Guest editors Manjit Kaur, Raman Singh, and Hassène Gritli introduce the Special Section on Meeting the Challenges of Ecosystem Management using Remote Sensing.

Temporal and spatial environmental factors have a substantial influence on crop yields, and an accurate prediction can benefit timely decision-making in global food production. Thus for better agricultural management, the precise estimation of the croplands is helpful. Mapping the cropland dynamics with regular requirement of crops is an important prerequisite for monitoring crops, yield estimation, and crop inventories. Remote sensing and geographic information systems play a significant role in tracing and understanding environmental impacts of agriculture. The use of machine learning aids in developing a model that can give precise predictions based on the historical data. The main objective of our study is to use these machine learning algorithms to make accurate predictions about the best crop types using the spectral, temporal, and polarimetric features. A big dataset incorporating optical and polarimetric aperture radar experimental values is used to train the machine learning classifiers for predicting the right crop type in the study area. It has been observed that these features aid in accurate mapping of the cropland. Our study involves the performance comparison of the various machine learning algorithms, and it has been observed that a single-layer neural network offers prediction accuracy of ∼99.6 % for this big PoISAR dataset.

Unmanned aerial vehicles (UAVs) that include mounted camera sensors enable a wide range of remote sensing application deployments. Due to UAVs’ capacity to explore distant locations such as woods, situational awareness for applications, such as search and rescue in wildfires, estimation of endangered flora and fauna, and emergency responses have undergone a paradigm change. A multi-layer perceptron (MLP)-Mixer architecture is suggested for classifying burned piles in dense forests. Convolutional neural networks (CNNs) and the more recent attention-based transformer models have produced cutting-edge outcomes in picture prediction. Convolutions and attention are used in the MLP Mixer architecture in an effort to overcome their drawbacks and improve performance. By including a new module of DePatch in the suggested MLP Mixer model, which separates the input images in a deformable pattern to identify forest fires at an early stage, the shallow learning of CNN layers and fixed-size patch embedding in transformers were eliminated. This suggests that the DePatch-based MLP classification model outperforms transformer approaches in terms of performance, achieving a substantial accuracy of 77.23. Our proposed classification system was evaluated on the pile photos dataset taken during a burning pile of debris in an Arizona pine forest.

The scope of our work is to compare Sentinel-2 and unmanned aerial vehicles (UAV) imagery from Western Tian Shan for monitoring applications in walnut fruit forests. Because acquiring field data on walnut fruit forests is difficult due to access limitations in the Tian Shan, remote sensing offers a unique opportunity to increase spatial and temporal coverage of ecological parameters. Sentinel-2 satellites launched in 2015 have been providing images at spatial resolutions of 10 to 20 m; however, difficulties remain as information retrieved from Sentinel pixels result from a mixture of objects that affect reflectance signals. We used submeter multispectral UAV images to assess the sensitivity of Sentinel-2 normalized difference vegetation index (NDVI) to subpixel vegetation. Results showed that Sentinel-2 data overestimates NDVI in regions with open terrain and grass, and underestimates NDVI in areas with trees. By implementing a bias correction method, the accuracy of the Sentinel-2 derived NDVI increased; R² values increased from 0.59 to 0.88 (p value < 0.001). We also showed that drought index derived from Sentinel-2 vegetation condition index (VCI) is well correlated with the ground-based standard precipitation index (SPI). Using this bias correction method, on average the correlation increased by 3% between VCI and SPI.

In the field of remote sensing, the classification of land cover is a pivotal and challenging issue. Standard models fail to capture global and semantic information in remote sensing images despite the fact that a convolutional neural network provides robust support for semantic segmentation. In addition, owing to disparities in semantic levels and spatial resolution, the simple fusion of low-level and high-level features may diminish the efficiency. To address these deficiencies, an attention-guided multi-level feature fusion network (AMFFNet) is proposed in this study. The proposed AMFFNet approach is designed as an encoder–decoder network with the inclusion of a multi-level feature fusion module (MFF) and a dual attention map module (DAM). A DAM models the semantic association of features from a spatial and channel perspective, and an MFF bridges the semantic and resolution gaps between high-level and low-level features. Furthermore, we propose a residual-based boundary refinement upsample module to further optimize the object boundaries. The experimental results indicate that the proposed strategy can considerably enhance the accuracy of land cover classification, achieving a mean intersection over union of 90.39% on the LandCover.ai dataset and 63.14% on the Gaofen Image Dataset with 15 categories (GID-15).

The application of remote sensing (RS) in the treatment of problems caused by natural phenomena, such as coastal erosion, has proven to be very effective. An attempt was made to reveal areas that are in a state of erosion along the coastline of the Prefecture of Achaia, elongated ∼185 km. The study area comprised the Gulf of Patras in the west and a part of the Corinth Gulf to the east, which are connected through the narrow and shallow Rion Straits. The system of the two Gulfs consists of a Plio-Quaternary asymmetric grabens trending northeast – southwest (NE-SW) and west-northwest (WNW) – east-southeast (ESE), respectively. The shoreline evolution is estimated for the 1945 to 2016 period with the usage of high-resolution data, using RS and geographic information system techniques. An extension of ArcGIS, the digital shoreline analysis system tool, version 5 beta, has been used for the analysis of shoreline changes, whereas the ERDAS Imagine 2014 was used for the image georeferencing and orthomosaicing process. Various statistical indices were computed including the end point rate and the linear regression rate. Although these tools did not take into concern any of the local geological parameters, the results are similar to the respective derived from conventional survey techniques. Eight sandy areas were revealed that are under erosion regime: the Araxos, Karnari, Kalamaki, Drepano, Loggos, Akoli, Diacopto, and Akrata, while the future shoreline position was estimated for each coast using the forecasting beta tool of the DSASv.5.0, after 10 and 20 years since 2016. We found that it is very likely all sites will show the same trend of change in the next 10 and 20 years, similar to those computed using the available data from 1945 to 2016. The tectonic behavior of the Gulf of Corinth seems that it mainly affects the area’s erosion/accretion regime, while changes in the areas adjacent to the Gulf of Patras are related to natural processes (waves, currents, coastal floods, extreme rainfalls, etc.) and anthropogenic interventions. Moreover, the Araxos and Karnari coasts are at 65.71% and 64.41% under moderate erosion regime with rates of change −0.62 and −1.10 m / year, respectively. Kalamaki, Akoli, and Loggos percentages of 82.35%, 86.67%, and 86.23% are under low erosion, with rates of change −0.28, −0.21, and −0.34 m / year, respectively. Drepano and Diacopto are under low erosion, with rates of change −0.26 and −0.16 m / year, corresponding to 54.55% and 38.67% of the coast, respectively. Finally, Akrata is a coast where erosion prevails at 60.42% with annual rate of change −0.16 m / year, whereas the 43.38% of the coast correspond to stable area.

Advanced remote sensing technologies, such as light detection and ranging (LiDAR), offer significant potential to mapping the alpine treeline ecotone (ATE) based on its actual definition (tree height ≥ 3 m) and contribute to the generation of baseline data for future change detection investigations. We propose an approach for combining LiDAR-derived absolute tree height data with elevation data to delineate the ATE in Uttarakhand, India. The approach was implemented using observations from the recently launched Global Ecosystem Dynamics Investigation system and validated with field measurements. The LiDAR-derived treeline was compared with the traditional normalized difference vegetation index (NDVI) treeline. The treeline derived from LiDAR was found to have root mean square error of ∼60 m with respect to the ground verified treeline location. The NDVI treeline was overestimated in comparison to the LiDAR treeline by an average surface distance of 290, 232, 257, and 237 m in the south, north, west, and east aspects, respectively. It is observed that the overestimation was higher at the lowest and highest elevation zones. We prove that LiDAR-based treeline mapping is an efficient method to delineate alpine treelines at a landscape scale.

Crop classification in small and fragmented agricultural lands is a challenging task. The presence of cloud cover poses additional challenges when classical techniques using multispectral images are applied. Multispectral data augmented with SAR images is capable of solving this issue. The use of robust deep neural network classifiers can further improve the classification performance. We propose a multichannel attention-based deep learning model to fuse time-series SAR and a single scene multispectral data at the feature level and classify tropical crops grown in smallholdings at pixel level with high accuracy. The attention-based transformer neural network for the fusion of Sentinel-1 and Sentinel-2 data is evaluated for crop classification in the Cauvery delta zone. The model achieves an overall classification accuracy of 92.6% and a kappa score of 0.906. The source code is available in a GitHub repository at: https://github.com/ajigeo/temporal-SAR-transformer.

On-orbit reconstruction is a hotspot technology in the field of space infrared remote sensing cameras technology. Since the space infrared remote sensing cameras perform detection work under the space radiation environment, the core part of its image processing unit, which is a static random-access memory type field programmable logic gate array (FPGA), is susceptible to the impact of high-energy particles, especially during single-event upset events. The occurrence of point-like granular noise in remote sensing images will affect the normal imaging detection performance of the system, and even lead to abnormal timing of infrared detectors in severe cases, thereby affecting the normal operation of infrared remote sensing cameras. Aiming at the reliability requirements of the normal operation of space infrared remote sensing cameras in the space radiation environment, we propose a modular architecture of heterogeneous dual FPGA for XilinxVirtex-5 series FPGA, which carries out periodic refresh, reconstruction, and uploading of Virtex-5 series FPGA through antifuse FPGA. The antifuse FPGA periodically refreshes and reconstructs the Virtex-5 series FPGA from a one-time programmable read-only memory and injects the program into the program storage NOR flash device, and periodically refreshes and reconstructs the Virtex-5 series FPGA from NOR flash. This method can refresh and reconstruct the space infrared remote sensing camera in orbit. The in-orbit flight data verification shows that this method can effectively solve the impact of single-particle turnover event and improve the reliability of the infrared remote sensing camera in orbit operation.

Cavity exploration is an important task for optical remote sensing that finds application in many fields, such as the assessment of disaster areas or the evaluation of subsurface cavities in terrestrial and extraterrestrial scenarios. In our work, we use single photon-counting avalanche diode cameras to record temporal signatures in the picosecond time domain and to exploit the impulse response of cavities. When illuminated with a short optical pulse, the photons are reflected several times inside the cavity before exiting it and being detected. In our datasets, we observe several intensity modulations correlated to specific optical paths with different numbers of reflections inside the cavity. We developed an analysis model based on simulation of the radiant transfer that provides a quick rough estimate of the cavity dimensions without costly measurements or computational reconstruction. Specifically, we developed a spherical model and a hemispherical model that can be used to estimate the cavity dimensions (diameter or height). Finally, we present experimental studies of eight different cavities with different dimensions, shapes, and surface materials. Our evaluations show that we can apply our analytical models to signatures of different cavities, which gives us an initial estimate of the cavity sizes before a detailed mapping is carried out.

The spatial resolution of a hyperspectral image (HSI) can be enhanced by utilizing a co-registered high-resolution multispectral image (MSI) via an image fusion algorithm. This is an active area of research with applications ranging from precision agriculture, to security issues, to mineral identification and mapping, etc. The nearest-neighbor diffusion (NNDiffuse) algorithm has seen tremendous success in being utilized as a pansharpening algorithm to fuse an MSI with a broadband high-resolution panchromatic image and is implemented in commercial software applications, such as ENVI (L3Harris). We extended NNDiffuse to the problem of hyprespectral–multispectral image fusion (HS + MS). Hyperspectral pansharpening is a special case of this: a single high-resolution broadband (panchromatic) image is fused with a low-resolution HSI. Deep-learning (DL)-based methods can achieve excellent results in the area of HS + MS data fusion, but DL algorithms are data hungry: they need extensive training data, require special computing architectures, are slower to implement, etc. NNDiffuse introduces much less spectral distortions compared to state-of-the-art methods, does not need any training data or expensive computing resources, and is significantly faster compared to DL-based approaches. We: (1) demonstrate the utility of NNDiffuse in low-resolution HSI–MSI sharpening; (2) test three deep learning-based image fusion algorithms and compare performance with NNDiffuse and other non-DL-based fusion algorithms using global/image-wide quality metrics; and (3) assess fusion performance using an application based approach: adaptive coherence estimator target detection. The proposed method is tested against several datasets of varying scene content and complexity, and we demonstrate that the NNDiffuse fusion method outperforms the other baseline methods when the application of target detection is considered.

Synthetic aperture radar (SAR) has become an important part of ship detection due to its all-day all-weather characteristics. Ship monitoring is important for coastal traffic control and territorial safety. Ship detection requires high accuracy, but the speed of ship detection is also an important parameter in making rapid decisions, such as those in maritime rescue and military strategy. However, fast networks tend to be less accurate. To address this problem, we propose an efficient localization method based on a spatial attention mechanism. This mechanism uses grid-guided localization instead of a frame regression mechanism, which makes the network faster but also lowers the accuracy to a level insufficient for direct application to SAR image target detection. A tested and selected spatial attention mechanism improves the detection accuracy while guaranteeing its speed. It is shown that four factors affect the design of the spatial attention mechanism, namely query and key content, query content and relative position, key content, and relative position. These factors are ordered, and the group with the best precision is added to the proposed network. The proposed network can achieve a good detection effect for sparsely distributed targets in SAR images, and the proposed algorithm can achieve AP_{50 ∶ 75}, AP₅₀, AP₇₅, and AP_S values of 71.3%, 96.9%, 86.6%, and 72.1%, respectively, on the SAR ship detection dataset.

Knowledge of demographic data is valuable information for planning initiatives. Typically, census, survey, and population projection exercises provide this information. In some developing countries, these operations pose a variety of economic and logistical challenges, thereby depriving authorities of accurate and timely information on their populations. To provide approaches for solving this situation, our study evaluates a population estimation method that is based on detection of residential geo-objects (houses) on very-high-resolution (VHR) satellite images using convolutional neural networks (CNN). The approach would be applicable to countries where a complete census is difficult to perform due to resource constraints or political instability. A 2008 VHR satellite image of Sudan is annotated according to seven classes of buildings to create a dataset that was used to train an object detection model, faster region-based CNN, by transfer learning. The model obtained mean average precision of 79% and 99% during training and validation, respectively. This unusual difference is due to the dominance of well detected classes in the validation dataset. The model was fine-tuned to detect the same building classes on images in 2021. A link between residential geo-objects and population size was established using 2008 population data and available field data. Subsequent characterization of the current population should assist in preparation of the 2023 census. Limitations of this approach were raised, but it could be used to improve the framework for population data collection in developing countries.

The rapid mapping of flood inundation is essential for timely assessment of damage and post-disaster recovery. Remote sensing technology provides clear spatial information for mapping flood inundation, but its real-time (RT) images are often not available due to the severe weather conditions during the disaster events; the combination of RT data can better compensate for the deficiency. In this study, a near RT (NRT) flood inundation probability mapping method based on post-event NRT remote sensing data combined with RT volunteer geographic information (VGI) was proposed for mapping the 2019 flood in Linhai City, Zhejiang Province. First, a probabilistic index distribution (PID) layer was constructed from high-resolution digital elevation model data using an inverse distance-weighted height filter based on each VGI point. Then, a quality evaluation method for non-reference data was introduced to evaluate the validity of the VGI data. The final flood probability map was generated by PID weighting. The results show that, by fusing NRT images and RT data, the proposed model for mapping flood inundation probability is more robust and enhances the spatial characteristics of the flood inundation probability index, allowing it to enable emergency responders to quickly identify areas requiring urgent attention.

The unstructured-grid finite-volume community ocean model (FVCOM) offers the potential to build a high-resolution operational prediction system because it can effectively adapt to complex boundaries and flexibly perform local mesh encryption. A matching assimilation system is an indispensable part of an operational forecast system, while the existing assimilation method of the FVCOM [e.g., the ensemble Kalman filter (EnKF)] has difficulty accommodating strong nonlinear model/observation operators and non-Gaussian errors. The localized weighted EnKF (LWEnKF) is a particle filter (PF) that combines the advantages of the EnKF and the PF, which can effectively address nonlinearity and non-Gaussianity and can provide a more accurate initial field for models. An observation preprocessing system and a new data assimilation system based on the LWEnKF for the FVCOM are developed. For the preprocessing system, a K-dimensional tree-based observation thinning method is proposed that can accommodate representative observation efficiently for unstructured grids. Through a 1-month global ocean data assimilation experiment, the effect of our assimilation system is compared with the EnKF and the Mercator analysis forecast product, which comes from a reduced-order Kalman filter system. The statistical verification includes five metrics for various sea surface fields, temperature, and salinity profiles. The results show that the system performs better than the EnKF, although there is still a gap with Mercator in some aspects.

Developing an efficient method to accurately estimate aboveground biomass in tropical forests is critical to monitoring the carbon stock and implementing policies to reduce emissions caused by deforestation. Thus, the objective of the present study was to estimate aboveground biomass in areas of the Amazonian Forest with selective logging, using the random forest (RF), support vector machine (SVM), and artificial neural network (ANN) machine learning algorithms, using light detection and ranging (LiDAR) data and these combined with OLI/Landsat 8 variables, as well as mapping the biomass for the years 2014 and 2017, allowing one to analyze its dynamics between the years of analysis. The RF and SVM algorithms obtained the lowest error values in all datasets. The association of the variables increased the RF performance. Analyzing the dynamics of biomass, it was observed that the oldest exploration units (2006, 2007, and 2008) have lower biomass stocks. The highest biomass losses in 2017 came from units operated between 2012 and 2013 (the most recent record). Thus, with the method used in this study, it was possible to infer that the machine learning algorithms were efficient in estimating the biomass, emphasizing the RF and the SVM.

Coastal salt marshes are valuable wetland ecosystems, and obtaining accurate information on their spatial and temporal distribution is crucial. However, due to the strong spectral similarity among various types of salt marshes, distinguishing them is difficult, especially in areas with mixed salt marsh growth. In this study, we developed a pixel-based unique phenological feature composite method (PUpf-CM). First, we identified a unique phenological period for each type of salt marsh by analyzing the normalized difference vegetation index time-series of Spartina alterniflora (S. alterniflora) and native salt marshes [Suaeda salsa (S. salsa), Phragmites australis (P. australis), and Imperata cylindrica (I. cylindrica)] in Yancheng National Nature Reserve, Jiangsu Province, China. Then, we generated a composite image for each unique phenological period, maximizing the spectral separability between the corresponding salt marsh and background species. Second, we used the composite images obtained at the unique phenological period of S. alterniflora as input to the random forest classifier, achieving high precision identification of S. alterniflora. Compared with the previously developed pixel-based PUpf-CM, the overall accuracy (OA) of PUpf-CM was 98.26% to 99.42%, and the Kappa was 0.940 to 0.975, indicating an improvement of 0.87% to 4.94% and 0.031 to 0.179, respectively. Finally, we tested the possibility of applying PUpf-CM to salt marsh classification by stacking four composite images obtained at different unique phenological periods. We found that the maximum spectral separability was achieved for the combination of four unique phenological periods, and the OA and Kappa of the PUpf-CM-based salt marsh classification reached 97.54% and 0.969, respectively.

Visible and near-infrared (vis-NIR) reflectance spectroscopy has been confirmed to be a convenient, rapid, and effective method for monitoring soil organic matter (SOM) content. However, its estimation accuracy is always affected by a number of factors, including soil moisture. The synchronized dewatering measured-value fuzzification (SDMF) method was investigated, with the aim being to reduce the effect of soil moisture during vis-NIR modeling. To magnify the effects of soil moisture, mixed samples (air-dried and moist samples) were used for SOM content estimation. Vis-NIR spectra were collected using an analytica spectra devices (ASD) FieldSpec 3 spectroradiometer. The soil moisture content values used were derived using the hue–saturation–intensity color-change values acquired from the soil spectra. Both the partial least squares regression (PLSR) method and the SDMF-PLSR method were then incorporated to build the SOM models. The optimal model (Group I-change; j = 1) of the SDMF-PLSR method was considered as the final model, which performed much better (root-mean-square error of validation 6.99 g / kg, mean relative error of validation 28.97%) than the original PLSR model. The SDMF method when combined with PLSR shows great potential for improving SOM content estimation accuracy.

Measurement of forest biomass is a time-, money-, and labor-consuming activity. Developing methodological approaches for quantifying biomass in natural ecosystems has motivated countless investigations in several vegetation types worldwide. In this study, we developed a model to estimate the above-ground biomass (AGB) in a study site spatially located in Western Pará state, Brazilian Amazon. We applied artificial neural networks and remotely sensed data for adjusting our model based on forest inventory data. We tested four vegetation indices retrieved from a Landsat-5 thematic mapper scene and assessed their correlations with the field measured AGB. We trained 286 artificial neural networks using the vegetation stratus (two phytophysiognomies: dense lowland ombrophilous forest and dense submontane ombrophilous forest) and the normalized difference vegetation index (NDVI) and aerosol-free vegetation index as predictors variables of the AGB. We selected the artificial neural network model showing higher correlation coefficient and concentration of residuals in the center classes, which indicates higher predictive capabilities of biomass estimation. The generated model architecture (4-13-1) was composed by 4 predictor neurons in the input layer, 13 neurons in the hidden layer, activated by logistic functions, and a single neuron (the AGB in ton · ha ^{− 1}) in the output layer, activated by an identity function. We observed that the NDVI and aerosol free vegetation index showed the best performance to estimate the AGB in the study area. Using artificial neural network and a Landsat-5 image combined, we were able to accurately predict (estimation error of ∼20 % ) AGB in tropical forest. This is a promising methodological approach that can be applied to assess ecosystem services related to carbon stock in tropical regions.

Accurate estimation of tropical cyclone (TC) intensity helps to understand the evolution of TCs throughout their life cycle and plays an essential role in mitigating TC impact. Although TC intensity estimation methods based on deep learning have made significant progress, the existing techniques do not apply good methods to overcome the intensity overestimation and underestimation problems caused by the unbalanced distribution of TC data. Therefore, we propose a dynamic balance convolutional neural network to overcome these issues. The model consists of two branches, one branch is the learning of the raw data, and the other is the learning of strong (weak) TCs that account for a few data samples. Finally, the model is dynamically adjusted by adaptive trade-off parameters, gradually from the learning of the raw data to the learning of strong (weak) TCs, thus reducing errors in underestimation (overestimation) of strong (weak) TCs. Furthermore, an attention mechanism is employed to obtain the correlation between channels to improve the accuracy of TC intensity estimation further. We used globally 1285 TC cases from 2003–2016 to train the model and globally 94 TC cases from 2017 as independent test data. The results showed that the root-mean-square error of TC intensity estimation was 8.32 kt, 35% lower than that of the advanced Dvorak technique and 26% lower than that of the deep learning method visual geometry group (VGG). For a subset of 482 samples (from East Pacific and Atlantic) analyzed with reconnaissance observations, a root-mean-square intensity difference of 7.95 kt is achieved. Finally, we explored the model’s feature learning process and the contribution of each component of the satellite image to the TC intensity estimation through model visualization.

The normalized difference vegetation index (NDVI) is essential for monitoring urban green space, forest cover, and crop growth from sowing to harvesting. High-resolution images from Sentinel-2 and Landsat-8 satellites are available nowadays to get field-level phenological information at short intervals. However, cloud cover is one of the biggest hindrances in vegetation monitoring using NDVI, resulting in data gaps. To address this issue, we propose an NDVI gap-filling methodology to generate cloud-free NDVI time series from Sentinel-1 SAR data using the pix2pix generative adversarial network (GAN) model. Pix2pix GANs with generators based on U-Net and ResNet were designed, and the performance of the models was compared for both the VV and VH polarizations. The generalization capability of the models was studied using synthetic aperture radar (SAR) and NDVI image pairs with different vegetation, field sizes, and shapes. Experimental results show that the generated synthetic NDVI images can effectively substitute the cloudy images for gap-filling. Compared to ResNet, the U-Net generator has given the best results with PSNR = 29.447 dB, RMSE = 0.055, and Pearson correlation coefficient ρ = + 0.909 for VH polarization.

Photogrammetric surveying using unmanned aerial systems (UAS) and structure from motion (SfM) is a rapidly emerging technique that has been widely used as a valuable research tool in geodesy and related disciplines. We studied surveying and three-dimensional (3D) modeling based on UAS and SfM. In particular, we focus on the accuracy and quality of UAS-SfM in 3D modeling and volumetric calculation of a building with an overhanging feature. For this purpose, two aerial surveys were performed using a fixed-wing and a multirotor aircraft with nadir and oblique camera axes. For validating the results, field surveys were carried out using terrestrial laser scanner (TLS), differential global navigation satellite system, and total station. To study the performance of SfM software on the accuracy and quality of point clouds, UAS images were processed using three software packages: Agisoft PhotoScan, Pix4Dmapper, and 3Dsurvey. As a result of photogrammetric image processing, 3D dense point clouds, mesh models, and digital surface models were generated. For validating the UAS-SfM results, the models were compared with the TLS model and field measurements; absolute accuracy assessments and model versus model comparisons were performed. The most accurate and high-quality 3D models comparable to TLS models were obtained from oblique images. PhotoScan generated relatively more accurate 3D models with the horizontal and vertical accuracy of root mean square error 0.025 and 0.02 m, respectively. The maximum discrepancies in dimensions and volume measurements from the UAS-SfM-derived point cloud were at the level of 0.01 m and 0.66 m³ (0.03%). The results highlighted that using a low-cost UAS and SfM photogrammetry, high-accuracy and high-quality 3D models can be generated. The results confirm that with respect to its rapid data acquisition and centimeter-level accuracy, UAS-SfM can be considered as a suitable alternative approach to conventional measurement techniques, especially in engineering and emergency response.

The increasing pollution of soil with heavy metals has greatly damaged the ecological environment and endangered human health. Heavy metal stress affects not only biochemical constituents (i.e., chlorophyll) but also physiological functions (i.e., photosynthesis). We aimed to detect heavy metal stress in rice by integrating these two aspects. We selected Zhuzhou, Hunan Province as the study area and collected Sentinel-2 images from 2019 to 2021 and situ-measured data. First, two indicators were computed from Sentinel-2 images characterizing biochemical components and physiological functions, respectively. Second, the original time series were decomposed using the complete ensemble empirical mode decomposition with adaptive noise. Finally, a time-spectrum feature space (TSFS) model based on signal decomposition in critical periods sensitive to heavy metal stress was developed to determine the level of heavy metal stress. It showed that, regardless of individuals years, the overall accuracy and kappa coefficients of the TSFS model were >85 % and 0.8, and compared with a single indicator, combining two indicators with the signal-decomposition method significantly improved the accuracy of stress detection. This demonstrates that the integration of biochemical components with physiological functions and signal decomposition techniques can provide an important reference for crop heavy metal stress monitoring.

Multispectral sensors have been invaluable for land cover mapping. Nonetheless, cloud cover is frequent in many parts in the tropics, meaning that relying on single-date multispectral data can be problematic. The limited number of variables, i.e., only two, in conventional dual-polarized synthetic aperture radar data usually drives low classification accuracies. Researchers have explored an augmentation strategy that uses textural features, such as Haralick’s Gray Level Co-occurrence Matrix and/or decomposition features. However, this also does not produce an acceptable outcome. Another approach involves short-term temporal stacking. This masks out atmospheric disturbance and means every pixel has a single value representing the object. Our research explored this last approach to identify woody vegetation in a rural area of Indonesia from Sentinel-2 data. Ten machine learning algorithms were explored to investigate limitations and opportunities. We investigated algorithms capable of reporting variable importance to indicate crucial spectral bands. This study indicated that highly important spectra include red edge and short-wave infrared bands and to a lesser extent, the red band. Part of these bands were confirmed by the variable importance. The spectral similarity between bands 8 and 8A could be seen as a drawback when both bands are employed. Variable importance indicated a negative impact due to this similarity, suggesting that only one should be used in the classification. Overall, classification accuracy was generally high (over 96%), except in the case of the benchmark, i.e., classification and regression trees. Highest accuracy (97.6%) was provided by C5.0 model, although the difference from random forests (RFs) was negligible. The RF was favored as it produced a better estimate of timber plantations. A few algorithms were unable to separate woody vegetation types, including the popular support vector machine. Successful separation was mainly achieved by ensemble tree learners, such as RFs and the variants of extreme gradient boosting.

Remote sensing image change detection (CD) is a technical method to analyze and compare remote sensing images covering the same area with different phases to determine the process of surface change, which has important application value in natural resource management, land monitoring, and natural disaster monitoring. At present, the commonly used supervised remote sensing image CD technique requires a large number of labeled samples to train the network, which brings a large human and material cost. In contrast, the unsupervised remote sensing image CD technique does not require the construction of labeled samples; however, there is no direct correspondence between the unsupervised learning task and the downstream CD. To address the aforementioned issues, a self-supervised remote sensing image CD method based on a high frequency feature enhancement module (HFFEM) and gate attention-guided optimization unit (GAGOU) Siamese-like network is proposed. First, the network is trained using a self-supervised learning diagram to extract features that are beneficial to remote sensing image CD and better serve the downstream CD task. After that, a clustering loss function and a contrastive loss function are used to optimize the network. Second, to enhance the feature extraction capability for change edges and to address the degradation of network performance due to the entrainment of redundant features during feature fusion, HFFEM and GAGOU are proposed, respectively. Finally, comprehensive simulation experiments were conducted on the IKONOS multi-spectral datasets Mina and Riyadh, the open CD dataset Onera satellite change detection (OSCD), and the heterogeneous dataset Shuguang, showing the effectiveness of the proposed algorithms based on evaluation metrics, such as overall accuracy, kappa coefficient, and F1 score.

The work is dedicated to the analysis of spatial energy characteristics of geodetic passive glass satellites of the BLITS type for high-precision laser ranging. The article covers principles of calculation of a retroreflector spherical system (RSS), which appears to be a set of concentric layers with different refractive indices. The peculiarity of the performed analysis allows for the specifics of the RSS space application requires maximum reflected power towards the receiver determined by the phenomenon of light velocity aberration. The reflected coherent laser radiation produced by the RSS in the far zone has been analyzed using the Zemax program. An equation is derived to estimate the spatial energy characteristic (cross-section) of the BLITS-type satellites. The authors introduce the concept of an effective diameter for the output beam, which generates a far-field diffraction pattern.

In recent years, multiclass target detection in remote sensing images has become a popular research topic, and it has been widely applied in military and civilian fields. Multiclass-oriented target detection in remote sensing images presents the following challenges: small densely parked targets (SDPT), multidirectionality, interclass unbalanced number of samples (ICUNS), and hard example detection. These problems will impact the result. Therefore, to solve the abovementioned problems, we propose a multiclass-oriented target detection method in optical remote sensing images. In the detection stage, an oriented bounding box (OBB) is used to predict the angle of the target, which can overcome the problems of SDPT and multidirectionality. A cascade refined module is proposed to solve the problem of network performance degradation caused using the OBB. Second, the smooth L1 loss function is used, which can complete OBB regression by adding an angle parameter. This method can improve network performance. Finally, gradient harmonized mechanism loss is applied to the OBB. It can solve problems, such as ICUNS and hard example detection. We describe experiments conducted on the DOTA public optical remote sensing dataset. The experimental results show that this method is effective for multiclass-oriented target detection in optical remote sensing images.

Cobble size distribution (CSD) on the surface of a cobble bed is an important part of the river system, with important implications for river ecology and geomorphology. A digital image edge detection algorithm based on markers and detection lines is proposed to realize automatic statistics gathering of CSD. By making multiple detection lines outward from the markers, the cobble edge detection process of 2D images is transformed into a 1D space for analysis and processing. To correct the error detection points, an edge correction algorithm based on mutation point detection is designed. Simultaneously, a cobble edge reconstruction algorithm is proposed to restore covered cobble edges. The results of a verification experiment and comparative analysis indicate that the relative error of D₅₀, which was estimated by the algorithm in our study, is 3% in the laboratory and 8% in the field. The proposed method is reliable for the estimation of CSD with grain size range of 5 to 50 cm. In conclusion, the proposed algorithm is suitable for river geomorphology research and physical habitat studies of cobble floodplains using an unmanned aerial vehicle for remote sensing photography.

A panchromatic image is a remote sensing image that is imaged at the entire visible light band. It often has the highest spatial resolution and is widely used in the fields of resource detection, urban planning, geographic information systems, military, and national defense etc. However, the feature of single-band imaging determines that panchromatic images are usually displayed in the form of grayscale and result in some detailed differences between distinct types of ground objects that are indecipherable. It causes many difficulties for object detection applications. To address this problem, an object detection-oriented style transfer network for panchromatic remote sensing image is designed. In full consideration of the actual requirements of object detection tasks for panchromatic remote sensing image, a style transfer network based on feature fusion is designed, where a style transfer model is trained to transfer the grayscale panchromatic image into the corresponding color style. Further data preprocessing and postprocessing operations are designed to improve the quality of the transferred images and thus prevent negative transfer. DOTA dataset is used to verify the performance of the proposed algorithm. Results show that after the style transfer, the accuracy of object detection on the panchromatic remote sensing images has been significantly improved.

The image navigation and registration (INR) performance assessment for the Geostationary Operational Environmental Satellite-R (GOES-R) Advanced Baseline Imager (ABI) is an image registration process at the subpixel level, which involves the registration of images from GOES-R ABI with corresponding images from Landsat at the same scene. We present an optimized image registration algorithm for GOES-R ABI INR performance assessment (GAIPA): a gradient descent algorithm with a Sobel edge-enhancement procedure and a normalized lookup table (NLUT) image transformation (GDSN), which is an area-based image registration approach with Pearson correlation as the similarity metric. Sobel edge enhancements increase the sensitivity of Pearson correlations that lead to much improved registration accuracy. NLUT transformations minimize radiance differences between GOES-R ABI images acquired at different times and Landsat reference images on image registration outcomes, which lead to improvements in correlation values and quality of image registration outputs. Evaluations of registration accuracy and quality metrics in GDSN show considerable improvements in computational efficiency, registration accuracy, and quality over the approach implemented on the image INR performance assessment tool set. Pearson correlations can be effective and robust similarity metrics in image registration at the subpixel level with implementing Sobel edge enhancement and NLUT transformation. The statistical analysis of the image registration outputs for GAIPA shows the new algorithm provides considerable improvements in reducing the uncertainties in the registration outputs.

High speeds have been an advantage for aircraft to fly out of the adversary’s airspace fast, to escape earlier generations of infrared (IR) guided missiles that could lock-on only from the rear aspect. Advancements in IR seekers in long-wave IR (LWIR, 8 to 12 μm) band with imaging sensors in focal plane arrays made them capable of tracking aircraft also from the frontal aspect, which is a tactical advantage. Aerodynamically heated airframe surfaces are LWIR signature hot spots that increase aircraft susceptibility from front and bottom views. The conventional belief is that IR signatures of aircraft nose is only due to temperature-based emission acquired by aerodynamic heating at high Mach number (M_amb). Our study reports the LWIR contrast of aircraft nose (with nose replaced skyshine as background) by considering: (i) aerodynamic heating (important for high ε_nose, M_amb) and (ii) reflected skyshine and Earthshine (important at low ε_nose, M_amb), where skyshine is more important relative to Earthshine. Earthshine irradiance is obtained using atmospheric transmittance and view factor between visible circular Earth-disc and nose surface. The role of a transition Mach number (M_trans) in emissivity selection is identified as follows: for M_amb < M_trans, LWIR contrast decreases with ε_nose and for M_amb > M_trans, LW-IR contrast increases with ε_nose.

In recent years, support vector machine (SVM) has been widely used for supervised classification and yielded lots of exciting results within hyperspectral images (HSI) community. It seeks the optimal hyperplane, which maximizes the margin between two investigated classes. In the existing works, spatial feature usually gets involved as input and shows excellent classification performance. Unlike those available where feature generation is independent with a classifier, in our work, a hybrid framework that couples SVM classifier training with Gabor feature learning for joint optimization is proposed. It is called hybrid SVM (HSVM). Along with HSVM, two data-driven schemes are further designed. For one thing, Kronecker decomposition is applied to convert the feature learning part into a bilinear form. It aims to discover the interfilter structure characteristic of Gabor feature. For another thing, a local regularization term is appended to capture discriminant information as well. The iterative strategy is employed to optimize the normal vector of HSVM and feature factor matrices successively. Experiments with two popular HSI datasets verify the effectiveness and superiority of the proposed HSVM.

Deep learning is widely used in remote sensing field of feature recognition. Symmetric encoder–decoder network, such as UNet, is one of the most commonly used image segmentation networks, but the accuracy is often low due to its simple structure. We combine two neural network models of convolutional neural network (CNN) and Swin Transformer called modified Swin Transformer using UNet structure (MST-UNet) to achieve accurate segmentation of water bodies from remote sensing data, with Xiamen City as study area. MST-UNet is based on symmetric encoder–decoder network. We use CNN and Swin Transformer blocks to extract features from input images and capture the interdependence among different pixels, respectively. More attention is paid to global information of images. By four times upsampling to obtain predictions, the results show that the accuracy of MST-UNet is better than UNet and its improved models. The Intersection of Union (IoU), mean IoU, and Dice score on test set reach 87.80%, 92.93%, 93.08%, respectively, which verifies the feasibility of the MST-UNet. This experiment has a reference value for related studies.

In road extraction from remote sensing images, the road environment is complex and blocked by trees, buildings, and other objects, making it impossible to extract practical (continuous and complete) road information. We propose a joint attention encoder–decoder network (JAED-Net) for road extraction from remote sensing images to solve these problems. First, JAED-Net encodes a modified residual network as the backbone for road feature extraction. A joint attention module is added to the encoder to enhance the network’s ability to learn and express road features. Then, strip convolution is added to the decoder, so the network retains more spatial features, such as the width and connectivity of roads during upsampling. Finally, a hybrid weighted loss function is introduced to train the network and ensure stability because of the unbalanced ratio of road and background pixels in remote sensing images. Experimental validation of the proposed network is performed on three publicly available datasets.

Owing to the complex environment of hyperspectral image (HSI) collecting area, it is difficult to obtain an extensive number of labeled samples for HSI. Recently, many few-shot learning (FSL) algorithms based on convolutional neural network (CNN) have been employed for HSI classification in the scenery of small-scale training samples. However, a CNN-based model is unsuitable for modeling the spatial-spectral information with long-range dependency. The transformer has proved its superiority in modeling the long-range dependency. Inspired by this, an improved spatial-spectral transformer for HSI few-shot classification (HFC-SST) is proposed to deeply extract the local spatial-spectral information with only a few labeled samples. The contribution of this letter is twofold. First, a local spatial-spectral sequence generation method based on spatial-spectral correlation analysis and adjacent position information is proposed to generate the input sequence for transformer. Second, a local spatial-spectral feature extraction network based on the transformer is proposed to further exploit the spatial-spectral feature information on the input sequence. Experimental results on HFC with four datasets confirm that our proposed HFC-SST algorithm can achieve higher classification accuracy than the traditional CNN algorithms and the HSI FSL algorithms.

Object detection is a fundamental part of autonomous driving algorithms, and with the promotions of transformers in a couple of years, numerous computer vision tasks are integrating transformers into object detectors to acquire a better generalization ability. Building a pure transformer-based detector seems to be a wonderful choice; however, transformers are not omnipotent, and they come with painful drawbacks. Its fundamental operator, multi-head self-attention (MHSA), suffers from the need for computational resources due to its quadratic complexity, which demands an unreasonably high memory usage and critically low throughput. To address this issue, we use a convolution operation to simulate MHSA from transformers by referencing the philosophy and principle of MHSA and making an application migration on convolutional neural networks (CNNs). This gives a detector with power and speed simultaneously. Furthermore, a multi-scale pyramidal feature extractor gives the detector a better view over various scales. In general, our proposed object detector mainly follows the philosophy of attention mechanism, which is implemented by a multi-scale feature pyramidal CNN encoder that simulates the transformer, and a real transformer query neck to extract all of the objects once and, eventually, feed them to the output heads. After training on the COCO2017 dataset, by combining the construction philosophy of the object detector and the philosophy and characteristics of the transformer, our FPDT-Tiny gives an average precision (AP) of up to 34.1 in 150 lower epochs, which is 16.0 and 10.8 higher than CNN-based YOLOv3-Base and SSD-300, respectively. Also, the AP given by our FPDT-Small is up to 37.7 under the same epoch, which is 10.4 and 7.9 higher than the transformer-based detector YOLOS-Small and DETR-ResNet-152, respectively, also demonstrating a comparable performance.

Remote sensing image (RSI) analysis heavily relies on change detection (CD) to identify differences between bitemporal images. Although significant progress has been made in CD methods, it remains challenging to distinguish between pseudo-changes caused by imaging angles, weather conditions, seasonal variations, etc. This is because most existing networks cannot fully utilize multi-scale features to narrow the semantic gap between cross-scale features. To address this issue, we propose a cascade cross-scale network (CCSNet) with a Siamese structure for CD in RSIs. Our CCSNet uses cascade Siamese encoders to extract multi-scale bitemporal features. These extracted features are then fused using the proposed Diff-Conc combination at the encoding stage. Cross-scale skip connections are used to continuously transmit shallow features to the decoders, in which a cross-scale attention module is designed to fuse shallow and deep features. This network structure effectively bridges the semantic gap between cross-scale features and helps eliminate pseudo-changes. We evaluated our method on two popular benchmark CD datasets (LEVIR-CD and WHU-CD). The experimental results demonstrate that CCSNet outperforms the state-of-the-art methods.

Accurately identifying the semantic information of complex objects is a challenging problem in semantic segmentation of remote sensing images. We propose a bi-encoder network for semantic segmentation of complex targets, called the SN-Unetformer. It combines ConvNeXt and Swin Transformer into a bi-encoder and constructs a feature fusion module (FFM) to fully integrate the semantic information of the bi-encoder by exploiting channel dependence. Moreover, an efficient attention mechanism has been introduced to model the global–local relationship. To the best of our knowledge, our proposed network is innovative, as it is the first method to combine two popular networks, ConvNeXt and the Swin Transformer, into a dual encoder. Our SN-Unetformer has been tested on large-scale Vaihingen and Potsdam datasets, as well as the LoveDA dataset, with significant challenges. Compared to current advanced methods for semantic segmentation for remote sensing images, our accuracy is significantly better. In particular, our method achieves 84.3% of mean intersection over union on the Vaihingen dataset, which is the best result currently available for this dataset.

Hyperspectral image (HSI) classification is a hot topic in the field of remote sensing applications. However, due to the high-dimensional and extensive spectral and spatial information of HSIs, effective feature extraction is difficult, which leads to a lower accuracy of HSI classification. In this work, a convolutional neural network (CNN) approach based on a multiscale hybrid 3D-2D-attention mechanism (MHAC) is proposed. Linear discriminant analysis dimension reduction is performed for HSIs, and the central pixel and its adjacent pixels are input into the network as a whole. Based on a multi-scale three-dimensional convolutional neural network (3D CNN) and a two-dimensional convolutional neural network (2D CNN), fine spectral and spatial features are automatically extracted, which improves the network classification capability. At the same time, the channel attention mechanism is embedded in the 2D CNN, and the spatial attention mechanism is embedded in the 3D CNN, which can assign different weights to the retrieved features, effectively utilizing the spatial spectral information to optimize features and improve the feature extraction ability. The experimental results demonstrate that the MHAC achieves an accuracy of 98.72%, 97.60%, and 97.90% on the WHU-Hi dataset and 98.93% on the Indian-Pines dataset based on only 100 samples. The MHAC model proposed in this paper has a good feature extraction ability that effectively reduces the impact of high-dimensional data and small sample problems on classification and improves the accuracy of HSIs classification.

An improved cloud shadow removal algorithm for high spatial resolution optical satellite data over land is presented. The method is based on the matched filter method, which consists of the calculation of a covariance matrix and the corresponding zero-reflectance matched filter vector and the computation of the shadow function. The new additions consist of the usage of an improved cloud shadow map and further evaluations performed on the shadow function. The performance of the cloud shadow removal algorithm incorporated in the software package Python-based atmospheric correction (PACO) is compared to the deshadowing algorithm in atmospheric correction on a set of 25 Sentinel-2 scenes distributed over the globe covering a wide variety of environments and climates. Furthermore, an evaluation of the relative ratio between clear and shadow pixels with and without deshadowing is performed. The visual, spectral, and statistical results show that the new additions performed on the deshadowing algorithm can improve the cloud shadow removal performance used so far.

Detection of high-speed maneuvering targets along with motion parameters estimation (MPE) are some of the challenging problems for modern radar systems. Such targets usually induce linear range migration (LRM), quadratic range migration (QRM), and Doppler frequency migration (DFM) problems in signal processing, resulting in severe performance degradation, especially when long time integration is needed for weak target detection. We propose an algorithm for coherent integration and MPE by combining the techniques of generalized dechirp (GD) and new axis rotation moving target detection (NAR-MTD), i.e., the GD-NAR-MTD algorithm, to solve the above problems. Specifically, the GD and NAR approaches can not only correct LRM, QRM, and DFM simultaneously with reduced rotation error in coordinate system transformation but also can abate the computational burden via frequency-domain implementation with rounding and complex addressing operations avoided, so the target’s echo energies can be well accumulated by MTD finally. Both simulation and practical radar experiments are conducted to validate the proposed algorithm. At the same time, the results by the other six typical algorithms are compared for demonstrating the superior performance of the proposed algorithm on detection and MPE. In addition, the equivalence between the practical radar experiment and the simulation experiment is analyzed by comparing the induced RMs and DFMs, which shows that the scaled equivalent experiment is persuasive in algorithm validation.

Superpixel segmentation is an important image preprocessing technology that can aggregate adjacent pixels with similar characteristics to reduce the complexity of subsequent tasks. However, with the increasing spatial resolution of remote sensing images, maintaining accurate boundary information of ground objects during superpixel segmentation has become a challenge. We propose a feature reconstruction method based on edge detection and adaptive morphological reconstruction to enhance feature representation for superpixel segmentation. Specifically, we employ the structured forest model to extract robust edges from the image, and an adaptive morphological reconstruction method is proposed to eliminate noise from the edge information. Furthermore, a feature fusion method based on wavelet transform is constructed to integrate the extracted edge information with the original image, which further enhances the segmentation accuracy of image edges in superpixel segmentation. Experimental results on three datasets of natural and remote sensing images demonstrate that the proposed method can effectively construct the feature representation of images and achieve better superpixel segmentation results than other methods.

Synthetic aperture radar (SAR) aircraft detection methods based on deep learning have become a current research hotspot. However, considerable challenges still remain due to the scattering feature of aircraft, variations in aircraft size, and interference from complex scenarios. To tackle these problems, the scattering feature extraction and fuse network (SFEF-Net) is proposed. First, considering the scattering characteristics of aircraft, we propose a scattering feature extraction and relation enhancement (SFERE) backbone based on the deformable convolution and the global context block. The SFERE backbone is used to extract the scattering feature of aircraft and model the correlation of scattering points. Furthermore, to enhance the detection ability for multi-scale aircraft targets in complex scenes, we redesign an attention bidirectional feature fusion pyramid (ABFFP). Two novel modules are proposed in ABFFP, namely, the attention guidance feature fusion (AGFF) module and the residual efficient channel attention (RECA) module. The AGFF module is proposed to suppress the interference of backgrounds and aggregate the multi-level feature maps. After the feature fusion operation, the output feature maps contain richer channel information, but there is some redundant information that could reduce the accuracy. Therefore, we adopt the RECA module to further select useful information in the channel dimension. To demonstrate the effectiveness of SFEF-Net, SAR aircraft images from the Gaofen-3 system are utilized in the experiments. The detection results show that the proposed model achieves competitive performance with an average precision of 95.5%.

To analyze the differences in the inversion of Gravity Recovery and Climate Experiment (GRACE) time-variable gravity field models, our study compares the results of ice sheet mass changes from six versions of GRACE monthly gravity field solutions, including the Center for Space Research (CSR) RL06, Jet Propulsion Laboratory (JPL) RL06, German Research Centre for Geoscience (GFZ) RL06, WHU-GPD01s, Tongji-Grace 2018, and HUST-Grace 2020 models, in Antarctica and Greenland. We analyzed the differences of ice sheet mass predicted rates in the temporal, spatial, and spectral domains. In the temporal domain, these differences have a systematic bias among various models and increase over time, especially after 2011. In the spatial domain, the Antarctic ice sheet mass loss rate of WHU-GPD01s is the most consistent with CSR, and the results of GFZ are significantly different than other models (∼15 Gt / year). The Greenland ice sheet mass loss rates of various models are consistent. In the spectral domain, the JPL and CSR models are in great accordance. The differences between the CSR model and the HUST-Grace2020, WHU-GPD01s, and Tongji-Grace2018 models are mainly distributed in the degrees of 20 to 30, 10 to 45, and 25 to 55, respectively. Moreover, the HUST-Grace2020 and CSR models have the smallest spatial discrepancy except for polar regions, and the strip errors for the trend rate of the global equivalent water height are small, which is suitable for research in the mid and low latitudes.