Species distribution models (SDMs) have been used extensively in the field of landscape ecology and conservation biology since its origin in the late 1980s. But there is still a void for a universal modeling approach for SDMs. With recent advancements in satellite data and machine learning algorithms, the prediction of species occurrence is more accurate and realistic. Presently, four machine learning and regression-based algorithms, namely, generalized linear model, maximum entropy, boosted regression tree, and random forest (RF) are used to model the geographical distribution of Rhododendron arboreum, which is economically and medicinally important species found in the fragile ecosystem of Himalayas. To establish complex relation between the occurrence data and regional climatic and land use parameters, several satellite products, namely, MODIS, Sentinel-5p, GPM, ECOSTRESS, and shuttle radar topography mission (SRTM), are acquired and used as predictor variables to the different SDM algorithms. The performance evaluation has been conducted using the area under curve (AUC), which showed the best result for Maxent (AUC  =  0.871) and poor result was observed for RF (AUC  =  0.755) among all. The overall prediction confirmed the distribution of Rhododendron arboreum in the mid to high altitudes of central and southern parts of the Garhwal Division. We provide crucial evidence that combining multisatellite products using machine learning algorithms can provide a much better understanding of species distribution that can eventually help the researchers and policymakers to take the necessary step toward its conservation.

A forest is a piece of land dominated by trees; factors like tree density, tree height, canopy cover, land use, and ecological function discriminate different forest types. The canopy cover is instrumental in the identification of forest in the satellite imagery. Multispectral satellite imageries are very useful in feature highlighting. The combination of bands is a powerful tool not only to enhance some earth features but also to suppress others. In our study, a novel index is proposed for the delineation of tree canopy using Sentinel-2 multispectral imagery. A new simple vegetation ratio is proposed for the extraction of the vegetated areas including forest, farms, and other green lands, the existing normalized difference built-up index is modified, and an open surface water index is utilized to formulate a novel tree canopy cover index for the identification of tree canopy.

The monitoring of mosquito breeding habitats requires the production of surface water maps on a regular basis and at a high-resolution using mapping algorithms. To map surface water, several machine learning (ML) algorithms were evaluated, taking advantage of frequently available synthetic aperture radar imagery from the Sentinel-1 mission with a 10-m spatial resolution and a large dataset of field observations of the water state (inundated/dry) in rice paddies and wetlands. One-class support vector machine, one-class self-organizing map, and multilayer perceptron with automatic relevance determination (MLP-ARD) algorithms were trained and assessed to examine their accuracy in detecting surface water. Results show the robustness of the MLP-ARD algorithm, which provides an overall accuracy of 0.974 for a single date and 0.892 for a 5-month study period from May to September. The accuracy of water detection was found to be mainly affected by the presence of dense and high vegetation in inundated fields and the presence of floating vegetation or algae.

Deep learning has been widely used in hyperspectral image (HSI) classification tasks and has achieved good results in various fields. However, the large gap between a large number of parameters to be adjusted and the limited labeled samples of HSI will greatly increase the risk of model overfitting and inevitably consume a large amount of time and memory. A lightweight hybrid channel expansion and squeeze network is proposed to solve the small sample set problem. First, a channel expansion and squeeze module (CESM) based on depthwise convolution is designed to obtain higher fine-grained features by increasing the width of network channels. At the same time, the introduction of efficient channel attention module can emphasize the importance of each channel and effectively improve the representation of CESM compression output. Second, multidimensional CESM is used to reconstruct the entire feature extraction process and further reduce the model complexity while maintaining the intrinsic physical characteristics of HSI. Finally, the extracted high-level abstract features are imported into the global average pooling classifier to obtain the prediction probability of each category. Experimental results on Indian Pines, Pavia University, and Salinas Scene datasets show that the accuracy of the proposed method can reach 98.66%, 99.61%, and 99.84%, respectively, under limited labeled samples, and the computational cost is significantly lower than other advanced methods.

Oblique ionogram inversion is an important tool for understanding the ionosphere. Vertical plasma density profiles are the most common inversion results for the majority of the inversion algorithms. However, the inversion of the dynasonde ionogram provides not only the vertical plasma density profile, but also two tilts profiles directly related to the horizontal gradients of the plasma density. The main reason is that the dynasonde ionogram inversion uses a 3D numerical ray-tracing technique that considers the geomagnetic field and the horizontal inhomogeneity of the ionosphere. The theory and the ray-tracing performance of the 3D numerical ray-tracing method are presented. The ray-tracing method uses a variable-step strategy to speed up the simulation. And three plasma density models with traveling ionospheric disturbances and small-scale irregularities and without them are considered when using the presented ray-tracing method to simulate oblique ionograms. The simulation results show that the ionosphere disturbances can eventually be embodied in the simulated oblique ionograms.

The temporal principal component analysis (TPCA) method in the field of geology can extract the temporal and spatial features of spatial-temporal data, and the ground subsidence in the Los Angeles area has typical temporal and spatial features. Based on the ground subsidence data of the Los Angeles area from 2016 to 2019 obtained by small baseline set (SBAS)-InSAR technology, TPCA and wavelet analysis were used to analyze the spatial and temporal evolution characteristics of ground subsidence in the Los Angeles area. Several results were found. First, the first principal component obtained by TPCA analysis reflected the spatial and temporal evolution characteristics of ground subsidence at that time series stage. Second, after wavelet analysis and the temporal trends of the eigenvectors of the second and third principal components, it can be seen that the second and third principal components mainly correspond to the periodic deformation and that the maximum amplitude of the periodic oscillation is 10 mm. Third, the correlation analysis of the eigenvectors of groundwater level, rainfall, and the fourth principal component reveals that there is a time phase delay between deformation and groundwater level change and rainfall.

Rapid and accurate tree crown detection is significant to forestry management and precision forestry. However, the variability of the data is challenging for traditional deep learning-based methods in cross-regional scenarios. To solve the problem of low accuracy of existing methods due to the distribution characteristics of tree crowns and the special background in which they are located, we propose a multi-adversarial domain adaptive (DA) crown detection model, which saves much of labeling efforts needed for covering different features in different regions. First, we embed a multilevel transferable attention mechanism to extract the multiscale transferable features of the tree crown and combat the negative transfer caused by the background. Second, a multi-adversarial instance alignment module with entropy regularization and a weighting factor is designed to match the cross-regional data distributions to reduce the misdetection and misidentification of complex backgrounds. The above-mentioned adaptations are integrated into a two-stage object detector to realize end-to-end training. We conduct comprehensive experiments with four transfer tasks. Our method improves F1-score to 95.68%, 83.12%, 78.56%, and 66.89% and performs 3.61% and 15.38% better than the state-of-the-art domain adaptation approaches without extra inference costs. The results demonstrate the great potential of our method for DA tree crown detection, which provides a reference for further research on forestry detection.

In recent years, remote sensing image observation technology has developed rapidly. Extracting coastlines from remote sensing images has become an indispensable means of port area measurement. Port segmentation from remote sensing images is an important method of coastline extraction and measurement. The remote sensing images used are panchromatic remote sensing images. Due to complex remote sensing image scenes and the large difference in feature information at different scales, traditional segmentation methods cannot perform effective extraction, and it is difficult to accurately segment the coastline in remote sensing images. We propose a multiscale feature fusion network for automatic port segmentation from remote sensing images. First, to reduce the redundant parameters and complex operation problems in traditional convolutional neural networks, we propose using MobileNetv2 as the base network for feature extraction to achieve a lightweight model. Then aiming at the feature differences of remote sensing images at different scales, we present atrous convolution as a convolution method for the entire network and combine a multiscale feature fusion method to extract the features of remote sensing images and improve the feature extraction ability. To reduce the problem of ships calling at the port being easily mistaken for port area and causing false segmentation, we propose a method of eliminating the ship area to reduce the interference. Finally, the comprehensive evaluation of a large number of Google port remote sensing data shows that compared with existing methods, the proposed method has the characteristics of being lightweight and having high precision.

Satellite observations, used worldwide in the atmospheric sciences, are extremely useful for providing aerosol information within a wide spatial range. However, the coverage of aerosol data by satellite observations is sometimes of inferior quality because of the effects of surface reflectivity and clouds. To fill the gaps in aerosol optical depths (AODs) retrieved from geostationary ocean color imager observations, this study applies operational statistical techniques, including radial basis functions (RBFs) with four different weightings (i.e., linear, multiquadric, thin-plate, and inverse), Poisson, and ordinary Kriging. Based on computation time and accuracy of the individual gap-filling techniques, Poisson and the liner RBF are selected as the two best methods and then averaged with weights using one-dimensional weighted average (1D-WAVG) and two-dimensional weighted average (2D-WAVG) root mean square errors. All methods produce reliable results, yielding a correlation coefficient between 0.74 and 0.87 over the entire research domain. Out of the individual techniques, the Poisson, with an initial estimation from a zonal mean of AODs, is the most accurate with the lowest computational costs, even for a large number of missing pixels and most regions, excluding East China (EC). The Poisson’s high bias over EC is compensated in 1D- and 2D-WAVGs by taking more accurate estimations of the linear RBF than those of the Poisson over the region. If we consider 1D- and 2D-WAVGs in our analysis, the highest correlation is obtained from the 2D-WAVG over all regions. Because of its reliability and fast computation time, applying the 2D-WAVG can be a good solution to provide spatial–temporal continuous aerosol information. In addition to air pollution studies, such as real-time air quality predictions, estimation of ground-level particulate matter concentrations, and other applications, the fast and operational gap-filling technique can also be expanded to remote sensing data obtained from satellite observations to provide helpful and useful information for the public.

Hyperspectral images (HSIs) have recently been exploited in several aspects as HSIs contain many contiguous and narrow discriminative spectral bands. The problem of dimensionality is a significant dilemma for HSIs due to there being plenty of irrelevant and redundant spectral bands and highly correlated bands that lead to Hughes phenomenon. To this end, we present an approach to selecting the most informative and relevant spectral bands for HSI dimensionality reduction using the Krill Herd (KH) algorithm. Moreover, KH is a heuristic search method that seeks to reach the optimum global solution within the search space and effectively evade falling into the local optima. Then an edge-preserving filter was employed to extract the spatial features while reducing noise and obtaining a suitable smoothing that improves the classification performance. Finally, the support vector machine classifier was performed at the pixel level for HSI classification. Furthermore, the proposed work was compared with the harmony search, genetic algorithm, bat algorithm, particle swarm optimization, and firefly algorithm. The experimental results demonstrated outstanding performance with an overall accuracy equal to 96.54%, 98.93%, 99.78%, and 98.66% on four hyperspectral datasets: Indian Pines scene, Pavia University scene, Salinas scene, and Botswana scene, respectively.

Aerosol optical depth (AOD) is an important indicator for air pollution monitoring. We obtain AOD from the GF-6 satellite, which was launched on June 2, 2018 by China. Beijing is selected as the study area. Based on the deep blue algorithm, the moderate-resolution imaging spectroradiometer surface reflectance product is used to correct and establish the surface reflectance library, and the AOD is retrieved from the GF-6 wide field of view (WFV) blue band. So, the application of the deep blue algorithm in a GF-6 WFV camera is realized. The aerosol is divided into three models: a custom model, a continental model, and an urban model in second simulation of the satellite signal in the solar spectrum model. To improve the accuracy of retrieval results, we recalculated the aerosol component concentration ratio based on GF-6 WFV data and ground observation as the custom model. The retrieval results of the three aerosol models show that the central and southern urban areas are higher than the mountainous areas in the northwest, indicating that human emission is the main source of air pollution in Beijing. The retrieval results of the custom model, continental model, and urban model are compared with the ground observation. The error of urban model retrieval is large, and the determination coefficient R² is only 0.3671. The results of the custom model and continental model are similar. The R² of the custom is 0.8942, which is higher than that of the continental (0.8872). We analyze the relationship between the retrieved AOD of the three aerosol models and top of atmosphere (TOA) reflectance and find that the custom AOD has the best sensitivity to TOA reflectance, whereas the urban AOD has the worst sensitivity. So, the custom aerosol model is closer to the real aerosol situation in Beijing.

Remote sensing scene classification has received more and more attention as important fundamental research in recent years. However, the redundant background information and complex spatial scale variability of remote sensing scene images make the existing convolutional neural network models, which mainly concentrate on global features, perform poorly. To effectively alleviate these problems, we proposed an MSRes-SplitNet model based on multiscale features and attention mechanisms for remote sensing scene image classification. First, MSRes blocks are constructed for the extraction of multi-scale features. Then, the multi-channel local features are fused by the Split-Attention block. Finally, the global and local feature information is aggregated by convolution, thus obtaining multi-scale features while alleviating the small-sample learning problem. Experiments are conducted on three publicly available datasets and compared with other state-of-the-art methods, showing that the proposed method MSRes-SplitNet has better performance while effectively reducing a large number of parameters.

This research considers the percentage of time that geosynchronous earth orbit (GEO)- and HEO-based optical sensors can track dim cislunar targets in L2 near-rectilinear Halo orbit (NRHO), taking into account observer constellation configuration—number of observer satellites and their orbits—as well as the sensors’ lunar exclusion angle (LEA). Simulations are created using systems tool kit (STK) to model the targets, observers, sensors, and all relevant orbits. Outputs of the simulations include target-sensor-Moon and target-sensor-Sun angles, as well as target signal-to-noise ratios (SNRs). Outputs are used for analyzing whether the observer constellation has access to the target (i.e., at least one sensor is not lunar- or solar-angle excluded), and the quality of the observation (i.e., the achieved SNR for every time step of the simulation). This research finds that LEA is the most significant variable influencing the extent to which tracking custody can be maintained over L2 NRHO and Halo targets. While utilizing larger OCCs does tend to increase the probability that targets will be trackable, this effect is insignificant relative to the impact of utilizing a smaller LEA. If a 1 deg LEA is implemented for these sensors, target tracking is successful through roughly 50% of the synodic period for targets in either orbit. Increasing the LEA beyond 3 deg decreases tracking uptime by a very significant margin for Halo targets, but not as significantly for NRHO targets.

In recent studies, remote sensing object detection methods based on deep learning have emerged as a primary concern in environmental monitoring, military investigation, and hazard response. However, many difficulties, such as complex backgrounds, dense target quantities, large-scale variations, and non-uniform distribution, lead to many parameters and complex network structures, thus limiting the accuracy of the detector and slowing the inference speed. To address these issues, we propose a lightweight and efficient object detector for remote sensing images. First, an asymmetric convolution with the visual attention mechanism is reconstructed to decrease the complexity and strengthen the feature representation ability. Then, an adaptive feature selection structure is designed to extract discriminative feature information, which can adaptively model the shapes of objects by introducing deformable convolution to obtain a stronger geometric feature representation. To reduce information loss across different channels and spatial locations, a hybrid receptive field module is also proposed to increase the receptive field model by mixing dilated convolutional layers with different dilation rates. Finally, experimental results on the DIOR dataset show that our approach significantly improves detection accuracy and running speed.

In the vastly growing field of remote sensing, automated field spectrometer systems are of high value as ground reference, e.g., when measuring reflectance or solar-induced fluorescence. To calculate reflectance of a target surface, these instruments are typically equipped with cosine receptors to measure irradiance corrected for the cosine function of the incident lights angle. However, available cosine receptors suffer from poor cosine characteristics or low light throughput. We created an innovative cosine diffusor concept, which follows simple design rules and allows iterative manufacturing of a diffusor with accurate cosine response. Using a specifically constructed goniometer, we characterized common cosine diffusors and compared their performance against the innovative type. The new cosine receptor was able to produce a near-perfect cosine response even at lower light angles and higher light throughput. The novel diffusor concept uses robust materials, which promise long-term stability in outdoor use. The presented research improves the collection of precise, hyperspectral irradiance measurements toward recording of extended time-series.

Measuring the microscopic and macroscopic optical properties of smoke aerosols is important in column concentration retrieval by remote sensing for environmental monitoring. Based on the radiation transfer theory, we constructed a physical model to correlate microscopic and macroscopic optical properties of smoke aerosols. According to the experiment, the mass extinction coefficient and mass angular scattering coefficient of smoke aerosols in the visible-near infrared (400 to 2400 nm) show the strongest extinction and scattering effects on the visible light, which are negatively correlated with wavelength. They were applied to smoke column concentration retrieval based on Landsat8/operational land imager and MODIS images by the established model. Then the optimal bands for smoke detection were analyzed. It indicates that the surface underlying the smoke has impact on the optimal band of smoke detection, and red-light band is optimal for vegetation background.

To improve the accuracy and operation efficiency of lithological classification using a thermal infrared airborne hyper-spectral imager (TASI) data, an innovative combinatorial algorithm (PCA-QPSO-LSSVM) based on principal component analysis (PCA), quantum-behaved particle swarm optimization (QPSO), and least-squares support vector machines (LSSVM) is proposed. After pre-processing, the emissivity data was extracted from TASI data, and 27 types of lithological units were selected in comparison with the measured spectrum and Johns Hopkins University spectrum library. The PCA was used for statistical analysis, and the appropriate n principal components were selected instead of the original 32 bands of TASI emissivity data. Based on the LSSVM classification algorithm, the improved QPSO algorithm was used to optimize the regularization parameter γ and the kernel function σ² in the classification process. Finally, test samples selected from TASI data were classified by PCA-QPSO-LSSVM. The results show that, compared with the traditional LSSVM algorithm, PCA-QPOS-LSSVM has a higher recognition accuracy and greater operation efficiency. The field verification of the study area was carried out in LiuYuan town, GanSu province, China, and the accuracy of the field verification was 74.36% for the classification result, which is more consistent with the actual lithological distribution compared with the LSSVM algorithm. However, there are some flaws in this paper, such as the low classification accuracy in Moyite and plauenite. In general, The PCA-QPSO-LSSVM algorithm is a practical algorithm for lithological classification of TASI data.

The Amazon–Cerrado transition region has a high rate of deforestation for agriculture, livestock, logging, and for the increase in urban areas of cities. The development of cities in this region in recent decades has increased the deforestation process and modified the local microclimate and the exchange of energy and water between the surface and the atmosphere. However, the magnitude of the change in energy and water exchanges due to urbanization is unknown for most Brazilian cities, especially in the Amazon–Cerrado region. Thus, the objective was to evaluate the spatiotemporal change in energy balance and in evapotranspiration (ET) patterns over three decades of urbanization in Sinop. The surface energy balance (SEB) and ET were estimated by the SEB algorithms for land with Landsat 5 Thematic Mapper images from the Sinop region, Mato Grosso, Brazil, in 1985, 2000, and 2010. The estimated SEB and ET were validated with measurements in a flux tower in the Amazon–Cerrado transition forest 50 km from the urban area of Sinop. The components of the SEB, except the soil/storage heat flux (G), had a strong correlation, and the ET had a moderate correlation with measurements in the flux tower. The surface albedo (a_sup), land surface temperature (LST), sensible heat flux (H), and G of the study area increased over the years, while the normalized difference vegetation index, latent heat flux (LE), and ET decreased over the years. Buildup and grassland areas had lower values of NDVI, Rn, LE, and ET and higher values of a_sup, LST, H, and G than forest areas. Therefore, the urbanization of Sinop has decreased available energy for ET and increased the energy to heat the air and soil.

Most remote sensing sea ice classification methods use single-source remote sensing data, such as synthetic aperture radar (SAR) data and optical remote sensing data. SAR data contain rich sea ice texture information, but the data are relatively single, making it difficult to distinguish detailed sea ice categories. Optical data include abundant spatial-spectral information, but they are often affected by clouds, fog, and severe weather. Hence, given the limitations of single-source data, the remote sensing sea ice classification accuracy cannot be further improved. A sea ice classification method based on deep learning and multisource remote sensing data fusion is proposed utilizing an improved densely connected convolutional neural network (DenseNet) to mine and fuse the multilevel features of sea ice. According to the characteristics of SAR data and optical data, a dual-branch network structure based on an improved DenseNet is employed for feature extraction, and a squeeze-and-excitation attention mechanism is introduced to weight the fusion features to further enhance the feature weights that can effectively distinguish different types of sea ice. The fully connected network is used to perform the deep fusion of features and classify sea ice. To verify the effectiveness of the proposed method, two sets of sea ice data are utilized for sea ice classification. The experimental results show that the proposed method fully excavates and fuses the multilevel characteristics of heterogeneous data using the improved dual-branch network structure, leverages the complementary characteristics of SAR data and optical data, significantly increases the sea ice classification accuracy, and effectively improves the influence of cloud cover on the sea ice classification accuracy. Compared with the typical single-source data classification method and other heterogeneous data fusion classification methods, the proposed method achieves superior overall classification accuracy (98.49% and 98.58%).

Spectral dimensionality reflects the intrinsic information content of imaging spectroscopy data. Prior work investigating spectral dimensionality has mostly either focused within a limited study area or at a single spatial resolution. Despite multiple spectrometer satellite missions in the development, the inherent information content of imaging spectroscopy images is still largely unknown. Leveraging data collected by the Global Airborne Observatory (GAO), we investigated the spectral dimensionality of common landscape types (i.e., urban, suburban, rural, agriculture, forest, shrub/grass, and wetland) across several spatial resolutions (i.e., 1 to 5, 10, 20, and 30 m). The results corresponded well to the size of the objects in the target landscape. At native spatial resolution (1 to 5 m), anthropogenic images dominated by fine-scale spectrally diverse objects usually had higher dimensionalities than natural images. But as resolution coarsened, the spectral dimensionality of images containing natural landscapes outranked anthropogenic ones, except for large-scale agriculture. Besides, there was a surprising difference in the magnitude of dimensionality decreasing trend between anthropogenic and natural scenes. Urban, suburban and rural cover types experienced dramatic drops at coarser resolutions, whereas vegetated landscapes preserved most of their dimensionality at 30-m spatial sampling. In addition to revealing the typical spectral dimensionalities of common landscapes, we provided insights into how much information might be achievable from spectroscopic analyses. Results of this study guide future imaging spectroscopy data collection and scientific applications.

Remote sensing image fusion aims to enhance the spatial resolution of multispectral (MS) images by extracting spatial detail information from panchromatic (PAN) images. To preserve the spectral information of MS images and enhance the spatial resolution, an innovative remote sensing image fusion method is proposed here. The proposed method is based on an image matting model, the non-subsampled shear wave transform (NSST) and a parameter adaptive pulse coupled neural network (PA-PCNN). In the proposed method, an intensity component, I, is first obtained by adaptively weighted averaging over each band of the MS image. An image matting model is then applied. There, I is used as the α channel to estimate the foreground color, F, and the background color B. Then, the I and PAN images are decomposed into low and high-frequency coefficients by NSST. Two fusion rules are proposed here and applied to fuse the low and high-frequency coefficients, respectively. The fused low and high-frequency coefficients are inversely transformed to get a fused image, which is used as the α channel in the reconstruction of the image. Finally, the F, B and fusion images are combined to reconstruct the final high-resolution fused image. The experimental results show that the proposed approach achieves superior results in both subjective visual effects and objective assessment. It can enhance the spatial resolution and effectively retaining the spectral information of MS images.

The unsupervised domain adaptation (UDA) of aerial image classification is of great significance in the traffic management and public safety applications. Due to the differences in the distribution of source and target domain datasets from different sensors and cities, and the acquisition difficulty and high annotation cost of target domain datasets, existing domain adaptive methods have difficulty obtaining high accuracy and stability. In this work, a modified cycle generative adversarial network (CycleGAN) (MGAN) based on the multiscale residual block (MSRB) and multipooling coordinate attention (MPCA) modules is proposed to generate fake images, which is the most critical stage of the UDA method. In the MGAN, the MSRB module can extract the multiscale information of the image, and the MPCA module can fuse the position and channel information in the image. These two modules better extract the features of edges, regions and semantic classes, which makes the generated fake images better retain the information of the source domain and possess the style of the target domain. Otherwise, the Deeplabv3+ model is used to obtain a pretrained and final segmentation model in the first and third stages. In the experiment, the proposed method is validated on the aerial images of the Potsdam and Vaihingen datasets in the International Society for Photogrammetry and Remote Sensing (ISPRS) two-dimensional semantic labeling benchmark and improves overall accuracy from 52% to 60% on the segmentation of the target domain, compared with the existing optimal method; it also achieves an accuracy of 77% for buildings. The results demonstrate that the proposed method can achieve a higher accuracy and stability.

To timely detect landslide hazards to start emergency rescue, an improved Faster R-CNN algorithm is proposed for remote sensing image landslide detection. First, the gamma transform and Gaussian filtering methods of image enhancement are used to improve the quality of the images. Second, the effect of batchsize size on the model is eliminated using the group normalization method. Finally, multiscale feature fusion is performed by adding a feature pyramid network structure to optimize the extracted landslide small target features, and then the backbone network is set as deep residual shrinkage network 50 to make the model more focused on information useful for landslide detection. The experimental results show that the improved model improves the accuracy rate as well as the average precision by 8.8% and 8.4%, respectively, compared with the unimproved Faster R-CNN, and compared with the first-stage models, such as you only look once version 4 and single-shot detector, which verify the superiority of the model in our study and can detect landslide targets well.

Algal blooms are pervasive in many freshwater environments and can pose risks to the health and safety of humans and other organisms. However, monitoring and tracking of potentially harmful blooms often relies on in-person observations by the public. Remote sensing has proven useful in augmenting in situ observations of algal concentration, but many hurdles hinder efficient application by end users. First, numerous approaches to estimate aquatic chlorophyll-a are available and can produce inconsistent results. Second, lack of quantitative in situ observations limits opportunities to train models for specific waterbodies, such that models developed for other systems must be used instead. We (1) implement univariate and multivariate logistic regression models to estimate the probability that aquatic chlorophyll-a concentrations exceed an accepted threshold beyond which harmful effects become likely and (2) evaluate the use of visually classified bloom/no-bloom satellite imagery to augment in situ training data. Using a binary classification of aquatic chlorophyll-a exceeding 10 μg / L, we found that (1) logistic regression models were ∼80 % accurate, (2) univariate models trained with visually classified data produce nearly the same accuracy (79%) as models trained with in situ observations (80%), and (3) augmenting in situ chlorophyll-a observations with visual classifications outperformed (82% accuracy) models trained on in situ observations alone (80% accuracy). These results provide a framework for evaluating multiple spectral indices in retrieving algal bloom presence or absence and illustrate that training data derived directly from satellite imagery can be useful in augmenting in situ observations.

Comprehensively assessing the ecosystem health of islands provides very important information to assist managers with ecological restoration efforts and to develop protection policies and recommendations for related marine administrators. However, evaluation of the ecosystem health of islands usually involves a single ecosystem health value, which does not assess the situation from the spatial and temporal perspectives. The present study analyzed ecological conditions and the effects of ecological restoration on Qinshan Island and the surrounding sea area using geographic information systems and remote sensing to assess the ecosystem health of Qinshan Island based on vigor-organization-resilience and water, sediment, and bio-ecological quality models. The results indicated that, from 2009 to 2018, the ecosystem health index of the land area increased from 0.351 to 0.567. The ocean health index of the surrounding sea area improved from 0.91 in 2009 to 0.82 in 2018. The island comprehensive ecosystem health indices were 0.5187 and 0.6427 in 2009 and 2018, suggesting that the ecosystem health of Qinshan Island changed from moderately degraded to healthy. The critical underlying factor for the improvement of the ecosystem health indices for Qinshan Island is the ecological restoration projects performed during 2011–2017, which effectively improved the environmental conditions.

Plateau pika (Ochotona curzoniae) is a key species of alpine meadow ecosystem in Qinghai–Tibet Plateau. In this study, an observation and classification system of plateau pika disturbance intensity was constructed by multidimensional stereoscopic surveying, visual interpretation, geostatistics, principal component analysis, and clustering analysis, which can monitor and evaluate the disturbance of plateau pika activities comprehensively, scientifically, and efficiently in the head-water region of the Yellow River. The results show that (1) digital elevation model, digital orthophoto model, and three-dimensional real scene model can extract the surface morphology information changed by pika activities such as the number of pika holes and molehills, the area and volume of molehills, the elevation variation coefficient, the surface relief, and roughness; (2) The main load factors of the plateau pika disturbance intensity index (PPDII) are the surface roughness, the area and number of the molehills, and the number of mousehole; (3) The kappa value of PPDII clustering verification was 0.866, and the clustering analysis results were consistent with reality characteristics of plateau pika activity and distribution; (4) The disturbance in Laji Mountain is mainly slight and mild, accounting for 47.6% of sampling area, and the disturbance in Henan County is mainly medium and severe, accounting for 42.6% of sampling area. The results can provide reference data for the restoration and management of alpine meadow degradation in the head-water region of the Yellow River.

The proliferation of grassland rodents has aggravated the degradation process of desert grassland. Additionally, it carries many viruses that threaten human and animal health. Accurate spatial distribution in grassland between rodent populations, vegetation, and bare soil is essential for developing rodent control measures. However, the traditional survey method of grassland rodent pest information is time-consuming and costly, and the period is long. In addition, satellite remote sensing cannot meet the accuracy requirements for the identification of grassland rat holes due to the limitation of spatial resolution. To realize intelligent grassland rodent infestation monitoring, this paper adopts an unmanned aerial vehicle hyperspectral remote sensing platform for data acquisition. Meanwhile, a transformer attention network (TAN) is proposed for grassland rodent infestation information extraction. The network adopts a two-stage feature extraction structure that effectively improves the classification performance of the model. In each stage, first, local features are extracted by a fixed convolution kernel to enhance detailed texture features; second, the extracted local features are refined using the contour convolution module to enrich feature information at the edges of the feature map; finally, the transformer attention module is used to focus on the global pixels, thus suppressing background information and enhancing effective information output. The results show that the overall accuracy (OA), average accuracy, and kappa coefficient of the TAN network can reach 97.71%, 98.44%, and 0.9538, respectively. Compared with several networks, such as two-dimensional CNN, three-dimensional CNN, HybridSN, and CTN, the OA values of the TAN network were improved by 2.59%, 2.45%, 2.94%, and 1.03%, respectively. The results of this study effectively improve the efficiency of the grassland rodent information survey and provide a solid theoretical basis for the investigation and statistics of grassland rodent infestation.

The state-of-art simultaneous localization and mapping (SLAM) methods have been demonstrated of satisfactory results in static environments. However, most of the existing methods cannot be directly applied in dynamic environments to account for moving objects. To solve the practical problem of SLAM in dynamic environments, an RGB-D SLAM method is developed based on moving objects removal and dense map reconstruction. Specifically, deep learning-based instance segmentation is performed to obtain the semantic information which is furthermore combined with multiview geometry theory to detect possible moving objects. Kalman filter and feature fusion algorithms are further used for moving objects tracking to improve the detection accuracy. Then the camera pose is estimated using the static feature points after motion removal to realize reliable localization in dynamic environments. Out of concern for the infeasibility of navigation and perception with sparse feature maps, each-single-view of dense map is constructed using the RGB image and the depth map collected by the RGB-D camera after motion removal. The dense environmental map is finally reconstructed through the registration of point cloud of selected key image frames. Experiments are conducted on public datasets and real scene data to evaluate the performance of the proposed method. Compared with the state-of-art method ORB-SLAM2, the absolute trajectory error and the relative pose error obtained by the proposed method are significantly reduced by at least 91%, and the localization accuracy reaches 0.001 m.

We propose an approach for millimeter wave radar targets classification based on the concatenated spectrogram and range-Doppler features. A publicly available dataset that contains raw single radar data for fast- and slow-walking people is utilized. First, the received channels are summed to obtain a higher signal-to-noise ratio. Next, range-Doppler and spectrogram plots are obtained after the two-dimensional fast Fourier transform of raw radar signals. Then the spectrogram dataset is normalized and augmented using a mixup algorithm and fed into the proposed encoder and decoder structures sequentially for the classification task. In addition to the bidirectional long short-term memory layer, dropout and batch normalization layers are added for regularization in the proposed network. Parameter optimization is done for each network for a certain number of parameters. Finally, the results show that with the proposed approach, a 0.961 mean F1 score is obtained, and it outperformed some state-of-the-art methods, such as convolutional neural networks (CNNs) and hybrid CNNs–long short-term memory networks. An ablation study of the proposed method is also presented.

Remotely sensed water properties are important for a variety of applications, including validation of Earth systems models (ESMs), habitat suitability models, and sea level rise projections. For the validation of next-generation, high or multi-resolution (30 to 60 km) ESMs in particular, the usefulness of operational forecasting products and directly observing satellite-based sensors for validation is limited due to their temporal availability and spatial resolution of <1 year (in some cases) and >30 km², respectively. To address this validation data gap, we developed a data-driven model to produce high-resolution (<1 km²) estimates of temperature, salinity, and turbidity over decadal time scales as required by next-generation ESMs. Our model fits daily moderate resolution imaging spectroradiometer (MODIS) Aqua reflectance data to surface observations (<1 m depth) from 2000–2021 in the Chesapeake Bay. The resulting models have similar error statistics as prior efforts of this type for salinity [root mean square error (RMSE): 2.3] and temperature (RMSE: 1.8 C). However, unlike prior efforts, our model is designed as a pipeline meaning that it has the advantage of producing predictions of water properties in future time periods as additional MODIS data becomes available. We also include novel geographically aware predictive features insofar as they capture geographic variation in the influence of flow and surface water exchange in upstream coastal watersheds.

The existing deep learning-based remote sensing image super-resolution (SR) reconstruction methods have some problems, such as insufficient feature extraction and detail loss. We propose an efficient and fast deep convolutional neural network (CNN) single-image SR model, which achieves relatively advanced reconstruction performance. The number of layers and filter layers of CNNs are completely optimized, and the computing cost is significantly reduced. The global and local features of the original image are extracted by multiscale features; then, contextual information is enhanced in the inference network, and the reconstruction is done in advance of the dimensionality reduction operation to reduce the computational effort. The reconstruction performance is further improved using skip connection network in the whole network to solve the inefficiency of information in the transmission process. The experimental results show that the algorithm is 0.32 to 4.38 dB higher than other algorithms in peak signal-to-noise ratio value, the structure similarity index measure value reaches 0.9618, and the reconstructed image extracts more features and is richer in details.

In bistatic inverse synthetic aperture radar (Bi-ISAR) imaging of targets with rotating components, the micro-Doppler effect generated by rotating components can produce interference bands in azimuth resolution, seriously affecting the imaging results and causing difficulties in target recognition. A unique Bi-ISAR target micro-Doppler separation method with rotating components is proposed by introducing the complex variational mode decomposition (CVMD) method. First, the echo signal is decomposed into different intrinsic mode functions (IMFs) by CVMD for the Bi-ISAR imaging target with rotating parts. Then, IMFs are divided into main and micromotion parts by setting the energy threshold. The IMFs belonging to the main part are extracted, and the imaging is performed using the range-Doppler algorithm. The simulation results show that CVMD outperforms the complex empirical mode decomposition algorithm and the complex local mean decomposition algorithm in the Doppler separation of Bi-ISAR targets with rotating components. In addition, CVMD exhibits good robustness and provides innovative ideas for future research on micro-Doppler separation technology.

Cloud detection is an important prerequisite for remote sensing image application. Any remote sensing image from which the information of ground object could be obtained will inevitably be preprocessed on cloud occlusion. In the traditional method, the segmentation of cloud and its shadow will be affected by the complex background. In the detection process, due to insufficient information extraction, misjudgment often occurs, and the cloud boundary processing is also very rough. In order to improve the accuracy of cloud and cloud shadow segmentation, we propose a multilevel feature context semantic fusion network. The network takes the residual network as the backbone networkand adopts the structure of encoding and decoding as a whole. In the model, we introduce the multibranch residual context semantic module, the multiscale convolution subchannel attention module, and the feature fusion upsampling module to strengthen the feature extraction, refine the cloud and cloud shadow edge information, and enhance the actual segmentation ability of the model. The experimental results show that the proposed method is more accurate than the previous network in the segmentation of cloud and cloud shadow and has good generalization ability on other datasets, which is of great significance for the study of cloud.

Generation and manipulation of digital images based on deep learning (DL) are receiving increasing attention for both benign and malevolent uses. As the importance of satellite imagery is increasing, DL has started being used also for the generation of synthetic satellite images. However, the direct use of techniques developed for computer vision applications is not possible, due to the different nature of satellite images. The goal of our work is to describe a number of methods to generate manipulated and synthetic satellite images. To be specific, we focus on two different types of manipulations: full image modification and local splicing. In the former case, we rely on generative adversarial networks commonly used for style transfer applications, adapting them to implement two different kinds of transfer: (i) land cover transfer, aiming at modifying the image content from vegetation to barren and vice versa and (ii) season transfer, aiming at modifying the image content from winter to summer and vice versa. With regard to local splicing, we present two different architectures. The first one uses image generative pretrained transformer and is trained on pixel sequences in order to predict pixels in semantically consistent regions identified using watershed segmentation. The second technique uses a vision transformer operating on image patches rather than on a pixel by pixel basis. We use the trained vision transformer to generate synthetic image segments and splice them into a selected region of the to-be-manipulated image. All the proposed methods generate highly realistic, synthetic, and satellite images. Among the possible applications of the proposed techniques, we mention the generation of proper datasets for the evaluation and training of tools for the analysis of satellite images.

Significant radiometric differences and weak grayscale correlations exist between optical and SAR images. As a result, there are severe spectral and spatial distortions in the fused images. We propose a fusion method of optical and SAR remote sensing images that couples the gain injection method and the guided filter. The proposed method is based on the fusion framework of generalized intensity-hue-saturation non-subsampled contourlet transform, and the gain injection is used for the low-frequency coefficient fusion to reduce the spectral distortion. Then, the divergence is used as the activity measure operator to calculate the initial weight template for the high-frequency coefficients. The guided filter is used to optimize the edge details of the initial weight template. The fused high-frequency coefficients are obtained by weighted average. Through comparison experiments with existing fusion methods, the results show that the proposed method has the best quality of fusion and the proposed method has the best performance.

Infrared small target detection (IRSTD) plays an essential role in many fields such as air guidance, tracking, and surveillance. However, due to the tiny sizes of infrared small targets, which are easily confused with background noises and lack clear contours and texture information, how to learn more discriminative small target features while suppressing background noises is still a challenging task. In this paper, a context-aware cross-level attention fusion network for IRSTD is proposed. Specifically, a self-attention-induced global context-aware module obtains multilevel attention feature maps with robust positional relationship modeling. The high-level feature maps with abundant semantic information are then passed through a multiscale feature refinement module to restore the target details and highlight salient features. Feature maps at all levels are fed into a channel and spatial filtering module to compress redundant information and remove background noises, which are then used for cross-level feature fusion. Furthermore, to overcome the lack of publicly available datasets, a large-scale multiscene infrared small target dataset with high-quality annotations is constructed. Finally, extensive experiments on both public and our self-developed datasets demonstrate the effectiveness of the proposed method and the superiority compared with other state-of-the-art approaches.

Salient object detection (SOD) in remote sensing images (RSIs) is a highly practical task. However, scale variations of salient objects and the diversity of salient objects in RSIs pose challenges for detection. To address these issues, an attention-based pyramid decoder network (APDNet) is proposed for SOD in RSIs. The APDNet consists of three key components. First, a multiscale attention block is constructed to extract multiscale information and relations between salient objects, suppressing the distraction of variations of object types and scales. Second, a pyramid decoder structure is designed to take full advantage of multilevel features by gradually fusing features from two adjacent layers for result prediction. This feature fusion allows APDNet to efficiently exploit multilevel feature information, thus enabling a better feature representation. Finally, a bidirectional residual refinement module is proposed to enhance the structural integrity and boundary retention of initial saliency predictions. Extensive experiments on two public datasets demonstrate the superiority and effectiveness of the proposed APDNet against other compared state-of-the-art methods.

Mapping changing land use and land cover (LULC) is important for land management and environment analysis. We tried to build deep learning models to classify LULC over time at an agricultural expansion area in Matopiba region, Brazil with MCD43A4 V006 moderate resolution imaging spectroradiometer (MODIS), and Sentinel-2 multispectral instrument (MSI) time series data. We collected time series MODIS data and Sentinel-2 A/B MSI data from 2015 to 2020 and prepared small patches with containing blue, green, red, near-infrared, and shortwave-infrared-1 bands as features. Then the both datasets were used to build and train the convolutional neural network (CNN) model, the CNN gate recurrent unit (CNN-GRU) model, and the CNN long short-term memory (CNN-LSTM) model, respectively. We evaluated these three trained models with ground truth data, and the CNN-LSTM model (overall accuracy: 91.29% from MODIS data and 89.47% from Sentinel-2 data) was better than the CNN-GRU model (overall accuracy: 89.19% from MODIS data and 88.61% from Sentinel-2 data) and the CNN model (overall accuracy: 89.17% from MODIS data and 86.02% from Sentinel-2 data). Our results also showed that the accuracy from cropland and savanna classes were higher than grassland and forest classes in all three models. These two classes generated from the CNN-LSTM model performed better than the other two deep learning models. The results from these two datasets indicated that the methods were reliable for both coarse and medium spatial resolution satellite images and time series remote sensing images worked better than single image for classification problems when considering LULC change over time. The results also provided an alternative way to prepare input data from satellite images for deep learning models. Furthermore, the classification results of the whole agricultural expansion area were reasonable and it can be used as an additional dataset for further environmental analysis at a regional scale.

Change detection (CD) is the operation of quantitatively analyzing the surface changes of a phenomenon or objects over two different times. Lately, CD based on deep learning has developed to become more and more powerful, and convolutional neural networks (CNNs) have dominated the field of remote sensing (RS) CD. In particular, in many fields of computer vision, neural networks based on U-Net network and skip connections have been generally used. However, despite the excellent performance achieved by CNN, it does not learn global and long-range semantic information interaction well due to the locality of convolutional operations. The recently proposed Swin-UNet in the field of medical image segmentation achieved excellent results, which is a U-Net-like pure transformer. In the face of the challenge of segmentation accuracy, the Swin transformer has demonstrated strong capabilities. The Swin transformer block (STB) consists of residual connected STBs used in SwinIR to enhanced training stability. We began to try to incorporate them into our network for RS CD. Finally, we propose a transformer-based multi-scale feature fusion model (TMFF), including decoder, encoder, and skip connection structure, for RS image CD. We modify the original U-Net architecture so that it can better aggregate semantic features at all levels. Our proposed TMFF achieves impressive results through experiments on three datasets;

Although the data on a particular complicated sea conditions are available, it is difficult to collect the data on various complicated sea conditions simultaneously. When the sea conditions change, the test results of the detection network based on the data of simple sea conditions may be unacceptable. It is proposed that the VGG16, pix2pixHD, and cycleGAN methods should be applied to establish related datasets of different complicated sea conditions. After the unified image quality assessment indicators are determined, it is found that the images generated by the VGG16 network are most in line with the subsequent target detection standards. The real images of complicated sea conditions have the smallest root-mean-square error, the largest peak signal-to-noise ratio, and the best regression coefficient (R²). Meanwhile, the improved Faster R-CNN is introduced for target detection of small sample datasets. First, the ResNet50_FPN module, as the backbone feature extractor, is used to improve the detection performance of small-sized objects. Moreover, because there are many small target objects to be detected, the ROI Align module is preferred, for it is more accurate. Finally, the Softer-NMS algorithm is selected to significantly improve the positioning accuracy through confidence estimation. Compared with some previous Faster R-CNN methods, the accuracy and missed detection rate of the proposed method outperform other networks, with a mean average precision of 85.11%.

Optical image dense matching is a crucial step in the process of generating digital surface models (DSMs). Many existing dense matching methods have adopted pixelwise matching strategy and have achieved precise matching results; however, the methods are time consuming and have limited efficiency in high surveying and mapping production. We introduce a bridge probability relaxation matching method for automatic DSM generation. The method adopts a coarse-to-fine hierarchical strategy and achieves high matching accuracy and fast processing speed simultaneously. This method builds a self-adaptive disparity surface model in a local area and constrains the disparity surface using the spatial relationship between feature points and adjacent pixels. Finally, the disparity is optimized by calculating the increment of the relaxation iteration probability. Experiments are based on different areas with different textures and terrain types. Compared with the DSMs derived from semi-global matching, our proposed approach achieves high levels of accuracy and efficiency in automatic DSM generation.

Extraction of rivers from remote sensing images is crucial for understanding the utilization of water resources. Because rivers have similar and consistent widths in a consistent range of remote sensing images, the stroke width transform (SWT) algorithm has a good effect on river extraction. However, the algorithm is susceptible to the influence of features around the river. A river extraction method combining relative total variation (RTV) and SWT was proposed to obtain river area information with high accuracy. First, the near-infrared channel of the image was selected as the input and the RTV model was used for texture smoothing and edge enhancement. Thereafter, the image edge map calculateds by the Canny operator was used as the input of the SWT to obtain the stroke width map. Subsequently, the connectivity components were labeled, and a specific filter was constructed according to the geometric characteristics of the river to filter out nonriver noise. The retained connected components were filled with holes, and a highly accurate river basin map was generated. In this study, six images of rivers flowing through four different landforms, including woodland, urban, mountainous, and cultivated land, were captured by the Gaofen-1 satellite for validation and comparison analysis. The experimental results demonstrated that the method’s completeness, accuracy, and extraction quality in this study could be maintained above 95%. Compared with the mainstream methods of the Otsu method, normalized difference water index, U-Net, and direct use of the SWT algorithm, the method in this study is significantly better in terms of extraction effect and algorithm stability. This can effectively improve the accuracy of river segmentation.

You only look once version 5 (YOLOv5) algorithm uses one convolutional layer for both classification and localization tasks. They share the same set of weights, leading to interference between the two tasks. Thus, the application of this method in remote sensing scenario is very limited. To solve the problem, we decouple the detection head of the YOLOv5 algorithm so that these two tasks can be separated. Furthermore, we achieve high quality detection in remote sensing scene with task refinement. Specifically, we propose a weight allocation method based on Huffman coding theory to ease imbalance of the number of categories and improve the classification accuracy. If the number of a category is large, a small weight is set. Otherwise, a large weight is given. To improve the localization accuracy, we use the efficient intersection over union (IOU) loss to alleviate the problem of penalty failure caused by central IOU loss. In this way, the predicted boxes are closer to the ground truth. A large number of experiments were carried out on DOTA and UCAS-AOD datasets to verify the effectiveness of our proposed method. Compared with baseline, the mean average precision totally increased by 2.29% in DOTAv1.5 dataset. Simultaneously, our method also showed superior performance when comparing with start of the art algorithms.

To effectively solve the accurate identification of gray ice, melt ponds water, floe, brash ice, and thin ice in the melting state of the Arctic Sea ice during summer, we propose adding a batch normalization layer and adaptive moment estimation optimizer of a U-NET (BAU-NET) method for Arctic Sea ice semantic segmentation in summer from remote sensing satellite optical images. The U-NET network structure is optimized to 18 convolution layers, and a batch normalization layer and a nonlinear activation function rectified linear unit are added behind each convolution layer. Then the hyper-parameters of the network structure are adjusted. The cross-entropy loss function based on SOFTMAX and L2 regularization are used in training the model, and the adaptive moment estimation optimizer is used for iterative training until the error convergence. The experimental results show that the accuracy, precision, recall, and F1 score of sea ice extraction results reaches more than 97.26%. Compared with the DeepLabv3 and U-Net methods, the sea ice prediction time efficiency is improved by 90.99s and 1.57s, respectively, and the accuracy is improved by 6.59% and 8.05%, respectively, which indicate that the sea ice prediction time efficiency and accuracy of the BAU-NET method are significantly improved.

The application of unmanned aerial system (UAV) remote sensing technology as a unique means of spatial data collection in various fields is based on high-resolution images captured by UAVs. However, due to the influence of atmospheric water vapor, haze, and other factors, the quality of UAV aerial images is limited. Considering that the model-based image fogging method developed according to the dark channel theory is widely used, the image edge after fogging incurs a “halo” effect owing to the excessive coarseness of the transmission map. We take advantage of the nonlocal structure tensor of the image to better protect the important details of the image edge while removing noise. The images we used are obtained from the public multi real world fog image dataset. Experiment results demonstrate that the proposed method can effectively eliminate the halo effect that occurs at the edges of an image after demisting.

The Baltic Sea is an optically very complex study object for watercolor remote sensing because of the high quantity of colored dissolved organic matter, two optically distinct phytoplankton seasons, high variability in concentrations of optically active substances, and low sun angles. Despite this, there are numerous remote sensing and modeled chlorophyll-a (Chl-a) products publicly available for the Baltic Sea. Sixteen openly accessible Chl-a products were tested with 267 in situ Chl-a measurements that were carried out in Estonian marine waters during 2016 to 2021. All modeled products and about half of the remote sensing products failed to produce reliable results. The best-performing remote sensing Chl-a product was Case2/Regional CoastColour produced from Sentinel-3 ocean and land color imager (OLCI) reflectance with R² = 0.55, root mean squared error ( RMSE ) = 4.5 mg m ^{− 3}, mean absolute percentage error (MAPE) = 74%. In addition, eight different band ratio algorithms were applied on Sentinel-3 OLCI and Sentinel-2 multispectral instrument data. The best remote sensing band ratio algorithm was derived from top-of-atmosphere reflectance of Sentinel-3 data using 665, 709, and 754 nm bands (R² = 0.67, RMSE = 3.9 mg m ^{− 3}, and MAPE = 63%). Our results show good suitability of Sentinel-3 for Chl-a retrieval. However, the high uncertainties suggest for the further product development and validation needs.

A high-resolution 96-GHz (3 mm) passive millimeter-wave imaging system is constructed. The system has a 96-GHz radiometric receiver, which is placed on the focal point of a 50-cm diameter parabolic reflector designed and created. A specially written program for the system controls the antenna via control hardware in desired steps by a 2-axis positioner motor. A balance weight has been calculated and mounted to the rear of this parabolic reflector antenna to prevent possible damage to the rotor of the positioner motor due to the antenna’s weight. The maximum field of view of the system is 120 deg in azimuth and 90 deg in elevation. The angular scan step can be varied from 0.1 deg to 1 deg. Although the system can shoot 1190 × 890 samples (pixels) images, which is close to the HD quality (1280 × 720 pixels) in the full field of view, the actual resolution of the image is limited by the antenna’s beamwidth (BW). Some measurements were performed to investigate the effect of scanning steps on PMMW image quality for different angular scan steps. It is observed that the PMMW image scan with the smallest step gives a detailed image that is the closest to the scene’s optical image as expected. Even though the angular scan step angle is less than half of the antenna’s BW at Nyquist sampling, it is observed that the image quality still improves due to averaging effect of the received signals giving a better estimated S/N. The resolution and scan step values of our system are compared with that of some systems found in the literature and it is shown that ours has the highest value among them. This system can be used to obtain high-resolution images in remote sensing and other applications.

We subjected a tool for 3D georeferencing accuracy assessment of tri-stereo primary images of Pléiades 1A/1B/Neo, SPOT 6/7, and Göktürk-1, developed in the MATLAB environment considering their private sensor-dependent orientation model. Although the orientation model has been developed by the image vendors, the major contribution of Geo3o1 was the two-step adjustment process. The basic stages of estimating the 3D georeferencing accuracy were (i) estimating the direct georeferencing accuracy by raw orientation parameters, (ii) pre-adjustment of look angles using ground control points (GCPs), and (iii) bundle adjustment. In addition, the efficiency and validation of exterior orientation parameters and also the correlation among interior and exterior orientation parameters could be analyzed. Geo3o1 has the capability of handling the images in stereo or tri-stereo modes. The case study has been carried out by the Pléiades 1A panchromatic tri-stereo primary images with 50-cm geometric resolution. About 90 points were used as GCPs/ICPs in two distribution scenarios. The accuracy on ground was estimated centimeter level at GCPs whereas the accuracy was naturally coarser at ICPs at ∼ ± 1 GSD. The efficiency and validation of exterior orientation parameters and the correlations were also shown by two samples. The research was carried out over an undulating, mountainous area in Zonguldak, Türkiye.

Wetlands have very important and irreplaceable ecological service functions in protecting biodiversity and maintaining ecological balance. Due to the unique hydrological characteristics of wetlands, large-scale wetland mapping methods are usually time-consuming, labor intensive, and expensive. With the rise of big data and cloud computing platforms, there are opportunities for long time series and large-scale spatial data processing that improve large-scale wetland mapping. In this study, the Sentinel-1, Sentinel-2, and digital elevation model derived from the Google Earth Engine were used to extract the wetland extent in the Zhalong Nature Reserve based on the Jeffries–Matusita (JM) distance for feature optimization and random forest. Research shows the following: (1) The JM distance for each feature showed that the optical features had the highest separation in all five features and (2) the classification after feature optimization performed best. User accuracy (UA) and producer accuracy (PA) for wetland reached 89.64% and 83.00%, respectively, and overall accuracy reached 90.76%. After feature optimization, PA and UA of wetland were each increased 4.26% and 1.07%, respectively, and the number of features was reduced from 46 to 28. The method used in this study is highly accurate in wetland mapping and can effectively avoid the problem of data redundancy.

The performance of hyperspectral image (HSI) classification can be enhanced by utilizing a metaheuristic technique with high exploration and exploitation capabilities for band selection. We design a framework for the improved classification of HSIs by utilizing a binary modified equilibrium optimizer (BMEO) for band selection. First, the proposed framework preprocesses the HSI data using the Gaussian filter to remove the undesired variations. Band selection is performed on the preprocessed HSI data using a BMEO, which exhibits balanced and high exploration and exploitation capabilities. The proposed V-shape transfer function is used for enhancing the exploration capabilities of the BMEO. An analysis for band selection is performed by comparing the reflectance spectra, correlation of each band with ground truth, and correlation between the bands before and after the band selection. The classification of HSI data by utilizing selected bands was done using the classifier selected by classifier analysis. The proposed framework outperforms 11 state-of-the-art techniques, including filter, wrapper, and deep learning-based techniques. The 95.85%, 98.13%, and 98.31% overall accuracies on the Indian pine, Pavia University, and Salinas datasets, respectively, show the significance of the proposed framework.

The seasonal variations of forest canopy spectral characteristics are critical to improving the utilization of remote sensing methodology to quantify forest physiology, especially forest carbon sink. However, the seasonal variations of forest canopy spectra are poorly understood. Combined field survey and EO-1 Hyperion imageries, we extracted the spectral curves of seven forest types of Changbai Mountain in China in seven periods. We also calculated various remote sensing indexes and analyzed their seasonal change of spectral characteristics among different forest types. Optimal indexes were selected to indicate the seasonal variation of forest carbon fluxes. Our results showed that there were differences in spectral curves among forest types. The reflectance of coniferous forests was lower than that of broad-leaved forests in growing season. Changbai Scotch pine forest owned the lowest spectral reflectance, whereas the reflectance of Mongolian oak forest was the highest, especially in the near-infrared region. The red edge slope (RES) of broad-leaved forest was higher than coniferous forest in spring and summer. The RES of broad-leaved and coniferous forests was similar in autumn. The red edge position of various forest types showed slight shift in different seasons. Four typical forest types showed different spectral characteristics with seasonal changes. The seasonal variation of coniferous forest spectral curves was not obvious. The seasonal variation of broad-leaved forest spectra was the largest. Most of the spectral indexes can indicate the seasonal variation characteristics of each forest type. Enhanced vegetation index (EVI) is better than normalized difference vegetation index (NDVI) to indicate the forest phenology. Seasonal curves of spectral indexes were different in all forest types. Spectral indexes of coniferous forests were most stable throughout the year. The curves of each index in broad-leaved forests showed significant difference in autumn, which may be influenced by the understory vegetation after their defoliation. For broad-leaved Korean (BK) pine forest, the scaled value of photochemical reflectance index (SPRI)*EVI owned the highest correlation with gross primary productivity (R = 0.99 and P < 0.01) and net ecosystem exchange (R = − 0.77 and P < 0.05), respectively. SPRI*NDVI showed the highest correlation with ecosystem respiration (R = 0.96 and P < 0.01). The seasonal variation of carbon fluxes of different forest types retrieved from the optimal remote sensing index were consistent, but their peaks occurred at different times.

Gaofen-3, the first high-resolution C-band multipolarization synthetic aperture radar satellite from China, has been in orbit for 5 years. It is significant for evaluating the positioning accuracy of Gaofen-3 products and determining the possibility of improvement. In this study, 338 Gaofen-3 images for six imaging modes from 16 test fields in China were selected, and 363 ground control points were used as independent checkpoints to verify positioning accuracy. The test results show that the average positioning accuracy of the six modes in the slant range direction was between −27.52 and −17.32 m. The average positioning accuracy of the six modes in the azimuth direction was between −17.21 and 2.4 m. The comparative experiments under the same imaging conditions demonstrate that the on-orbit state of Gaofen-3 is relatively stable. The internal electronic delay of the instrument and the azimuth “bistatic error” of the Gaofen-3 image rational function location model have not been eliminated. There is room for further improvement in the positioning accuracy of the Gaofen-3 standard product.

Automatic target recognition (ATR) is a challenging task for several computer vision applications. It requires efficient, accurate, and robust methods for target detection and target identification. Deep learning has shown great success in many computer vision applications involving color RGB images. However, the performance of these networks in ATR with infrared sensor data needs further investigation. In this paper, we propose a multistage automatic target detection and recognition (ATDR) system that performs both target detection and target classification on infrared (IR) imagery using deep learning. Our system processes large IR image frames where targets take <1 % of the total number of pixels. First, we train a state-of-the-art object detector you only look once (YOLO) to localize all potential targets in the input image frame. Then, we train a convolutional neural network (CNN) to identify these detections as targets or false alarms. In this second phase, we adapt and analyze the performance of three CNN architectures: a compact and fully connected CNN, VGG16 with batch normalization, and a wide residual neural network (WRN). We also explore the use of a loss function that optimizes directly the area under the receiver operating characteristic (ROC) curve (AUC), and adapt it to our ATR application. To enhance the robustness of the proposed ATR to perturbation and variations introduced during the detection stage, we train our CNN classifiers on automatically detected targets using YOLO, in addition to ground truth bounding boxes and apply selected data augmentation techniques. To simulate real testing environments, where the spatial location of the targets within the image frame is unknown, only YOLO-detected boxes are used during validation. We evaluate our ATDR on a real benchmark dataset that includes different vehicles captured at different resolutions. Our experiments have shown that YOLO can detect most of the targets at the expense of generating a high number of false alarms. We show that the VGG-16 network with batch normalization, which is the best performing model, can correctly identify the classes of the targets, as well as classify the majority of YOLO’s false detections into an additional nontarget class. We also show that the proposed training modification to optimize an AUC-based loss function for ATR proved to be advantageous mainly in identifying difficult targets.

Convolutional neural networks (CNNs) are very important deep neural networks for analyzing visual imagery. However, most CNN-based methods have the problem of over-smoothing at boundaries, which is unfavorable for hyperspectral image classification. To address this problem, a spectral-spatial multiscale residual network (SSMRN) by fusing two separate deep spectral features and deep spatial features is proposed to significantly reduce over-smoothing and effectively learn the features of objects. In the implementation of the SSMRN, a multiscale residual convolutional neural network is proposed as a spatial feature extractor and a band grouping-based bi-directional gated recurrent unit is utilized as a spectral feature extractor. Considering that the importance of spectral and spatial features may vary depending on the spatial resolution of images, we combine both features with two weighting factors with different initial values that can be adaptively adjusted during the network training. To evaluate the effectiveness of the SSMRN, extensive experiments are conducted on public benchmark data sets. The proposed method can retain the detailed boundary of different objects and yield competitive results compared with several state-of-the-art methods.

Rapid and comprehensive lockdowns to contain the coronavirus 2019 (COVID-19) pandemic reduced anthropogenic emissions and, thereby, decreased the aerosol optical depth (AOD) in Xiangyang, Hubei Province. However, their complicated interactions make quantifying the contribution of decreased aerosols to crop growth challenging. Here, we explored the indirect effects of decreased aerosol concentrations on the gross primary productivity (GPP) and water use efficiency (WUE) of winter wheat by quantifying the contributions of key environmental factors. Our results showed high temporal and spatial associations between aerosols (represented by AOD), GPP, and WUE before, during, and after the COVID-19 pandemic. AOD decreased by 23.8 % ± 10.1 % , whereas GPP and WUE increased by 16.5 % ± 5.8 % and 17.0 % ± 15.3 % , respectively. The GeoDetector model revealed that photosynthetically active radiation (PAR) had a major impact on GPP and WUE, followed by precipitation, surface soil moisture, subsurface soil moisture, and surface temperature. Moreover, causality analysis showed a causal relationship between AOD and the dominant factors (PAR and precipitation) during the lockdown, thereby indicating a positive effect of decreased aerosols on GPP and WUE changes of winter wheat. Our findings assist in understanding the mechanisms causing GPP and WUE changes, given the environmental factors that changed significantly during the pandemic.