The compact Reconnaissance Imaging Spectrometer for Mars (CRISM) has significantly advanced our understanding of the mineralogy of Mars. With its enhanced spectral and spatial resolution, CRISM has enabled the identification and characterization of various minerals on the Martian surface, providing valuable insights into Mars’ past climate and geologic history, as well as the evolution of the planet’s atmosphere and climate. We present a comprehensive review of mineral identification on Mars using CRISM data. We discuss the data description, pre-processing techniques, different spectrum libraries, geological characteristics used for mineral identification, challenges, and methodologies used for mineral classification, such as learning models, probabilistic methods, and neural networks. We highlight major findings of minerals on the Martian surface and discuss validation techniques. We conclude with a discussion of further research to address the existing gaps and challenges in this field. Overall, we provide a general understanding of mineral classification using CRISM data and could serve as a helpful resource for researchers and scientists interested in planetary remote sensing and mineral identification on the Martian surface.

Non-ponded aerobic rice offers the opportunity for substantial global savings of irrigated water. Site and environment specific soil tension thresholds have been identified for non-ponded rice production to save water without adverse yield consequences. However, soil tension readings are point-source measures that are not necessarily representative of a given irrigation area/paddy and become costly when required in each paddy. Remote sensing offers spatial advantages and has been demonstrated in many crops to accurately estimate crop evapotranspiration (ETc), which is often used to schedule irrigation. This study uses soil moisture tension (SMT) data measured in a commercial aerobic rice experiment conducted in temperate Australia on a heavy clay soil to demonstrate the relationship between soil tension and remotely sensed estimated cumulative ETc between irrigation events using open access satellite imagery. The model was used to predict SMT in the second growing season during the vegetative period after establishment until PI with a root mean squared error of ±5.8 kPa. This demonstrates the ability of irrigators to schedule irrigation in water saving rice using open satellite-derived data that can be incorporated into an automated irrigation system, without the need for costly in-field soil moisture sensors. Future research to integrate forecast crop evapotranspiration could enable soil moisture forecasting days in advance, providing irrigators more time to plan irrigation activities.

Social distancing is a feasible solution to break the chain of the spread of coronavirus disease 2019 (COVID-19). A human crowd detection model was trained with a computational load that can be handled by a companion computer on the unmanned aerial vehicle (UAV) to minimize the spread of COVID-19. The model is designed to be able to measure social distance between people, whether it exceeds predetermined safe limits (1.5 m). The convolutional neural network model was trained using a dataset of 9600 images featuring humans, cyclists, and motorcyclists, with an allocation of 200 images each for testing and hyperparameter tuning. The image dataset was extracted from videos recorded above the UAV in the Institut Teknologi Bandung area, capturing diverse crowd scenarios throughout the day. The pre-trained model for transfer learning method is a single shot detector with MobileNet, ResNet50, and ResNet101 architectures. The measurement of the estimated social distance uses the Euclidian distance with the average Indonesian human as a reference, which is 1.6 m. MobileNet V2 was chosen as a crowd detection model with a lightweight size, which is only 19 MB and the average detection runtime for a single image is only 0.606s, in accordance with the load for the onboard companion computer. MobileNet V2 is also able to detect crowds of people well with the precision value reaching 84.9% and the recall value reaching 87.8%.

Semantic segmentation of high-resolution remote sensing images based on deep learning has become a hot research topic and has been widely applied. At present, based on the structure of the convolutional neural network, when extracting target features through multiple layer convolutional layers, it is easy to cause the loss of small target features and fuzzy boundary of ground object classification. Therefore, we propose a remote sensing image semantic segmentation method P-Net to detect small target and enhance edge feature. The proposed network was based on an Encoder-Decoder structure. The decoder included the following components: a progressive small target feature enhancement network (IFEN), a boundary thinning module (BRM), and a feature aggregation module (FIAM). Firstly, the dense side output features of the encoder network were utilized to learn and acquired small target feature information and target edge features. Secondly, the pyramid segmentation attention module was introduced to effectively extract fine-grained and multi-scale spatial information. This module enhanced the feature expression of small targets and obtained high-level semantic feature information. The boundary refinement module was designed to refine the low-level spatial feature information extracted by the encoder. Finally, in order to improve the accuracy of remote sensing image object segmentation boundaries, skip connections were used to fuse high-level semantic information and low-level spatial information acrossed layers. These skip connections had the same spatial resolution but different semantic information. In this paper, six evaluation indices including mean intersection over union, frequency weighted intersection over union, pixel accuracy, F1, recall, and precision were used to verify on two public datasets of high-resolution remote sensing images, Gaofen image dataset (GID) and wuhan dense labeling dataset (WHDLD). In the GID dataset, each index reached 78.90%, 78.87%, 87.76%, 87.74%, 87.51%, and 88.04%, respectively; in the WHDLD dataset, each index reached 63.21%, 75.20%, 84.67%, 75.79%, 76.56%, and 75.45%, respectively. The results show that the performance of proposed method is better than that of DeepLabv3+, U-NET, PSPNet, and DUC_HDC methods. More precisely, the recognition performance of small target features is better, and the boundary obtained between object categories is clearer.

Surface cover diversity and the complexity of geological structures can seriously impact the accuracy of mineral mapping. To address this issue, we propose a method for lithological classification and analysis based on random forest (RF) and multiple features. Feature vectors, including spectral, polarization, texture, and terrain features, are constructed to provide multidimensional information. Subsequently, these feature vectors are screened based on their discriminative properties for different lithologies to reduce feature redundancy. Finally, the results of lithological classification can be obtained using the RF algorithm based on the selected features. In the experiments conducted in the Qulong copper deposit area, data from Sentinel-1A, Sentinel-2A, and Terra satellites were used to extract multidimensional features. After calculating the Bhattacharyya distance and analyzing the probability density distribution, 17 features selected were input into the RF classifier, achieving an accuracy of 88.83% in lithological classification. This represents a 7.5% improvement compared to exclusively relying on spectral features, and suggests that the proposed method of combining spectral, polarization, texture, and terrain features provides new possibilities for improving the accuracy of field lithological classification.

Land cover classification is a vital application of polarimetric synthetic aperture radar (PolSAR) images in various fields, such as agriculture monitoring and urban assessment. We introduce a modified and enhanced PolSAR image classification method, combining six decomposition techniques, a support vector machine (SVM) based classifier, and a probabilistic voting ensemble (PVE) model. Our method addresses the challenges posed by the complexity of PolSAR data and the limited availability of labeled samples. The core of our approach lies in integrating multiple decomposition techniques as feature extractors, aiming to capture diverse scattering behaviors and uncover valuable information related to land cover characteristics. These techniques include the Huynen, Cloude, Freeman and Durden, HAAlpha, Yamaguchi, and Vanzyl decomposition methods. The extracted features are then utilized as inputs for training the SVM base classifier. To enhance classification performance, a PVE model is used to combine predictions from each decomposition technique, considering the individual prediction confidence and the characteristics of the decomposition methods. The decision fusion process is applied to integrate diverse predictions based on the majority voting and estimated class probability, providing a more robust and reliable final label prediction and thereby improving the overall accuracy of the classification process. Experimental analyses are conducted on airborne and spaceborne PolSAR images, covering various bands and land cover types, to evaluate the effectiveness and robustness of our proposed method. The experimental results demonstrate that our approach yields more confident class predictions than alternative methods.

Modifications in submerged aquatic vegetation (SAV) spatial and temporal abundance patterns indicate changes in marine environmental conditions or physical disturbances and need to be monitored. Vegetation percent cover (%cover) is recognized as one of the key parameters in SAV monitoring. Coastal waters of the Baltic Sea are often turbid and contain high amount of colored dissolved organic matter. These factors significantly reduce the water depth, where benthic parameters can be detected by remote sensing. Field campaigns were carried out in a low-transparency Pärnu Bay area to assess to what extent multispectral Sentinel-2 (S2) satellite can be used for SAV %cover mapping in such waters. An average depth restriction for S2 benthic vegetation detection remained near 1.5 to 2.0 m. Empirical and physics-based methods were applied to S2 imagery to compare their performance for SAV %cover retrieval. Both methods identified similar %cover patterns. Model validation results showed that R² of the best-performing models remained between 0.56 and 0.66 and root-mean-square error between 22.11 and 28.06. As physics-based inversion models do not require extensive set of training data for model calibration, those can be used for retrospective time series analysis across multitemporal images.

Accurate lithological mapping is a difficult task through standard image processing techniques. We utilize the application of different machine learning (ML) algorithms on dual polarimetric synthetic aperture radar (SAR), optical data, and surface elevation images to map various lithologies in parts of Jaisalmer district of Rajasthan, India. Different SAR-derived textural and decomposition parameters were also used to improve the discrimination of various lithology units. Further, to improve the classification accuracy, different ML-based feature importance models, such as XGboost, decision tree, and random forest were implemented to select the useful bands for the classification of lithology. A total of 14 different ML classifiers were applied, and the best classifier was chosen after comparing their accuracies (overall accuracy, kappa coefficient, F1 score, and ROC-AUC curve) to map the lithology. Out of all of the classifiers used in this study, light gradient boosting machine (lightgbm) performed better in discriminating lithology (OA = 0.80, kappa coefficient = 0.75, and F1 score 0.79). In addition, the AUC values (>0.9 in all lithology units) were obtained for the “lightgbm” model, which is indicative of accurate discrimination of different lithological classes.

The ARCSTONE project objective is to acquire accurate measurements of the spectral lunar reflectance from space, allowing the Moon to be used as a high-accuracy SI-traceable calibration reference by spaceborne sensors in low-Earth and geostationary orbits. The required spectral range is 350 to 2300 nm with 4-nm sampling. The ARCSTONE approach is to measure solar and lunar spectral irradiances with a single set of optics and determine spectrally resolved lunar reflectances via a direct ratioing method, eliminating long-term optical degradation effects. Lunar-irradiance values, derived from these direct reflectance measurements, are enabled by independently measured SI-traceable spectral solar irradiances, essentially using the Sun as an on-orbit calibration reference. In an initial attempt to demonstrate this approach, a prototype ultraviolet-visible-near infrared (348 to 910 nm) instrument was designed, fully assembled, characterized, and field tested. Our results demonstrate that this prototype ARCSTONE instrument provides a dynamic range larger than 10⁶, which is necessary to directly measure both the solar and lunar signals, and suggest uncertainties better than 0.5% (k = 1) in measuring lunar spectra can be achieved under proper operational scenarios. We present the design, characterization, and proof-of-concept field-test of the ARCSTONE instrument prototype.

The advancement of near-ground remote sensing and artificial intelligence techniques has revolutionized field surveys, replacing traditional manual methods. Nevertheless, understanding and exploring the growth patterns and intricate morphology of natural elm tree crowns present significant challenges, especially when attempting to extract their features, which are often susceptible to interference from surrounding grass and vegetation. In addition, existing detection and segmentation models based on convolutional neural networks exhibit redundancies in their network architectures and employ less efficient algorithms, such as mask region-based convolutional neural networks. As a result, these models may not be the most suitable options for analyzing extensive and highly detailed remote-sensing image data. We focus on detecting trees in semi-arid regions and extracting their canopy parameters, such as canopy width and area. A training set is established by outlining a total of 20,594 tree canopies on high-spatial resolution unmanned aerial vehicle images. A two-stage instance segmentation model is proposed to develop a method for individual tree detection and efficient extraction of canopy parameters in complex natural environments. The results demonstrate the method’s capability to accurately detect the location, number, and canopy parameters (e.g., crown width and area) of individual trees in diverse natural scenes. The model achieves a detection speed of 13.3 fps@1024, with the model weight parameters totaling 8.08 M and computation requiring 8.96 Giga floating point operations per seconds (GFLOPs). Moreover, the detection accuracy and segmentation accuracy of individual trees on the validation set are reported as 0.463 and 0.465, respectively. Compared with Mack RCNN and Mask Scoring RCNN, the proposed method reduces the weight parameters and computational complexity of the model by 82.4%, 83.5% and 96.8%, 92.8%, respectively, while increasing the inference speed by 47.4% and 26.3%. This method offers an efficient and accurate solution for obtaining the structural parameters of individual trees.

Landslide susceptibility prediction (LSP) is crucial for hazard prevention and geological risk assessment. Support vector machine (SVM) is widely used for LSP, but its parameter optimization problem affects the prediction accuracy and generalization ability of the model, and variations in parameter combinations may result in different prediction outcomes, which brings some challenges to the application of the model. We present a procedure for LSP using the gray wolf optimization (GWO) algorithm to optimize SVM models in the Sanshui District, Foshan City, China. Fifteen factors affecting landslide susceptibility are selected and processed by the natural breakpoint method and normalization method. To prevent overfitting and improve the generalization ability of the model, the five factors with high correlation are excluded using the Pearson correlation coefficient. The grid search method and GWO are used to optimize the SVM parameters and establish the GWO-SVM model. The results indicated that the GWO-SVM model, which incorporated the normalization method (referred to as GWO-SVM-NOR), demonstrated superior predictive accuracy, achieving an impressive area under the curve value of 0.886. The gray wolf algorithm improves the fitting accuracy of SVM and optimizes the model prediction performance with better stability, which is suitable for predicting areas susceptible to landslides.

The effect of uranium mining on ground deformation is a relatively unexplored area, especially in terms of surface subsidence related to subsurface ore removal. We use interferometric synthetic aperture radar and spatiotemporal techniques to characterize subsidence signals at the McArthur River underground mine in Canada and the Four Mile in situ leach mine in Australia. We enhance the signal-to-noise ratio of our datasets via time-series techniques and compare results from active periods with results during inactivity to establish a baseline for mining-related signals. We then relate observed surface subsidence to subsurface volumetric strain rates via a voxel parameterization and Bayesian, geostatistical inversion. We use priors on our volumetric strain rates to identify whether these rates are best attributed to ore removal or if additional factors are contributing to subsidence at these sites. We find that the subsidence at McArthur River is best explained by a combination of ore removal and thermal contraction resulting from ground freezing practices. Ore removal via solution extraction alone explains the subsidence at Four Mile, although the localized subsidence pattern and resulting strain rates suggest an intricate combination of sinks and sources in the field, possibly from injection and production well locations and the subsequent flow of solution.

In precision agriculture, vegetation and soil are monitored by multispectral sensors that can observe outside the visible bands. In contrast to satellites and manned aircraft, unmanned aerial vehicles (UAVs) allow anyone to easily acquire near-real time data at a reasonable price. However, UAV images do not account for the anisotropic reflectance and solar irradiance from the ground surface, so extracting the reflectance of vegetation is difficult. To solve this problem, this study developed a bidirectional reflectance distribution function (BRDF) that expresses the anisotropic reflectance of the Earth’s surface as a function of the geometric relationship with the UAV sensor and the Sun. To compensate for the effect of changes in solar incident energy due to clouds and solar irradiance, the solar irradiance was measured and corrected on the ground rather than in the air to avoid errors due to the flight attitude. Before processing by the BRDF and correcting for the solar irradiance, the UAV obtained striated orthomosaic images for which the vegetation indices were affected by the position and attitude of the Sun and the UAV sensor. After the correction, consistent values were calculated for the vegetation indices throughout the images. The accuracy of the UAV data was analyzed by comparison with Sentinel 2A. Reflectance differences are 0.02% to 6.37% from the image without correction. After applying the correction, it reduced to 0.27%, 0.61%, 0.16%, and 0.65% from the blue, green, red, and near-infrared bands, respectively. This study is valuable for obtaining accurate values for vegetation indices under a wide range of weather and geometric conditions at different sites because UAVs to collect images are a rare case under optimal conditions.

Soil maps are essential sources for a diverse range of agricultural and environmental studies; hence, the detection of soil properties using remote sensing technology is a hot topic. Satellites carrying hyperspectral sensors provide possibilities for the estimation of soil properties. But, the main obstacle in soil classification with remote sensing methods is the vegetation that has a spectral signature that mixes with that of the soil. The objective of this study is to detect soil texture properties after eliminating the effects of vegetation using hyperspectral imaging data and reducing the noise by fusion. First, the endmembers common to all images and their abundances are determined. Then the endmembers are classified as stable ones (soil, rock, etc.) and unstable ones (green vegetation, dry vegetation, etc.). This method eliminates vegetation from the images with orthogonal subspace projection (OSP) and fuses multiple images with the weighted mean for a better signal-to-noise-ratio. Finally, the fused image is classified to obtain the soil maps. The method is tested on synthetic images and hyperion hyperspectral images of an area in Texas, United States. With three synthetic images, the individual classification results are 89.14%, 89.81%, and 93.79%. After OSP, the rates increase to 92.23%, 93.13%, and 95.38%, respectively, whereas it increases to 96.97% with fusion. With real images from the dates 22/06/2013, 25/09/2013, and 24/10/2013, the classification accuracies increase from 70.51%, 68.87%, and 63.18% to 71.96%, 71.78%, and 64.17%, respectively. Fusion provides a better improvement in classification with a 75.27% accuracy. The results for the analysis of the real images from 2016 yield similar improvements. The classification accuracies increase from 57.07%, 62.81%, and 63.80% to 58.99%, 63.93%, and 66.33%, respectively. Fusion also provides a better classification accuracy of 69.02% for this experiment. The results show that the method can improve the classification accuracy with the elimination of vegetation and with the fusion of multiple images. The approach is promising and can be applied to various other classification tasks.

Aquatic algae and cyanobacteria can impair water-quality and pose risks to human and animal health. Several metrics of in-situ water-quality, including chlorophyll-a, phycocyanin, turbidity, Secchi depth, phytoplankton taxonomy, and hyperspectral reflectance, were collected in coordination with Sentinel-2 satellite overpasses to ascertain water-quality conditions and calibrate satellite detection and estimation of chlorophyll-a concentration. The performance of multiple satellite chlorophyll-a detection indices was evaluated by comparing satellite imagery to field observations of chlorophyll-a concentrations. Seventeen chlorophyll-a spectral indices were implemented using the ACOLITE atmosphere correction; the top performing indices were selected for further evaluation using the Sen2Cor and MAIN atmosphere corrections. The Moses three-band spectral index delivered the strongest linear agreement with field measurements of chlorophyll-a concentration across all reservoir sampling sites (R² = 0.70). Compared to open-water sites, the Moses three-band spectral index delivered better linear agreement with chlorophyll-a field measurements at inlet sites where there was a greater abundance of near surface aquatic chlorophyll-a concentrations, and the overall chlorophyll-a hyperspectral reflectance signal was stronger. Chlorophyll-a concentration estimates were implemented in a cloud-computation remote sensing platform designed for regional scale remote sensing analysis to map spatiotemporal patterns of aquatic chlorophyll-a across 10 study reservoirs located primarily in north Texas.

Detecting small objects in remote sensing images is a challenging task. Existing object detectors for remote sensing images suffer from two issues: (1) insufficient feature extraction for small objects in the backbone network and (2) feature misalignment and information loss for small objects in the neck network, leading to poor detection performance on small objects. To address these challenges, a small object detector named CABDet for remote sensing images that combines context and attention mechanisms is proposed. Specifically, an enhanced ResNet50 is designed as a novel backbone network that adaptively adjusts the size of receptive fields to fully extract feature information of small objects. Additionally, an adaptive multiscale feature pyramid network (AM-FPN) is proposed. To alleviate the problem of feature misalignment for small objects, AM-FPN leverages self-attention mechanisms to establish semantic and spatial dependencies between adjacent feature layers. Then to mitigate the issue of information loss for small objects, AM-FPN captures semantic dependencies between subregions of current layer features through self-attention mechanisms to preserve channel information. Extensive experiments were conducted on two demanding remote sensing datasets, namely dataset for object detection in aerial images and UCAS-high resolution aerial object detection dataset, to demonstrate the effectiveness of the proposed methodology in achieving superior detection performance when compared with contemporary state-of-the-art approaches.

With the emergence of keypoint matching technology, the correspondence-based point cloud registration (PCR) method has gained increasing attention. However, the correspondences generated by keypoint matching technology contain extremely high outlier rates, resulting in the correspondence-based PCR method facing issues of high computational complexity and low precision registration. We propose a correspondence-based PCR method using a coarse-to-fine outlier removal strategy with O ( N ) complexity. First, we propose a coarse outlier removal module based on linearly related properties, i.e., we build a deviation matrix that can measure each correspondence deviation the degree away from the ideal inlier. The module can reduce the number of correspondences and the outlier rates. Then, we propose a fine outlier removal module that adopts each correspondence to identify outlier based on the spatial geometric mapping invariance. Finally, to increase registration accuracy, we introduce a graduated non-convexity with Tukey’s biweight method. It can avoid the solution falling into the local minimum and better reduce the influence of outliers. Experimental results show that the proposed method is robust at outlier rates above 99% and is faster than state-of-the-art methods.

Two-dimensional compact variational mode decomposition (2-D-C-VMD) is a new data mining and signal analysis technology. It is suitable for feature extraction and classification of hyperspectral images (HSIs). However, given an appropriate prior parameter K value is the primary condition for us to achieve 2-D-C-VMD for feature extraction. Different prior-parameter K values will produce results beyond our expectations in different application scenarios. Consequently, combining HSIs as an application scenario and aiming for feature extraction and classification recognition effect, we determine the optimal prior parameter K* of 2-D-C-VMD and construct a set of optimal high-pass, low-pass, and band-pass filters using the optimal threshold K* to improve the classification accuracy. In the experiment, three open HSIs data sets with different characteristics were used to verify the feasibility and advantages of the scheme, and the test analysis was carried out under different training sets. The research results show that the scheme has better competitiveness in the comprehensive classification accuracy.

Timely monitoring and evaluation of ecological restoration in mining areas is crucial. Based on remote sensing data and deep-learning models, the dynamic changes of bare rock area and vegetation in open-pit mine can be quantitatively monitored and analyzed. Current mining area feature extraction algorithms are limited by single-scale approaches and insufficient information fusion, resulting in low recognition rates. To address this, we proposed an improved Bilateral Segmentation Network (BiSeNetV2) semantic segmentation model (BiSeNetV2 + MSFE + SegHead, BMS), which combines multiscale feature extraction (MSFE) module and segmentation head (SegHead) structures. We utilized BMS model to conduct research on the classification and change monitoring of vegetation areas and mining areas. Our results demonstrated that the accuracy evaluation indicators aAcc, mAcc, and MIoU of the BMS model were better than those of the BiSeNetV2 model, with improvements of 3.5%, 5.5%, and 7.9%, respectively. Meanwhile, compared to the short-term dense concatenate and Twins-PCPVT deep-learning models, the BMS model improved aAcc, mAcc, and MIoU by 3.4%, 8.0%, and 7.3% and 4.4%, 1.1%, and 8.6%, respectively. Accurate and efficient research on ground object classification methods enables quantitative evaluation of mining area environment recovery, providing crucial technical support for ecological monitoring, planning, and governance.

OS-CFAR is widely used in synthetic aperture radar automatic target detection. It rank orders the data to obtain the k’th order statistic (OS). But it suffers from high computational cost because of its sorting process. We introduce a partial mean and multi-section (PMMS) rank order filtering method, which can effectively reduce the computational cost. The PMMS finds the k’th OS by dividing the data into several sections and only using a subset of the data to calculate the mean. A theoretical analysis of computational complexity is given. Simulation results show that, on average, the proposed method reduces execution time by 29.6% for uniformly distributed data and 34.4% for exponentially distributed data. Also, the best time reduction methods for uniformly and exponentially distributed data are presented.

To better cope with significant nonlinear radiation distortions and severe geometric distortions in multisource remote sensing image matching, we propose a high-precision image matching method for multisource remote sensing images based on extended phase consistency (EPC). First, feature detection was performed based on an anisotropic weighted moment equation. We used the EPC model to construct multi-orientation weighted phase sharpness map and the multiscale phase orientation index map and built a gradient location orientation histogram-like local feature descriptor to obtain the corresponding features based on the feature maps. The spatial relationship constraint function was introduced to screen the obtained initial matches to reduce the negative effects caused by the high proportion of outliers. The remaining outliers were removed using the marginalizing sample consensus (MAGSAC) method, and the transformation parameters of the sensed image were calculated based on the iteratively optimized homography matrix model. Based on the feature detection results obtained from two distinct groups of synthetic data, it is evident that the proposed detector outperforms the three classical detectors in terms of repeatability and uniformity. The registration performance experiment data were divided into six groups, and the proposed method was compared with the log-Gabor histogram descriptor, radiation-variation insensitive feature transform, histogram of absolute phase consistency gradients, and local phase sharpness orientation methods. The experimental results showed that the proposed method is significantly better than the existing algorithms in terms of image matching performance and achieves high registration accuracy.

How to fully capture high-frequency information is an important issue in the remote sensing image super-resolution (SR) task. Most of the existing convolutional neural network based methods usually apply the attention mechanism to capture the high-frequency information. However, it is often insufficient since the remote sensing images usually contain more high-frequency information than natural images. Recently, some studies try to transform the original image into the wavelet domain to capture more high-frequency information. However, we observe that these methods usually apply similar network structures to learn different wavelet components, which will be difficult to fully capture the different features. To solve this issue, we propose a method named multi-frequency parallel wavelet transform network (MFPWTN) for remote sensing image SR. Specifically, we initially design two different network structures to reconstruct the high-frequency and low-frequency wavelet components, which can fully capture the characteristics of different frequencies. Subsequently, we introduce a high-frequency fusion module to enhance the information transmission among different high-frequency wavelet components. In addition, we employ the dilated convolution to establish the network structure for reconstructing the low-frequency wavelet component, which allows us to capture different receptive fields by using relatively few parameters. The experimental results on two public remote sensing datasets, UCMerced-LandUse and NWPU-RESISC45, show that the proposed MFPWTN can get superior performance over many existing state-of-the-art algorithms.

Small ship instance segmentation from synthetic aperture radar (SAR) images is a challenging task. Because small ships have smaller scales, indistinct contours, and weak feature response. In addition, background interference and clutter make feature extraction of small ships more difficult. To solve this issue, we propose a scale-aware dimension-wise attention network (SA-DWA-Net) for better small ship instance segmentation in SAR images. SA-DWA-Net has two subnetworks to ensure its desirable instance segmentation of small ships. The first is a scale-aware subnetwork that can fully use low-level location-sensitive information to achieve representative small ship features. The second is a dimension-wise attention subnetwork that can fully utilize high-level semantics-sensitive information for refined small ship feature expression. We perform experiments on two open SSDD and HRSID datasets to verify the effectiveness of the proposed method. Quantitative experimental results show the state-of-the-art SAR ship instance segmentation performance of the proposed SA-DWA-Net. Specifically, SA-DWA-Net surpasses the existing best model by 2.2% box detection average precision (AP) and 5.0% mask segmentation AP on SSDD and by 2.9% box detection AP and 3.7% mask segmentation AP on HRSID. Especially, the small ship mask segmentation AP of the proposed SA-DWA-Net is higher than the existing best model by 4.4% on SSDD and 3.7% on HRSID.

Compared with the image captured in the natural scene, the image obtained by unmanned aerial vehicle (UAV) aerial photography has a more complex background and many dense small targets, which puts forward higher requirements for the detection accuracy of the target detection algorithm. However, because the UAV is a kind of small mobile device, how to ensure its real-time detection effect has been a problem. Aiming at these problems, the lightweight YOLOv7 algorithm, namely LRT-YOLOv7, is designed. First, the enhance feature fusion module and the transformer efficient layer aggregation networks module are proposed to improve the performance of feature extraction and fusion to enhance the efficiency of small target detection. Second, aiming at the problems of small target size and complex background in the UAV images, the detection head structure is redesigned in the YOLOv7-tiny algorithm to enhance the multi-scale feature fusion ability of the algorithm and thereby improve the algorithm’s detection accuracy for small targets. Finally, ablation, comparison, and visualization validation experiments were conducted using precision, recall, mean average precision, and frames per second (FPS) as evaluation indicators. The results show that the detection speed of the LRT-YOLOv7 algorithm on the self-made traffic target dataset is 133.8 FPS, and the precision indicator is 84.58%. Therefore, the LRT-YOLOv7 algorithm has high accuracy and real-time performance in traffic target detection tasks for UAV aerial imagery.

In order to solve the problems of redundant data acquisition and sparse ground points in dense vegetation areas by conventional unmanned aerial vehicle (UAV) path planning methods, an UAV-airborne LiDAR route optimization method for dense vegetation areas is proposed. First, based on the high-resolution true color remote sensing images of the study area, the “fuzzy” calculation of vegetation coverage for route planning is completed. Then, an optimized ant colony algorithm is proposed for route planning, which introduces vegetation coverage as a reference for route planning and optimizes the pheromone initialization, state transfer rules, pheromone calculation, and update strategies in the classical ant colony algorithm to obtain more ground points. The experimental results show that this method can take into account the vegetation coverage of the flight area and find the area with low vegetation coverage to complete the route planning and efficiently use the sweeping principle of three-dimensional laser scanning to improve the probability of ground point acquisition, with faster iteration speed than the classical ant colony algorithm, and improve the efficiency of ground point acquisition in dense vegetation areas.

The road segmentation task has become increasingly important in fields such as urban planning, traffic management, and environmental monitoring. However, most existing deep learning-based methods suffer from issues such as poor temporal effectiveness and connectivity, making it a significant challenge to achieve high-precision and high-efficiency road segmentation. We propose a road segmentation model based on a detail-enhanced lightweight transformer. Through the connectivity enhancement module, the issue of spatial information loss is addressed, enhancing the modeling capability of the road network connectivity. The model incorporates a detail-enhancement strategy to capture the relationship between roads and the environment, enhancing the perception and expression of details while maintaining low computational complexity. Furthermore, the use of a lightweight multiple feature fusion module promotes information fusion from features at different scales while a maintaining lightweight design. Extensive experiments on two publicly available datasets demonstrate that our method achieves the best performance in terms of real-time effectiveness and accuracy.

Remote sensing (RS) scene classification is a highly challenging task because of the unique characteristics of RS scenes, such as high intra-class variability, large inter-class similarity, and various objects with different scales. Attention, interpreted as an important mechanism of the human visual system, can emphasize meaningful features of deep neural networks, which is beneficial for boosting the classification performance. Motivated by it, we present a multi-attention aggregation network (MAANet), which contains various specially designed attention models, for precise RS scene classification. First, a gated attention fluid coding structure is constructed for mining hierarchical gated attention features from RS images. Second, a progressive pyramid refinement architecture is designed to explore correlations of cross-layer attention features to learn enhanced multi-scale representations. Third, a two-stream attention aggregation structure, equipped with three different attention models, is developed to guide the generation of aggregated features. Finally, a scene label prediction module is proposed for scene label prediction. We conduct extensive experiments on three famous RS scene datasets, and the experimental results show that our MAANet outperforms a number of current representative state-of-the-art approaches for the RS scene classification task.

Efficient compression is pertinent for the convenient storage, transmission, and processing of modern high-resolution hyperspectral images (HSI). We propose a high-performance HSI compression method using library-based spectral unmixing and tensor decomposition. Unlike the existing approaches, our proposed work incorporates unmixing in the compression framework and achieves significantly higher compression performance with negligible loss. The proposed library-based unmixing method includes an index for accurate endmember number estimation, followed by exact library pruning and a sparsity regularized formulation with norm-smoothing to compute the abundance maps. As the spectral library is available at the reconstruction (decoder) side also; compressing the abundance maps is as good as compressing the original HSI data. Since the abundance constraints used for the unmixing indicate the correlation of the abundance maps, compressing all abundance maps seems to cause redundant computation. A metric using the image smoothness and information measures is used here to identify the abundance map hardest to compress and the remaining part is left uncompressed. Subsequently, the work compresses the abundance map tensor using orthogonal PARAFAC decomposition with optimal rank determination. The orthogonalization process ensures that the factors span independent subspaces and reduces redundancy, whereas the rank selection prevents noisy or insignificant components. Extensive experiments are carried out to demonstrate that the unmixing workflow leads to negligible loss due to accurate endmember number estimation, exact library pruning, and accurate physically meaningful sparse inversion. Comparative assessments of compression efficacy suggest that the proposed work corresponds to better compression performance and higher classification accuracy.

Pansharpening is proposed to fuse a high spatial resolution panchromatic (PAN) image and a low spatial resolution multispectral image to generate a high resolution multispectral (HRMS) image with both high spatial resolution and high spectral resolution. Many previous studies have focused only on the global or local relationship between PAN and the corresponding HRMS images in the intensity or gradient domains. However, we observe that the relationship between PAN and HRMS images can be better explored in the high-pass domain through adaptive coefficients. We propose high-pass injection fidelity (HPIF) with adaptive local dual-scale coefficient (LDSC) estimation, which can adequately model the complex relationship between PAN and HRMS images in the high-pass domain and efficiently preserve spatial details. In addition, we propose a new spectral correction term to assist HPIF in avoiding spectral distortion. Specifically, we first compute corresponding LDSC from every input, and then the LDSC assists HPIF to extract spatial and spectral information. Finally, we add a total variation term to assist our proposed HPIF and spectral correction terms, which together make the final pansharpening model. We optimize our model by an alternating direction method of multipliers-based algorithm. Through comparative experiments with existing state-of-the-art pansharpening methods on QuickBird, GaoFen, and WorldView, we demonstrate the superiority of our proposed method in terms of both quantitative metrics and subjective visual effects.

Frequency diversity array (FDA) radar can automatically scan an area of interest without phase shifters by utilizing its range-angle-dependent beampattern, which is more convenient in terms of system implementation relative to the conventional phased array (PA) radar. However, the FDA cannot track a target continuously as the PA does because of the periodicity characteristic shown in the FDA’s beampattern. We address the problem of stable tracking of multiple moving targets, aiming at pursuing a high-resolution target imaging approach. First, we establish an inverse synthetic aperture radar (ISAR) imaging model applicable to multiple repeated subpulses based FDA ISAR (MRS-FDA-ISAR) radar. In the procedure of moving target imaging, the proposed MRS-FDA-ISAR scheme is able to not only avoid the problem of range-angle-coupling but also achieve high-level energy accumulation by compensating the phase of the target. Finally, the back projection algorithm is utilized to achieve high-resolution two-dimensional imaging of multiple moving targets. Numerical experiments demonstrated the effectiveness of the proposed approach and it is shown that this approach is superior to conventional ISAR imaging methods due to its high-level energy utilization and relatively low hardware overhead.

Multispectral remote sensing can enable myriad applications, including land cover change mapping and monitoring of vegetation activity and forest fires, to name a few. Although the acquisition of multispectral remote sensing data could empower developing countries toward better management of their natural resources, acquiring this type of data is not trivial. To assess a potential solution to lowering the barrier to entry for these types of missions, Quetzal-1, a 1U CubeSat and Guatemala’s first satellite, tested a payload prototype on a relatively lower budget. This prototype’s approach was based on housing a series of bandpass filters on a carousel that was rotated by a piezoelectric motor, so that a sensor could acquire data at different wavelengths. We provide a guideline for replicating this prototype (including engineering drawings) and the logic behind its design, describing how the data collection wavelengths were selected and introducing key optics and photography concepts needed for the optimal setup of the sensor, lens, and filters. The governing software architecture and preflight testing (including computer-based simulation and vibration testing) is also described. On-orbit performance parameters, such as power consumption, are provided and contrasted to preflight test results, as well as the impact of payload operations on the satellite’s attitude. Quetzal-1’s piezoelectric motor rotated more than 1800 times in space, and the payload imaged hurricane Iota as it was hitting Guatemala. Although the development of this payload is still work in progress, the lessons learned from the development and operation in space of this first prototype can serve other teams designing their own multispectral imaging CubeSat missions, which by changing the light filters may implement different remote sensing applications.

Urban buildings significantly influence land surface temperature (LST) by modifying heat exchanges due to their 3D structures. Our study calculates six three-dimensional building form (3DBF) indicators in grid cells. They include average height (AH), average volume (AV), building structure index (BSI), floor area ratio (FAR), fluctuation degree (FD), and spatial congestion degree (SCD). Two spatial regression models: spatial lag model (SLM) and spatial error model (SEM), along with three machine learning regression models: support vector regression (SVR), random forest (RF) regression, and artificial neural network (ANN) regression were used in our study. We investigated the qualitative and quantitative effects of urban building 3DBF on Beijing’s primary urban region’s thermal environment. The seasonal relationship between 3DBF indicators and LST was also examined. The ideal grid scale for investigating the impact of 3DBF on LST in Beijing is 200 m. Both SEM and RF regression models exhibit superior performance. The SLM model exhibits superior performance compared to the ANN and SVR models, which demonstrate comparatively poorer results. An increase in AH and AV often results in a decrease in LST, but a higher SCD is associated with an increase in LST. Seasonal variability affects the relationship between BSI and FD on LST. The relationship between FAR and LST does not exhibit a uniform and predictable pattern. In the SEM, the variables of SCD, FD, and AV exhibit a stronger impact on LST. The BSI, AV, and SCD variables are crucial factors in the building of the RF regression model. The cooling effects of AV and AH exhibit their maximum intensity during the summer season, with values reaching up to 1.1°C and 0.5°C, respectively. In winter season, the cooling impact of AH and the warming impact of SCD are notably evident, reaching maximum values of 0.8°C and 0.6°C, respectively.

In the field of remote sensing, hyperspectral image (HSI) classification is a widely used technique. Recently, there has been an increasing focus on utilizing superpixels for HSI classification. However, noise pixels in superpixels may lead to unsatisfactory classification results. To address this issue, an effective superpixel sparse representation classification method with multiple features and L₀ smoothing is proposed. In this method, multifeature extraction utilizes the diversity of HSIs’ spectral–spatial information, band fusion effectively reduces redundant information and noise of HSIs, and L₀ smoothing improves superpixel segmentation results by strengthening homogeneous neighborhoods and edges. Meanwhile, simple linear iterative clustering is adopted to acquire superpixels of HSIs. Finally, the majority voting strategy is adopted to determine the final classification result, improving the classification accuracy. To verify the performance of the proposed method, three hyperspectral datasets are selected for experiments. The experimental results show that the proposed method is superior to some famous classification methods.

We provide a speckle aware image segmentation algorithm for synthetic aperture radar (SAR) data. It uses search based segmentation using a three-component machine learning model where speckle noise is considered as discrete component of the feature description. This method allows for the removal of the need for a de-speckling filter during the feature extraction process for SAR images, resulting in a more efficient and accurate approach. A three-component model is used to efficiently represent a feature in SAR data. The algorithm is used to segment different crops from Sentinel-1 C-band SAR data. We describe the search-based segmentation algorithm, three-component model, and its design using K-NN algorithm. We tested the proposed algorithm against K-NN based segmentation on Sentinel-1 images de-speckled using widely used Lee, Refine Lee, Frost, and Gamma-MAP filters. The proposed method is found to produce better classification accuracy compared to results from K-NN and commonly used de-speckling filters.