach is used to classify fused pictures and perform an in-depth investigation of the spatiotemporal evolution features of urban LUCC in Zhengzhou, Henan, China. According to the findings, the extent of urbanized land in Zhengzhou has expanded dramatically during the last 20 years. The share of urbanized land has risen from 9% in 2000 to 22% by 2020. The comprehensive dynamic degree and single dynamic grade of land use display varied features in different areas and counties; the comprehensive index of the extent of land use demonstrates more evident regional disparities. The research findings can expose the man-land system’s inherent conflicting interaction mechanism and give data to promote urban-related research.

The southeastern Fennoscandian Shield is a specific region dominated by glacial and periglacial originated Quaternary deposits. In dense boreal vegetation zones, direct remote decoding of Quaternary sediments is impossible. Our aim is to assess the efficiency of machine learning algorithms for identification of types of Quaternary deposits based on spectral data of vegetation cover and digital elevation model data. A comparative analysis of two classification methods, discriminant analysis (DA) and classification and regression trees (CARTs), was performed. Two models, DA and CART, were constructed based on a set of spectral variables. They included principal components and spectral indices, such as the normalized difference vegetation index, the normalized difference moisture index, and the clay minerals ratio. When the absolute height and the topographic position index (DA+ and CART+ models) are added to a set of independent variables, the classification of Quaternary sediments becomes more accurate. The nonparametric CART method was shown to be more accurate in differentiating loose deposits. The correctness of the CART+ model for the reference sample of observation points was as high as 90.5%. According to the trained data, mapping of Quaternary deposits was carried out. Kappa analysis showed that the agreement with the map of Quaternary deposits for the CART+ and DA+ models was 0.5 and 0.44, respectively. For the DA and CART models, the agreement was much lower. It is noted that the sets of spectral data for the vegetation cover more clearly show the grain-size composition of loose sediments. To make the classification of Quaternary deposits more correct, morphometric indicators were added. Furthermore, the CART+ method was used to estimate the maximum height of glaciolacustrine sediments (117.5 m), which corresponds to the maximum absolute mark of the periglacial reservoir.

Accurate temperature prediction is of great significance to human life and social economy. A series of traditional methods and machine learning methods have been proposed to achieve temperature prediction, but it is still a challenging problem. We propose a temperature prediction model that combines seasonal and trend decomposition using loess (STL) and the bidirectional long short-term memory (Bi-LSTM) network to achieve high-accuracy prediction of the daily average temperature of China cities. The proposed model decomposes the temperature data using STL into trend component, seasonal component, and remainder component. Decomposition components and the original temperature data are input into the two-layer Bi-LSTM to learn the features of the temperature data, and the sum of prediction of three components and the original temperature data prediction result are added using learnable weights as the prediction result. The experimental results show that the average root mean square error and mean absolute error of the proposed model on the testing data are 0.11 and 0.09, respectively, which are lower than 0.35 and 0.27 of STL-LSTM, 2.73 and 2.07 of EMD-LSTM, 0.39 and 0.15 of STL-SVM, achieving a higher precision temperature prediction.

The advanced baseline imagers (ABIs) on board the geostationary operational environmental satellite (GOES)-R series, GOES-16 and GOES-17 satellites, have accumulated long data series that have been updated throughout their missions to enhance their utility for imaging the Earth’s weather, oceans, and environment. In preparing for a reprocessing effort aimed at improving the consistency and image quality and reducing radiometric uncertainties necessary for some Earth science applications, we developed a vicarious validation technique for five of the six reflective solar bands. This technique uses Sonoran desert surface reflectance data derived from the moderate resolution imaging spectrometer bidirectional reflectance distribution function parameters product along with other atmospheric products to generate a predicted at-sensor radiance for both GOES-16 and GOES-17 for comparison with ABI observations and to enable ABI intercomparisons. This technique identified known instrument biases and demonstrated absolute validation over the course of a year, showing its capability to detect changes, monitor stability, and assess radiometric uncertainties for operational data applications and future reprocessed data sets.

As a key variable in the surface hydrological cycle and energy balance, soil moisture is often analyzed using data assimilation to improve precise and high spatiotemporal resolution from multisource data. However, accurate estimation of model error in soil moisture data assimilation is not sufficient in spatial and temporal dimensions. This paper proposes a spatiotemporal error estimation method that integrates the spatial and temporal dimensions of the study area into model error estimation. The in-situ data was regarded as the “true value” for the temporal dimensions model error estimation. The triple collocation (TC) technique was used to estimate the spatial error of the model. Based on the aforementioned estimation, a linear method to construct the relationship between spatial and temporal information was proposed to fuse the spatiotemporal model error. Result from the proposed method was compared against the overall model error estimation and spatial estimation methods in ensemble Kalman filter (EnKF) data assimilation. The ERA-Interim reanalysis data and in-situ soil moisture data from Washington State in 2016 were used to evaluate the performance of EnKF data assimilation. Using statistical metrics, including the root mean square error (RMSE), the mean bias error (MBE), and the Pearson correlation coefficient (i.e., R-value), the results show that our proposed approach, which uses spatiotemporal approach to model error estimation in EnKF data assimilation, yielded more accurate soil moisture estimates. When ERA-Interim reanalysis soil moisture data was used as the reference data, compared with the overall model error estimation EnKF Data Assimilation Experiment, RMSE and MBE of spatiotemporal model error estimation EnKF Data Assimilation Experiment were reduced by 0.0062 and 0.0061 m³ / m³, respectively, and R remained basically unchanged. When using in-situ measurements as reference data, RMSE and MBE decreased by 0.0034 and 0.0038 m³ / m³, respectively, and R increased by 0.0365. Compared with the other estimation approaches in the experiments, the spatiotemporal error estimation approach greatly improved the estimation accuracy of soil moisture analysis, yielding the closest values and variation trends with the in-situ measurements. Fully considering the temporal and spatial variation of model error and improving the estimation accuracy of model error can provide more accurate model information for data assimilation. The proposed method and framework in this paper can be used in model error estimation and data assimilation, particularly in the fields of numerical weather forecasting, natural environment monitoring, and geographical environment research.

Many shadow detection algorithms pertaining to remotely sensed imagery exist. Several of these algorithms exploit the spectral characteristics of shadows within imagery to identify shaded regions. However, these algorithms can have problems when water is also present within the imagery because water shares similar spectral characteristics with shadows. Many of these algorithms are applied to small image subsets instead of the image as a whole and frequently are applied to urban environments that require additional use of the normalized difference water index or other features to affect the removal of water from the shadow mask. This diversity of the image scene content coupled with the complexity and wide variety of environmental conditions that satellite imagery can acquire makes reliably separating shadow and water within larger images a complex problem. Thresholding various spectral indices to produce segmentation maps can be challenging when large, complex, and often imbalanced scenes are captured, which may require manual adjustment of algorithmic parameters for different image areas. We present a methodology that makes use of the near-infrared channel using a watershed segmentation algorithm, local binary pattern measurements, and a support vector machine to classify shadow and water within a full WorldView scene. Results are promising with initial accuracies above 97% for both shadow and water.

Interferometric synthetic aperture radar (InSAR) is an important resource for rapidly obtaining forest height information. The coherence amplitude method relies solely on interferometric coherence amplitude information for forest height inversion; however, interferometric data are commonly affected by decorrelation factors including time, baseline, and noise decorrelation, as well as observation geometry errors, leading to errors in estimated forest height. Thus, in this paper, we propose a method to improve the accuracy of forest height estimation by correcting the decorrelation and vertical wave number using a priori forest height knowledge. UAVSAR-L cross-polarization channel HV data from the AfriSAR project are used as interferometric data to invert the forest height using the coherence amplitude method, and the relative height variable RH100 from land, vegetation, and ice sensor light detection and ranging (LiDAR) is used for validation. We optimize the coherence amplitude method by iteratively setting different steps for the nonvolume decorrelation (γ_d) and the correction parameter (τ) for the vertical wavenumber (k_z). The optimal compensation parameter is identified when the root mean square error (RMSE) between the inversion height and LiDAR height is minimized, and the stability of the returned parameter is evaluated through an independent validation sample. Our results indicate that enhancing the coherence amplitude method using a semiempirical iterative approach can effectively improve inversion accuracy. In the validation results, all compensation schemes exhibit a significant improvement in the inversion results compared with those without parameter compensation. The R² increases by 0.13 and the RMSE decreases by 9.88 m when compensating only γ_d, whereas the R² value does not change when only compensating k_z, but the RMSE decreases by 19.24 m. When compensating for both γ_d and k_z, the R² increases by 0.08, and the RMSE decreases by 19.73 m. This changing pattern is consistent with that recorded in the training sample, indicating that our proposed parameter compensation scheme for the coherence amplitude method is effective. With the widespread usage of satellite data, such as ALOS-2 and SAOCOM, as well as the future TanDEM-L and BIOMASS satellites and NISAR program, the combination of ICESat-2 and GEDI forest height data to compensate and optimize inversion results and model parameters is expected to greatly improve the efficiency of forest resource monitoring in the future.

There are two heterogeneous data types in hyperspectral image (HSI): rich spectral data and spatial information. Recent research has shown that the application of spectral–spatial information significantly improves HSI accuracy; thus, multiple-feature combination-based methods have been favored by researchers in the field of HSI classification. A multiple-feature combination approach, based on the particle swarm optimization (PSO) algorithm, is proposed for improving the accuracy of HSI classification. The proposed method couples a multiple kernel support vector machine (SVM) with a PSO algorithm to assign optimal weights to different kernels. Moreover, it also solves the problem of artificially selecting weights when learning multiple features by implementing adaptive weights on different datasets. In addition, it has fewer parameters and a shorter training time than deep learning methods, thus, the model is smaller and easier to train. The proposed method was tested on four datasets, containing two and three kernels. The experimental results show that our optimized method improves the classification accuracy; additionally, the kappa performance of the classification is also better.

To get the best performance for a broadband wireless access, this research aims to scale a Ku-band real-time rain attenuation data to higher frequencies and estimate the degree of fading that can be observed under such conditions. Data of simultaneous measurement of satellite radio beacons and rainfall parameters, carried out between January 2017 and December 2018, at Physics Department, Federal University of Technology (07.17°N, 05.18°E, 358 m) Akure, were used for the work. The radio beacon measurement, of horizontally polarized Eutelsat W4/W7 satellite signals at 12.245-GHz frequency, and received through a 90-cm parabolic antenna positioned at 53-deg elevation angle, was conducted at 1 s integration time using Tektronix Y400 NetTek spectrum analyzer. Measurement of rainfall parameters was done at 1-min sampling intervals using Davis Vantage Vue automatic weather station, which is co-located with the receiving antenna of the satellite receiver unit. Comparing measured and estimated attenuation values at 12.245 GHz for 1%, 0.1%, and 0.01% exceedance margins result in about 64%, 40%, and 55% relative error, respectively. Similarly, relative error values between estimated and scaled attenuation values for the same exceedance margins are 72%, 53%, and 27% for 16-GHz frequency, respectively, with similar values for the 20-, 30-, and 40-GHz frequencies as well. A power-law relationship between measured rain attenuation and rain rate was developed and used to generate a location-specific expression for computing rain attenuation. This was equally found to give better performance than the International Telecommunication Union model. The results obtained in this work provide useful information for radio engineers on the rain-induced attenuation needed for the design of radio communication system that can withstand poor weather conditions.

In terms of the target with micromotion parts, the readability of inverse synthetic aperture radar (ISAR) image of the main body is influenced by micro-Doppler (m-D) effects. Some facts, such as radar working modes and electromagnetic environment, may lead to sparse aperture, which further induces high side lobes and makes the removal of m-D effects more difficult. To jointly eliminate m-D effects and the interference introduced by sparse aperture, a novel m-D effects removed ISAR imaging algorithm is proposed. In this technique, sparse ISAR imaging with the removal of m-D effects can be established as an optimization model of joint low-rank and sparse representation, which is a quadruple constraint optimization problem, including the low rank of the high-resolution range profile (HRRP) of the main body, the sparsity of the HRRP of micromotion parts, the sparsity of the imaging result of the main body, and the noise constraint of the echo signal. Furthermore, the matrix factorization method is theoretically presented to simplify the optimization process of the nuclear norm, and the alternating direction method of multipliers is utilized to solve all constructed models efficiently. Experimental results based on both simulated and measured data demonstrate the superiority of the proposed algorithm.

We propose SiamGauss, a Siamese region proposal network with a Gaussian head for single-target visual object tracking for aerial benchmarks. Visual tracking in aerial videos faces unique challenges due to the large field of view resulting in small size objects, similar looking objects (confusers) in close proximity, occlusions, and fast motion due to simultaneous object and camera motion. In Siamese tracking, a cross-correlation ration is performed in the embedding space to obtain a similarity map of the target within a search frame, which is then used to localize the target. The proposed Gaussian head helps suppress the activation produced in the similarity map on confusers present in the search frame during training while boosting the confidence on the target. This activation suppression improves the confuser awareness of our tracker. In addition, improving the activation on the target helps maintain tracking consistency in fast motion. Our proposed Gaussian head is only applied during training and introduces no additional computational overhead during inference while tracking. Thus, SiamGauss achieves fast runtime performance. We evaluate our method on multiple aerial benchmarks showing that SiamGauss performs favorably with state-of-the-art trackers while rating at a frame rate of 96 frames per second.

The development of remote sensing sensor techniques allows us to now readily capture many types of indoor and outdoor scene images, which often include many weak texture regions with notable geometric distortions. Obtaining qualified matches from these difficult stereo images using existing methods is challenging. The recent achievements of deep-learning models have shown that the convolutional neural network (CNN) is adept at the image matching task. However, in practical applications, the following challenges remain: first, it is difficult to detect features in the weak texture regions of an image, and existing CNNs fail to extract discriminative image information from the quantized features of weak texture; second, as a result of the complex distortion across wide-baseline stereo images, it is difficult to match feature primitives detected in the image pair. To solve these problems, we propose the perspective invariant local feature transformer (PILFT) algorithm. Our method includes four main steps. (1) The affine scale-invariant feature transform is proposed to automatically extract the corresponding features from images, and then the perspective of the matched image is corrected to eliminate as much geometric deformation as possible. (2) The residual network is used to extract potential features from stereo images to obtain coarse and fine feature maps at different scales. (3) Using an attention mechanism, location and context information are added to the coarse level features, which are predicted by a dual-softmax function layer. (4) The features are precisely predicted on the fine feature map using the coarse reference, and the final matching results are determined by calculating the matching probability. A large number of experiments on wide-baseline weak texture images demonstrate that the proposed method has advantages over the existing algorithms in the number of matches, correct match rate, and matching accuracy. The pseudocodes of PILFT are available at https://github.com/KiltAB/PILFT.

Synthetic aperture radar (SAR) image super-resolution (SR) has broad application prospects in military reconnaissance, marine monitoring, crop monitoring, and other fields. However, when the scale factor is large, current SAR image SR has limitations, such as insufficient feature extraction, difficulty of feature representation, and modeling difficulty of SR reconstruction. These problems have severely restricted the SR performance of SAR images. In addition, optical remote sensing images share higher resolution and richer high-frequency information than SAR remote sensing images. Therefore, we adopt high-resolution (HR) optical images as auxiliary images of SAR image SR to improve the performance of SAR image SR. Our research is conducted from two aspects: the extraction efficiency of HR optical image-guided information and the representation ability of low-resolution (LR) SAR image features. We propose an optical guidance residual networks (OGRNs) based on residual learning for SAR image SR. The OGRN uses optical images during network training, which enhances the performance of SAR image SR, even when optical images are missing during network testing. First, we extract the texture feature of SAR images through the designed residual learning module. Second, we extract the weight map of feature information for SAR image SR from HR optical images by the convolutional layer. Finally, the weight map is used to guide the weighted supervised SR reconstruction for SAR images. Extensive experimental results demonstrated that, under the guidance of optical images, OGRN achieves excellent performance in terms of both quantitative assessment metrics and visual quality.

Based on the low imaging resolution of bistatic inverse synthetic aperture radar (Bi-ISAR) and the failure of pulse correlation under the condition of sparse aperture cause that of the traditional self-focusing algorithm, a Bi-ISAR sparse aperture self-focusing algorithm with the combined constraint of image quality optimization and sparsity is proposed. First, the proposed algorithm establishes the Bi-ISAR sparse aperture self-focusing signal model, reconstructs images through fast sparse Bayesian learning (FSBL), uses the minimum Tsallis entropy and constraints the reconstruction process, iteratively updates the phase error, and performs self-focusing to realize the initial phase correction of Bi-ISAR images. Simulation results show that the proposed algorithm has a fast convergence speed, strong robustness to noise, and high accuracy in reconstructing images.

The proposed objective was to develop a near-automated workflow for orthorectification of archived aerial photographs that overcomes the limited resources problem. Automation was implemented using the oriented fast and rotated brief (ORB) and random sample consensus (RANSAC) for conjugate points detection. Because it is difficult to determine the RANSAC threshold, the RANSAC threshold search for two step triangulation method was developed to find the threshold that resulted in the minimum three-dimensional misclosures at check points from the threshold candidates. For triangulation, the bundle block adjustment (BBA) with a minimal constraint and the absolute orientation were implemented. Although the absolute orientation still requires manual measurement of control points, the required number is usually less than that of traditional BBA, especially for longer strips. The control points were measured in relatively recent high-accuracy orthophotos and the digital elevation model. In the experiments, the archived aerial strip photographs were orthorectified using a differential rectification method with ∼1 m resolution. The first and second orthorectification experiments showed 1.31 m (X), 1.14 m (Y), and 1.88 m (X), 1.51 m (Y) of RMSE. The proposed workflow suggested is significant because it can save cost, time, and labor required for the orthorectification of archived aerial photographs while achieving reasonable accuracy.

Scene classification is an important tool for remote sensing image interpretation, and it has fundamental applications in research and industry. However, given complex backgrounds and scale variations, remote sensing images have large intraclass diversity and interclass similarity, which bring challenges to accurate classification of remote sensing images. We proposed a scene classification method using joint learning and multiscale attention to alleviate the aforementioned problems. To fully utilize the multiscale information of the image and improve the adaptability of the proposed method to objects with various sizes, different from general methods that fuse different scales of features for classification, joint learning using multiscale features is developed to optimize the whole network. Specifically, we leverage a pretrained deep convolutional neural network as the feature extractor to extract low-level, medium-level, and high-level feature maps from the images. Then, due to the poor semantics of low-level and medium-level feature maps compared with the high-level feature maps, we design a multiscale attention module to enhance the semantic information and suppress the noise information. Finally, the global mean pooling is used to obtain the feature vectors and different classifiers are used for different feature vectors. And the decision-level fusion is adopted to obtain more reliable predictions. The experimental results on the AID and NWPU-RESISC45 datasets show that the proposed method makes a significant improvement in terms of overall accuracies compared with the baselines. And the overall accuracies of our method on the two datasets are 97.49% and 95.20%, respectively, which achieves state-of-the-art performance. The code will be public at a Github repository available at https://github.com/Cbanyungong/JLMSAF.

Ship classification in high-resolution (HR) synthetic aperture radar (SAR) images is an important and challenging task. The deep learning methods rely heavily on massive labeled data, which is not practical for SAR applications due to the lack of feature consistency under different sensors, angle of view, and scenarios, while the statistical methods are vulnerable to noise, image quality, and radiometric calibration and have poor robustness. To overcome these problems, a more robust superstructure scattering feature, named scattering bright line feature, is proposed to realize the convenient classification of ships in engineering applications. Based on this, a decision tree classifier for three types of merchant ships is developed. Four types of HR SAR slices are used to verify the validity of the new feature by both the decision tree and support vector machine classifiers, and the results show that the proposed feature has good classification performance in merchant ships with an overall accuracy higher than 80%.

Sentinel-2 and Sentinel-3 are two remote sensing satellites implemented by the European Space Agency for global observation. The temporal resolution of Sentinel-2 images and the spatial resolution of Sentinel-3 images may not be sufficient for local and precise monitoring. With spatiotemporal image fusion of Sentinel-2 and Sentinel-3 sensors, images with 1.4 days temporal resolution and 10-m spatial resolution can be produced. However, strong temporal change is a challenging factor for spatiotemporal fusion. The aim of this study was to compare the success of the deep learning-based DMNet model with flexible spatiotemporal data fusion (FSDAF) 2.0 and reliable and adaptive spatiotemporal data fusion (RASDF) algorithms for the spatiotemporal fusion of Sentinel-2 and Sentinel-3 images over strong temporal changes. Thus, the Kansas dataset was developed for the spatiotemporal fusion of Sentinel-2 and Sentinel-3 images in this study. It contained a large number of surface changes due to large wheat harvests. The results of this investigation show that in case of strong temporal change, the deep learning-based DMNet model performed better than the FSDAF 2.0 and RASDF methods. On the other hand, in the case of less temporal change, the FSDAF 2.0 and RASDF methods had a very high success compared with the DMNet model.

Object detection plays an important role in the field of remote sensing (RS) images analysis. The advancement of object detection task for RS images is extremely challenging due to object scale variation and complex background. Almost all detection frameworks use the neck network to fuse the feature maps extracted by the backbone to obtain better features, among which feature pyramid network (FPN) is the most widely used. Although traditional FPN has shown great potential in multi-scale object detection based on deep learning, it has unsatisfactory detection accuracy for small objects and confusing objects in RS images because of the lack of rich semantic and contextual information. We propose an architecture, called semantics reused context FPN that is portable to any FPN-based detectors to boost the detection performance in RS images without parameters increasing significantly. It includes two blocks, namely, context feature enhanced block, which uses dense connection and a learnable branch structure to extract rich context features with multiple receptive fields, and semantic feature reused block, which enhances semantic information of shallow feature maps by reusing later-layer features. Comprehensive evaluations on three benchmark datasets of geospatial object detection demonstrate that our method is superior to the existing methods.

The description of context information affected by speckle and class imbalance under labeled data makes the pixelwise classification for high-resolution (HR) synthetic aperture radar (SAR) image a challenging task. To address these issues, we propose a global-context pyramidal and class-balanced network (GPCNet) for HR SAR image classification. The proposed structure follows an encoder–decoder architecture. In the encoder module, the multiscale convolutional and global-local cross-channel attention (GCA) blocks are employed to capture the global-context and distinguishable deep feature statistics, while reducing the impacts of the random fluctuation in the homogeneous region. The channel information of different scale convolutional layers is efficiently learned by local cross-channel interaction in the GCA block. Besides, a sampled class-balanced loss, associating with the effective number, is utilized for alleviating the class imbalance of HR SAR images. The experiments carried out on a TerraSAR-X image classification dataset demonstrate GPCNet is able to yield superior performance when compared with other related networks.

Considering the fragmentation of karst mountainous areas, a Dajiang Innovation Mavic 2 Pro quadcopter drone was used to collect visible light images of a modern intensive agricultural planting area in response to the high labor intensity and low efficiency of manual assessment of agricultural plastic film mulching (PFM). First, the characteristics of RGB values of 10 main types of features in the sample area, such as PFM, crops, buildings, and roads, were analyzed based on the color index, geometric size, and texture characteristics. Second, based on the construction principle and form of the excess green index, we constructed the excess blue index (ExB), which comprehensively utilized the three visible light bands of red, green, and blue. By calculating the ExB of the image, the characteristics of the PFM were enhanced. Third, Gaussian High-Pass Filter (GHPF) was used for ExB images to retain high-frequency information of PFM targets and reduce the influence of noisy ground types such as plant ashes, blue-roofed buildings, asphalt roads, and water on identifying PFM. Then, combined with field measurements of PFM data, the GHPF images were segmented using Otsu and color slices. The polygon area method was used to eliminate polygons that were too large or too small. Combined with random samples from field surveys, the threshold segmented PFM polygons were verified. The accuracies of the PFM area by methods of Otsu and human–computer interaction threshold segmentation (HC-ITS) are 85.44% and 97.36%, respectively, and the average accuracies of the verification areas B-I and B-II are 90.03% and 82.32%, respectively. It is proved that the PFM ExB proposed in this study has good applicability. We find that the HC-ITS method is suitable for scenes with complex ground object types, and the Otsu method is suitable for scenes with fewer ground object types.

Spatiotemporal fusion is a simple, feasible, and economical method for balancing the temporal and spatial resolutions of satellite images. However, the current spatiotemporal fusion methods have some disadvantages, such as their insufficient feature extraction abilities and unsatisfactory fusion image effects. To obtain higher-quality spatiotemporal fusion images, a spatiotemporal data fusion method based on deep learning is proposed. This method combines an attention mechanism and a residual learning strategy to design an asymmetric spatial channel attention residual network (SCRNet). For different input images, the SCRNet extracts rich feature information to fuse high-quality images in a more emphatic and scientific manner than other approaches. Specifically, we design an efficient and flexible spatial channel attention mechanism that not only focuses on spatial information but also takes channel information into account. The optimized residual network can further enhance the ability of the network to extract feature information. In addition, we design a compound loss function and propose an edge loss to further enrich the extracted feature information and improve the resulting image quality. Experimental results on two datasets show that the proposed method outperforms existing fusion methods in subjective and objective evaluations.

Hyperspectral images with an immense number of spectral bands provide abundant discriminant information for accurate land-cover classification in the remote sensing field. However, these narrow and adjacent bands contain a large amount of redundant information. Analyzing these images always requires a huge storage space with expensive computational costs. Furthermore, their high correlation coefficient would lead to the Hughes phenomenon, hindering the improvement of classification performance. We propose a linear semi-supervised hyperspectral feature extraction method L³ME to learn latent local manifold embeddings. Although labeled samples are beneficial to construct learning models, their number is always limited in real-world tasks. The motivation of this paper is to jointly enhance the contributions of labeled and unlabeled samples for learning local manifold structures of hyperspectral images. Features in labeled samples are extracted by two procedures, the adaptive patch alignment framework and integrated intraclass-interclass relationships, from different perspectives. The former aims to solve the problem of the uneven distribution of classes by introducing spectral angle based adaptive parameters. The latter aims to solve the problem of the uneven distribution of samples by constructing several adjacency graphs. The locality preserving projection is capable of preserving the local neighborhood structure of samples. A penalty for sparse regularization is cleverly integrated into the proposed linear discriminant objective function, which is optimized using a novel updating strategy. The convergence of L³ME is proved in detail and analyzed in this paper. Experiments on three typical hyperspectral datasets illustrate the effectiveness of the proposed method over some state-of-the-art techniques. The implementation of L³ME is available at https://github.com/biowby/L3ME.

In real-world applications, the hyperspectral image (HSI) classification faces two critical problems: one is the Hughes phenomenon and the other is the large spatial and spectral variabilities among the pixels of the HSI, especially for the case where only few labeled pixels per class are available. To address these problems, we propose a united framework for the hybrid representation model based on two dictionaries in this paper, where the input pixel (or its feature) is explicitly decomposed into the class-specific component and the variation component via the Bayesian approach. Based on the proposed framework, two hybrid representation-based classification approaches, named as the sparse and dense hybrid representation-based classification and the hybrid sparse representation-based classification, are proposed based on different priors on the separate components over the associated dictionaries. Moreover, the techniques for designing the associated dictionaries and estimating the hyperparameters are also present so the practicality of the proposed approaches is further improved. Experimental results show that the proposed two approaches outperform the conventional classifiers and are robust to the spatial and spectral variabilities, based on four real hyperspectral datasets.

To fully use the contextual information of hyperspectral images (HSIs), we propose a U-shaped network model combined with attention mechanism to achieve image-level HSI classification. First, the entire HSI is input into the network for end-to-end training, and the classification results of the entire scene are directly output. Then, the context information is used to improve the classification accuracy, while reducing many redundant calculations. Second, to improve the classification accuracy, considering two dimensions (i.e., space and channel), a hybrid attention module, mixing spatial and channel, is designed. Third, three datasets of the University of Pavia, Indian Pines, and Salinas are selected for the classification experiments. The experimental results show that, compared with other methods, the proposed method can obtain higher classification accuracy, and its training and testing efficiency is higher.

Multispectral images have numerous features and a wide range of applications. However, traditional image compression methods, such as JPEG2000 and 3D-SPIHT, do not make effective use of spectral information. We propose a deep compression framework based on interspectral prediction to take full advantage of spectral correlation when using temporal correlation for interframe prediction in video compression. First, two-dimensional and three-dimensional convolutions were used to obtain spatial and spectral information for predicting the original image. Then, we applied a residual neural network to compress the residual information of the image. Subsequently, a decoder was employed to reconstruct the multispectral image based on the compressed image and residual information. All components were jointly trained by a single loss function that considered the tradeoff between the compression bit rate and decoded image quality. The experimental results showed that our proposed method outperformed other traditional compression algorithms, including JPEG2000, 3D-SPIHT, and PCA+JPEG2000, in terms of peak signal-to-noise ratio and spectral angle and is equivalent to or even better than some image compression algorithms based on deep neural networks.

Anomalous objects detection for hyperspectral imagery is a significant branch in the area of remote sensing. Although enormous advancements have been developed, issues of redundancy of spectral information and correlation between pixels should be further explored and improved. To address these problems, we proposed a method that is on the basis of integrating collaborative representation with multipurification processing and local salient weight. Multipurification processing consists of spectral bands purification (SBP) and background purification (BGP). First, to alleviate the interference of redundant spectral information, we remove unnecessary spectral bands by adopting SBP based on considering the global spectral intensity of each band. Then, we remove the outliers in the local dual window by BGP to avoid the effect of heterogeneous pixels. Simultaneously, we obtain the local salient weight by calculating the similarity and difference of pixels in the dual window. Next, we obtain the initial detection result by a collaborative representation, which has been testified to be very effective. Finally, combined with the local salient weight map, the initial detection map is improved to the final detection map. To demonstrate the superiority of the proposed method, we conducted the comprehensive experiment on three public benchmark datasets that contain 15 hyperspectral images.

Due to the special characteristics of hyperspectral images (HSIs), hundreds of continuous bands, and low spatial resolution, it is of great importance to explore the coherence among hyperspectral bands and extract the spatial information as far as possible to reconstruct the high-resolution (HR) HSIs, in which most of the methods failed. We propose an HSI super-resolution (SR) method termed NLB-HMS3D, which consists of two main parts named the spatial–similarity features module and the spatial and spectral correlation utilization module. Different from the majority of existing methods that stack multiple parallel two-dimensional convolution layers to blindly extract more spatial features, we introduce the nonlocal block to expand the receptive fields to thoroughly dig the spatial–similarity features from the image itself. This block not only greatly improves the effectiveness but also reduces tons of parameters. To better preserve the spectral details, we further propose a new block called multiscale spectral features fusion block using the separated three-dimensional convolution with different convolution kernel sizes to explore the diverse spatial–spectral features and fuse them to recover better spectral details. The experiments and data analysis demonstrate that NLB-MS3D can obtain superior performance over many existing state-of-the-art algorithms.

Recently, visible and infrared thermal (RGB-T) images have drawn wide publicity for vehicle fusion detection in traffic monitoring because of their strong complementarities. How to fully use RGB-T images for vehicle fusion detection has generated enormous publicity. However, the infrared thermal dataset is relatively insufficient. Moreover, the most important requirements of vehicle fusion detection are accuracy, fast speed, and flexibility. Therefore, to address these difficulties, we propose a concise and flexible vehicle fusion detection method in RGB-T images via spare network and dynamic weight coefficient-based Dempster–Shafer (D–S) evidence theory. It combines the detection results of RGB-T images based on a decision-level vehicle fusion strategy. In this work, we focus on vehicle detection using infrared thermal images and the vehicle fusion detection strategy. For the former, we construct an applicable network for vehicle detection in infrared thermal images with sparse parameters (weights) and high generalization ability. For the latter, a fusion strategy via dynamic weight coefficient-based D–S evidence theory is proposed to fuse the two detection results of the RGB-T images. In the vehicle fusion detection strategy, we do not directly fuse the two detection results but judge the detection accuracy in advance. Finally, we introduce the VIVID, VOT2019, and RGBT234 datasets to verify the proposed vehicle fusion detection method. The vehicle fusion detection results show that the proposed method presents superior results compared with several mainstream approaches.

The advanced baseline imager (ABI) on board geostationary operational environmental satellite-16 (GOES-16) provides high quality reflective solar band (RSB) and thermal emission band (TEB) image data. Intensive field campaigns for postlaunch validation of the ABI Level-1B spectral radiance observations were carried out during March to May of 2017 to ensure the Système International traceability of the ABI. Radiometric calibrations of the RSBs and TEBs of the ABI were evaluated by comparison with the spectral measurements of the airborne visible infrared imaging spectrometer-next generation (AVIRIS-NG) and scanning high-resolution interferometer sounder (S-HIS) on board the high-altitude aircraft ER2, respectively. The comparison between ABI RSB mesoscale (MESO) data and AVIRIS-NG measurements showed that the mean biases are within 2% with uncertainty <2.5 % for ABI CH01, CH03, CH05, and CH06. The brightness temperature differences between GOES-16 ABI and S-HIS measurements were shown to be within 0.45 K with uncertainty <0.35 K at the 300 K scene equivalent for ABI CH08 to CH10 and CH13 to CH15. The ABI uses large focal plane arrays with the detector number varying from hundreds to more than a thousand per channel, which makes the evaluation of detector uniformity a challenge. Reanalysis of the GOES-16 ABI north-south scan data from the field campaign and lunar observation was performed for ABI CH01 to CH03 with two versions of calibration coefficients. Significant improvements in detector uniformity with the updated solar calibration algorithm were verified.

Moderate resolution imaging spectroradiometer (MODIS) continues to be an important instrument for NASA’s earth observing system. Terra and Aqua MODIS have produced more than 22 and 20 years of global datasets that have significantly helped scientists better understand the Earth’s systems, respectively. The MODIS level-1B (L1B) algorithms use the uncalibrated, geolocated Earth scene observations as input and convert the instrument response into calibrated reflectance and radiance, which are used to generate the downstream science products. The sustained calibration and characterization activities undertaken by the MODIS characterization support team have resulted in several upgrades to the L1B algorithms to maintain accurate calibration in the data products. We present an overview of the L1B algorithm designated as Collection 7. Various algorithm enhancements both in the reflective bands and thermal bands characterization are currently under science testing and evaluation. Once applied in data processing (projected in early 2023), they are expected to manifest in improved science products, both in terms of radiometric accuracy and long-term stability.

Traditionally, high-resolution electrooptical (EO) sensors are placed in Sun-synchronous orbits (SSO) as they enable consistent solar illumination during imaging and near-global coverage for various remote sensing applications. However, these sensors inherently offer limited swath coverage resulting in poorer imaging efficiency and often require a large number of satellites in constellation for improving revisit time over region of interest (RoI). Compared with constellation of satellites in SSO, a smaller number of satellites in lower inclination orbits can provide significant advantage in terms of imaging efficiency in the tropical and near-tropical RoIs, which is desirable for many civilian and strategic applications. In this work, detailed analytical simulations have been performed for inclined orbit constellation of high-resolution EO sensors, and higher imaging efficiency has been demonstrated against SSO constellation. Rigorous mathematical models are developed to enable optimization of constellation design with proper fine-tuning of orbital parameters to maximize imaging efficiency in the RoI, and the model output is validated through STK simulation. For quantitative evaluation of imaging efficiency, two figures of merits, namely, average number of access per day and number of days with no access, have been estimated for various design cases. This study shows significant improvements in imaging efficiency parameters up to a factor of 3 in RoI as compared with traditional SSO constellation. This study is generic in nature and is applicable for any inclination angles and latitudes in the tropical and near tropical RoIs. This study can significantly aid in design of satellite constellations for future remote sensing missions.

Continuously improving the accuracy is a hot topic in hyperspectral image (HSI) classification with small-scale samples, due to the high label noise of traditional labeling systems and the high cost of expert labeling systems. We focus on constructing a smaller and more informative training sample set, so an iterative sample selection method guided by uncertainty measurement (ISS-Un) is proposed. The method learns shallow and deep features in the spectral and spatial domains via a convolutional neural network (CNN), where an uncertainty measurement algorithm such as least confidence (LC), marginal sampling (MS) or entropy (Ent) is used to iteratively select high-quality samples for the training set. In addition, we propose a more efficient uncertainty measurement algorithm named margin-entropy fusion (MEF) algorithm to integrate multiple-criteria information. The proposed method is compared with the conventional random sampling method. Experimental results on three HSI datasets show that the proposed ISS-Un method can significantly alleviate the redundancy of training samples and form a more compact and efficient training set, thus improving the classification performance of pixel-oriented HSI. Meanwhile, training sets constructed based on different uncertainty measurement algorithms are applied to five popular CNN models to verify the quality and generalizability of the selected samples. The results show that these training sets work better than random training sampling. Moreover, the proposed MEF algorithm outperforms the LC, MS, and Ent algorithms in selecting samples and is the main recommended scheme.

A model of lunar spectral irradiance incorporates data from multiple spacecraft and surface telescopic observations. Using 12 data sources derived from 90,000 lunar images, models that are smooth across both geometry and wavelength and involve only about 35 derived coefficients are found with a mean weighted residual of <0.5 % . An irradiance libration model derived from lunar orbiter observations is used to reduce the number of coefficients required. Derivation uses iterative assignment of a single scaling factor for each band in each instrument, which is effectively the long-term lunar calibration coefficient for that band. Calibration of 26 instruments and two other published models reveals that although eight instruments agree within about 3% over 400 to 840 nm, some large biases exist. The model provides a sensitive assessment of instrument response trends.

Deep learning technology has been widely used in synthetic aperture radar automatic target recognition (SAR ATR) tasks due to its good performance and high efficiency. However, the development of semiconductor technology cannot completely satisfy the demand for electronic hardware with high computational capability to conduct deep neural networks. To solve this dilemma, we propose an optronic convolutional neural network (OPCNN) that can perform SAR ATR tasks directly in optical platform. In OPCNN, all computational operations are implemented in optics with high speed and low energy cost. Electronic platforms are used only to control devices and transmit data information without a massive computational burden. As in digital CNNs, the convolutional layer, downsampling layer, nonlinear activation layer, and fully connected layer are all contained in OPCNN. The simulations demonstrate the feasibility of our OPCNN in solving SAR ATR problems. The good performance in experiments, which achieves 87.4% and 93.8% recognition accuracy on original and denoised moving and stationary target acquisition and recognition dataset, validates the application ability of OPCNN in practical scenario. In one-time recognition tasks, the processing time of our OPCNN is only 0.26 s with the speed of light and the power consumption of our prototype is also far less than digital processer, which is 954 W. Through analysis, our OPCNN obtains the higher processing speed and lower energy cost than digital CNNs with the same structure due to the advantages of optical technology. Also, the scalability of optical structure contributes to build more complex networks to solve complicated dataset without the demand of advanced electronic hardware.

Although the exploitation of mineral areas brings wealth to society, it inevitably leads to the degradation of the surrounding natural environment. To understand and assess the influences of mining activities on the geological and ecological environment, land cover classification in open-pit mine areas (LCCMA) is of great significance. This research proposes an intelligent classification framework for LCCMA based on an object-oriented method and multitask learning (MTL), named the MTL Classification Framework (MTLCF). With the help of MTL, each land cover type in open-pit mine areas obtains its exclusive and receivable object-oriented feature sets using the model-agnostic method. After that, the feature sets are fused with the original images. EfficientNet, a spatial pyramid pooling module, and a global attention upsample module are assembled as the segmentation models with the structure of the encoder and decoder to classify intelligently each land cover type in open-pit mine areas. Finally, the models were trained, and ablation experiments were performed. The experimental results show that our proposed framework -MTLCF was effective for classification in LCCMA, and the overall accuracy and the mean of F1 score for the MTLCF in LCCMA were 85.6% and 86.06%, respectively.

The surface subsidence caused by coal mining will cause serious environmental problems, and an effective monitoring and prediction system is indispensable. Aiming at this phenomenon, a mining subsidence monitoring and dynamic prediction model combining time series stacking differential interferometric synthetic aperture radar (TSS-DInSAR) technology, small baseline subset InSAR technology (SBAS-InSAR), and golden section method-Holt–Winters (GSM-HW) model was proposed. First, based on the nine scene images of Sentinel-1A satellite in the study area, the time series subsidence of 811 work-face in Guobei coal mine from April 5 to July 10, 2021, was obtained using TSS-DInSAR and SBAS-InSAR monitoring technologies, respectively. Then, the original training samples of the GSM-HW model were generated by combining the cumulative surface subsidence results monitored by TSS-DInSAR and SBAS-InSAR, and the monitoring and prediction of surface subsidence were realized using this model. The experimental results show that the application of TSS-DInSAR and SBAS-InSAR monitoring technology to the impact of surface mining in mining areas has certain reliability, and the GSM-HW prediction model can make up for the deficiency of a single HW model in parameter optimization. The maximum fitting accuracy and prediction accuracy of the model for the 10 monitoring pilot sites are 96.9% and 98.4%, respectively, which can provide a reference for the design of surface monitoring and prediction system in the mining area.

For satellite imaging instruments, the tradeoff between spatial and temporal resolution leads to the spatial–temporal contradiction of image sequences. Spatiotemporal image fusion (STIF) provides a solution to generate images with both high-spatial and high-temporal resolutions, thus expanding the applications of existing satellite images. Most deep learning-based STIF methods throw the task to network as a whole and construct an end-to-end model without caring about the intermediate physical process. This leads to high complexity, less interpretability, and low accuracy of the fusion model. To address this problem, we propose a two-stream difference transformation spatiotemporal fusion (TSDTSF), which includes transformation and fusion streams. In the transformation stream, an image difference transformation module reduces the pixel distribution difference of images from different sensors with the same spatial resolution, and a feature difference transformation module improves the feature quality of low-resolution images. The fusion stream focuses on feature fusion and image reconstruction. The TSDTSF shows superior performance in accuracy, vision quality, and robustness. The experimental results show that TSDTSF achieves the effect of the average coefficient of determination (R² = 0.7847) and the root mean square error (RMSE = 0.0266), which is better than the suboptimal method average (R² = 0.7519) and (RMSE = 0.0289). The quantitative and qualitative experimental results on various datasets demonstrate our superiority over the state-of-the-art methods.