IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing最新文献

{"title":"Robust Representation Learning Based on Deep Mutual Information for Scene Classification Against Adversarial Perturbations","authors":"Linjuan Li;Gang Xie;Haoxue Zhang;Xinlin Xie;Heng Li","doi":"10.1109/JSTARS.2025.3564376","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3564376","url":null,"abstract":"Remote sensing scene classification enables data-driven decisions for various applications, such as environmental monitoring, urban planning, and disaster management. However, deep learning models used for scene classification are highly vulnerable to adversarial samples, resulting in incorrect predictions and posing significant risks. While most current methods focus on improving adversarial robustness, they face a trade-off that compromises accuracy on clean, unperturbed images. To address this challenge, we utilized information theory by incorporating a mutual information (MI) representation module, which allows the model to capture high-quality, robust features. Furthermore, a domain adversarial training strategy is applied to promote the learning of domain-invariant features, reducing the effect of distribution differences between clean images and adversarial samples. We propose a novel algorithm that accurately differentiates between clean and adversarial scenes by introducing the MI and domain adaptation-guided network. Extensive experiments demonstrate the effectiveness of our approach against adversarial attacks, revealing a positive correlation between adversarial perturbations and image information entropy, and a negative correlation with robust accuracy.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"11963-11978"},"PeriodicalIF":4.7,"publicationDate":"2025-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10977989","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144084788","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"CASSNet: Cross-Attention Enhanced Spectral–Spatial Interaction Network for Hyperspectral Image Super-Resolution","authors":"Zhanxu Zhang;Linzi Yang;Guanglian Zhang;Jiangwei Deng;Lifeng Bian;Chen Yang","doi":"10.1109/JSTARS.2025.3564379","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3564379","url":null,"abstract":"Deep-learning-based super-resolution (SR) methods for a single hyperspectral image have made significant progress in recent years and become an important research direction in remote sensing. Existing methods perform well in extracting spatial features, but challenges remain in integrating spectral and spatial features when modeling global relationships. In order to take full advantage of the higher spectral resolution of hyperspectral images, this article proposes a novel hyperspectral image SR method (CASSNet), which integrates convolutional neural networks and cross-attention mechanisms into a unified framework. This approach achieves comprehensive integration of spectral and spatial information, with extensive exploration at both local and global levels. In the local feature extraction stage, parallel 3-D/2-D convolutions work in tandem to efficiently capture detail information from both spectral and spatial dimensions. In addition, a spectral–spatial dual-branch module employing the cross-attention mechanism is designed to capture the global dependencies within the features, where the reconstructed spectral–spatial module and the spectral–spatial interaction unit can effectively promote the interaction and complementarity of spectral–spatial features. The experiments on three publicly available datasets demonstrated that the proposed method obtained superior SR results, outperforming state-of-the-art SR algorithms.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"11716-11730"},"PeriodicalIF":4.7,"publicationDate":"2025-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10979241","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144073295","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"GF-2 Remote Sensing-Based Winter Wheat Extraction With Multitask Learning Vision Transformer","authors":"Zhihao Zhao;Zihan Liu;Heng Luo;Hui Yang;Biao Wang;Yixin Jiang;Yanqi Liu;Yanlan Wu","doi":"10.1109/JSTARS.2025.3564680","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3564680","url":null,"abstract":"Accurate mapping of winter wheat is essential for the advancement of precision agriculture and food security. However, classical semantic segmentation models frequently encounter difficulties in precise edge extraction, omission, and classification due to the presence of dense distributions and intraclass diversity. This study proposes a novel method for the extraction of winter wheat from remote sensing data using the GF-2 satellite. The method incorporates a multitask learning framework-Vision Transformer-based model (namely MCFormer) that combines semantic segmentation and boundary detection. Furthermore, the normalized difference vegetation index (NDVI) and land surface temperature (LST) derived from Landsat 8 images was included to enhance the representation of winter wheat's spectral characteristics. The method is evaluated in comparison to frequently used U-Net-, SegNet-, SegFormer-, and MANet-based winter wheat extraction methods in northern Anhui Province. The results indicate that the MCFormer-based method achieves the intersection over union (IoU), F1 score, recall, precision and overall accuracy (OA) of 0.9790, 0.9893, 0.9953, 0.9835, and 0.9900, respectively, outperforming the U-Net-, SegNet-, SegFormer-, and MANet-based methods. The incorporation of multitask learning with NDVI and LST data has been demonstrated to enhance several key performance metrics, including improvements in the IoU, F1 score, recall, precision, and OA by 5.95%, 3.65%, 3.75%, 2.79%, and 2.24%, respectively. Our proposed approach improves the accuracy of winter wheat extraction from remote sensing images, which has the potential to facilitate precision agriculture and enhance food security.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"12454-12469"},"PeriodicalIF":4.7,"publicationDate":"2025-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10979367","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144125360","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Hybrid Products for Enhanced and Unified Full and Compact Polarimetric SAR Data Exploitations","authors":"Ramin Sabry","doi":"10.1109/JSTARS.2025.3564491","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3564491","url":null,"abstract":"A synthetic aperture radar (SAR) data analytic methodology that serves a variety of exploitation objectives is developed within a unified polarimetric framework. New hybrid and multiaspect exploitation products are introduced and examined by applying multimodal and multisensor SAR data. These products include hybrid feature-based classification and detection maps. Hybrid multichannel ML-based, and phenomenology-based change detection and classification products are also developed and examined. Results indicate enhanced exploitation via added-value products for multiaspect and customized detection and change analysis. It is shown that information-rich exploitation products are achieved by the virtue of properly fusing SAR phenomenology-based characteristics. The results also highlight the multimodal and multipurpose nature of hybrid products that offer possibility of multisensor joint data exploitation for enhanced analysis. In particular, means for SAR-based multimodal anomalous analysis for change detection can be provided.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"12059-12073"},"PeriodicalIF":4.7,"publicationDate":"2025-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10979214","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144131621","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Projecting Forest Fire Probability in South Korea Under Climate Change, Population, and Forest Management Scenarios Using AI & Process-Based Hybrid Model (FLAM-Net)","authors":"Hyun-Woo Jo;Myoungsoo Won;Florian Kraxner;Seong Woo Jeon;Yowhan Son;Andrey Krasovskiy;Woo-Kyun Lee","doi":"10.1109/JSTARS.2025.3564852","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3564852","url":null,"abstract":"Climate change-induced heat waves and densely forested areas near urban centers in South Korea create complex challenges for wildfire response systems. Various forest fire models have been developed to address this, each with unique strengths and weaknesses. Process-based models offer high interpretability through human domain knowledge but require extensive optimization, while machine learning models automatically identify important features but have limited interpretability. To leverage the strengths of both models, this study aimed to integrate human domain knowledge into a machine learning framework. IIASA's wildfire cLimate impacts and Adaptation Model (FLAM)—a process-based model incorporating biophysical and human impacts—was developed as a neural network called FLAM-Net. Enhancements included improving backpropagation for optimization and introducing algorithms for national-specific fire ignition dynamics. FLAM-Net was applied at multiple scales and integrated through U-Net-based architecture, named deep neural FLAM (DN-FLAM), to produce downscaled predictions. The optimization revealed spatial concentration of fires near metropolitan areas and the east coast, with temporal concentration in spring due to agricultural burning. Integration of multiscale features through DN-FLAM achieved optimal performance with Pearson's <italic>r</i> values of 0.943, 0.840, and 0.641 for temporal, spatial, and spatio-temporal validations. Future projections based on shared socioeconomic pathways indicated increasing fire frequencies until 2050, followed by a decrease due to increased precipitation. This study demonstrates the benefits of the hybrid approach, providing interpretability, accuracy, and efficient optimization. These hybrid models offer scientific evidence to guide locally tailored decision-making for climate change-induced forest fires and lay the groundwork for global application through their optimization capabilities.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"13003-13022"},"PeriodicalIF":4.7,"publicationDate":"2025-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10979267","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144178905","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"New Advance in Land-Controlled-Source Audio-Frequency Magnetotellurics Exploration: Measurement at a Shortened Transmitter–Receiver Offset and Three-Dimensional Inversion","authors":"Shengtao Wang;Changhong Lin","doi":"10.1109/JSTARS.2025.3564862","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3564862","url":null,"abstract":"Ambient noise significantly affects the quality of land-controlled-source audio-frequency magnetotellurics (CSAMT) data. Shortening the transmitter–receiver offset can enhance the raw data signal-to-noise ratios (S/N). However, most of CSAMT explorations still collect field data using a large separation. To enhance the S/N of the raw electric and magnetic field and the corresponding Cagniard apparent resistivity, we propose shortening offset to 2–4 km during CSAMT fieldwork and inverting apparent resistivity rather than only electric field. We tested this approach with three experimental models with varying offsets, simulating 3-D responses. Results showed that longer offsets predominantly produce data anomalies in the far-field zone, whereas shorter offsets shift these anomalies toward the transition-field zone with some in the far-field or near-field zones. Shortened-offset responses exhibit stronger electric and magnetic fields. Synthetic datasets with different noise levels were generated for inversion analysis. Results indicate that shortening the offset, particularly in noisy environments, improves data S/N. Inversion results reveal that a shortened CSAMT transmitter–receiver offset allows for data collection with most anomalies in the transition-field zone, facilitating the accurate prediction of the true model in the 3-D inversion. In low-noise environments, inversion results from data with a shortened offset are comparable to those with a long offset. However, in high-noise environments, the shortened-offset approach yields more accurate results due to improved data S/N. Field data inversion from Yanqing, China, further confirms the effectiveness of our scheme. The shortened-offset approach achieves higher S/N datasets and inversion results that align with actual geological structures.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"12224-12240"},"PeriodicalIF":4.7,"publicationDate":"2025-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10979203","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144125642","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Multichannel Aligned Feature Fusion Method for Salient Object Detection in Optical Remote Sensing Images","authors":"Weining Zhai;Liejun Wang;Panpan Zheng;Lele Li","doi":"10.1109/JSTARS.2025.3563591","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3563591","url":null,"abstract":"Salient Object Detection (SOD), an important preprocessing part of image processing, identifies and labels the most attention-grabbing objects by simulating human vision. Because remote sensing images (RSIs) have different characteristics from natural images such as the limitation of shooting angle and variable scale. RSI-SOD often faces problems such as incomplete structure, missing semantic information, and blurred edges. Our Multilevel Complementary Cooperative Network (MCoCoNet) is capable of balancing semantic and detailed information to reduce noise interference to ensure semantic integrity through feature fusion in a multi-channel aligned manner. And it is adapted to the network requirements for more targeted feature extraction. Specifically, the Neighbourhood Feature Co-Extractor (NFCoE) is designed between the encoder and the decoder to utilize features from neighbouring layers to complement the missing semantic information as well as the detail information within the current layer, thus ensuring the integrity of the structure. The Parallel Refinement Block (PRB), as a decoder, which is combined with contextual information to gradually refine the target edges. It is shown by extensive experiments and visualisations that MCoCoNet provides new improvement ideas for existing RSI-SOD models.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"12740-12754"},"PeriodicalIF":4.7,"publicationDate":"2025-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10979684","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144139970","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Dual-Path Interactive U-Net for Unsupervised Hyperspectral Image Super-Resolution","authors":"Wenchen Deng;Jianjun Liu;Jinlong Yang;Zebin Wu;Liang Xiao","doi":"10.1109/JSTARS.2025.3564589","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3564589","url":null,"abstract":"Combining low-spatial-resolution hyperspectral image (LrHSI) with high-spatial-resolution multispectral image (HrMSI) serves as an effective strategy for enhancing the spatial fidelity of LrHSI. Nevertheless, most existing methods still face challenges in effectively leveraging the complementary information between the two distinct modalities and maintaining internal consistency, leading to suboptimal fusion results. Previous studies have demonstrated that U-shaped networks can capture spatial structural features within images. Inspired by this, we propose a dual-path interactive U-Net architecture to preserve spatial and spectral integrity. Specifically, we use a standard U-Net and a reversed U-Net as the backbone to extract image information and generate abundance maps of the input images. By enabling interaction between the encoders and decoders of both U-Nets, our architecture integrates information across different scales and modes, leading to enhanced fusion results. To further improve the feature extraction capability, we construct a multimode decomposition and reconstruction module, which adaptively fuses the features of LrHSI and HrMSI. This module extracts and combines correlations between the images through canonical polyadic decomposition and attention mechanism, capturing global features across different modes. In addition, we design a weight-sharing U-Net that leverages the similarities and differences between two abundance maps, ensuring internal consistency while reducing computational cost. Thorough evaluations conducted using four publicly available datasets, along with one real-world dataset, and under various noise conditions confirm the validity of our proposed model.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"11751-11766"},"PeriodicalIF":4.7,"publicationDate":"2025-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10978013","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144073126","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Arctic Sea Ice and Open Water Classification From Dual-Polarization Synthetic Aperture Radar Imagery and Deep Learning Models","authors":"Yiru Lu;Biao Zhang;William Perrie;Jinyu Sheng","doi":"10.1109/JSTARS.2025.3564847","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3564847","url":null,"abstract":"In this study, we present a deep learning (DL)-based method for classifying Arctic sea ice and open water using spaceborne C-band RADARSAT-2 dual-polarization (HH, HV) synthetic aperture radar (SAR) images. The HH- and HV-polarization radar. backscatter and cross-polarization ratios (HH/HV) are used as input for the DL model. First, we combine an unsupervised clustering method with an adaptive thresholding technique to generate accurately labeled samples, thereby minimizing subjective errors and reducing the time required for visual interpretation. Second, we employ an enhanced U-Net architecture to develop the proposed classification method. The modified Atrous spatial pyramid pooling module is integrated into a customized dilated U-Net to create a multiscale feature extraction model (MS-DUNet). MS-DUNet is then trained and validated using 11 485 labeled patches extracted from 3565 RADARSAT-2 SAR images. In addition, 5004 SAR images are used to test the model. Compared to representative DL models, MS-DUNet demonstrates better performance, achieving an overall classification accuracy of 99.3% and an intersection over union value of 98.6%. Furthermore, we compare MS-DUNet's predictions with the daily sea ice extent data from the Interactive Multisensor Snow and ice mapping system, achieving an average accuracy of 91.9%. The results indicate that the proposed approach effectively distinguishes between sea ice and open water, using wide-swath SAR imagery. The method also demonstrates strong performance in the complex marginal ice zone during the melting season, and in distinguishing sea ice leads from the surrounding areas.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"11803-11815"},"PeriodicalIF":4.7,"publicationDate":"2025-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10979201","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144073121","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"FRANet: A Feature Refinement Attention Network for SAR Image Denoising","authors":"Shuaiqi Liu;Yu Lei;Qi Hu;Ming Liu;Bing Li;Weiming Hu;Yu-Dong Zhang","doi":"10.1109/JSTARS.2025.3564846","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3564846","url":null,"abstract":"Since synthetic aperture radar (SAR) images have complex noise and have no clean reference images, SAR image denoising is very challenging. With the development of deep learning, several denoising algorithms based on deep learning are proposed to achieve a better SAR image denoising effect. However, most networks are prone to gradient disappearance and explosion in the training process. The deep network model will produce an excessive amount of computation. The denoising time is also too long. Since most of the denoising algorithms based on deep learning use simulated images for model training, it is difficult to effectively suppress speckle noise in the real SAR image while a balance between denoising and detail preservation cannot be achieved. To address the mentioned problems, we propose a novel feature refinement attention network named FRANet. In FRANet, a feature refinement network is first used to refine the input noise image to extract more useful features while accelerating network training. Second, a feature attention encoder–decoder network is constructed for deep feature extraction. This network uses an asymmetric encoder–decoder structure to expand the receptive field, which can improve the information extraction ability and reduce the number of parameters effectively. Finally, the final denoised SAR image is obtained by global residual learning. Compared with other denoising algorithms, the proposed algorithm can achieve better results in denoising performance and running time.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"18 ","pages":"12343-12363"},"PeriodicalIF":4.7,"publicationDate":"2025-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10979213","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144125487","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}