Pub Date : 2025-12-11DOI: 10.1109/JSTARS.2025.3642993
Xianqi Meng;Yuefeng Zhao;Kaifa Cao;Qifei Wang;Junjie Wang;Nannan Hu
Remote sensing image change captioning (RSICC) aims to generate textual descriptions of changes between bitemporal images. However, accurately describing fine-grained changes while capturing interscale relationships as well as distinguishing real changes from spurious changes (e.g., illumination, seasonal variations) are still major challenges for current methods. To address these issues, we propose DualFocus-CapNet, a novel model tailored for RSICC. DualFocus-CapNet employs a dual-stream architecture, where each stream is dedicated to processing a distinct pair of bitemporal features. Crucially, we introduce a scale-wise progressive convolution (ScalePro Conv) that employs a progressive scale-specific approach to decompose remote sensing features into pixel-level variations, regional continuities, and linear structures. Unlike conventional parallel multiscale processing methods, ScalePro Conv adopts a serial progressive structure to establish interscale relationships, thereby avoiding the fragmentation of feature information. Then, the bi-directional difference guided transformer (BDiGTrans) is proposed to eliminate interference from spurious changes by dynamically masking invariant regions and extracting bidirectional differential features. Furthermore, the cross-temporal adaptive fusion module (CTAF) is introduced to dynamically balance bitemporal features using learnable gating to enhance change discrimination and robust caption generation. Comprehensive experiments on the benchmark datasets LEVIR-CC and WHU-CDC show that our DualFocus-CapNet surpasses state-of-the-art change captioning methods in various evaluation metrics.
{"title":"DualFocus-CapNet: A Dual-Stream Network for Real Change and Interscale Relationship-Aware Change Captioning in Remote Sensing Images","authors":"Xianqi Meng;Yuefeng Zhao;Kaifa Cao;Qifei Wang;Junjie Wang;Nannan Hu","doi":"10.1109/JSTARS.2025.3642993","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3642993","url":null,"abstract":"Remote sensing image change captioning (RSICC) aims to generate textual descriptions of changes between bitemporal images. However, accurately describing fine-grained changes while capturing interscale relationships as well as distinguishing real changes from spurious changes (e.g., illumination, seasonal variations) are still major challenges for current methods. To address these issues, we propose DualFocus-CapNet, a novel model tailored for RSICC. DualFocus-CapNet employs a dual-stream architecture, where each stream is dedicated to processing a distinct pair of bitemporal features. Crucially, we introduce a scale-wise progressive convolution (ScalePro Conv) that employs a progressive scale-specific approach to decompose remote sensing features into pixel-level variations, regional continuities, and linear structures. Unlike conventional parallel multiscale processing methods, ScalePro Conv adopts a serial progressive structure to establish interscale relationships, thereby avoiding the fragmentation of feature information. Then, the bi-directional difference guided transformer (BDiGTrans) is proposed to eliminate interference from spurious changes by dynamically masking invariant regions and extracting bidirectional differential features. Furthermore, the cross-temporal adaptive fusion module (CTAF) is introduced to dynamically balance bitemporal features using learnable gating to enhance change discrimination and robust caption generation. Comprehensive experiments on the benchmark datasets LEVIR-CC and WHU-CDC show that our DualFocus-CapNet surpasses state-of-the-art change captioning methods in various evaluation metrics.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"19 ","pages":"2045-2059"},"PeriodicalIF":5.3,"publicationDate":"2025-12-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11297771","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145886568","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}
Pub Date : 2025-12-10DOI: 10.1109/JSTARS.2025.3642571
Rong Deng;Yunwei Pu;Mingda Hao
The accuracy of change detection (CD) in high-resolution remote sensing imagery can be compromised due to complex surface textures, variations in illumination, and atmospheric disturbances. Existing methods struggle to simultaneously achieve noise suppression, long-range dependence modeling, and computational efficiency. In response to the aforementioned challenges, this study proposes a spatial–frequency collaborative detection network (SFCNet) that combines VMamba’s linear-complexity global modeling with frequency-domain techniques. The designed network adopts a VMamba-based encoder to efficiently extract multiscale features and further integrates a frequency–spatial interaction module to perform content-aware noise suppression and detail enhancement through spatial-domain-guided adaptive frequency filtering. A progressive decoder, equipped with a hybrid pooling channel attention module, integrates multilevel features through hybrid pooling and progressive fusion strategies. To validate the effectiveness of the proposed SFCNet, extensive experiments were conducted on three widely used public datasets: LEVIR-CD, WHU-CD, and CLCD. The network achieved F1-scores of 91.17%, 94.29%, and 74.98%, along with intersection over union values of 83.78%, 89.21%, and 59.98%, respectively. These results clearly demonstrate that the proposed method enhances detection accuracy and robustness under complex scenarios.
{"title":"Spatial–Frequency Collaborative Network for Remote Sensing Image Change Detection","authors":"Rong Deng;Yunwei Pu;Mingda Hao","doi":"10.1109/JSTARS.2025.3642571","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3642571","url":null,"abstract":"The accuracy of change detection (CD) in high-resolution remote sensing imagery can be compromised due to complex surface textures, variations in illumination, and atmospheric disturbances. Existing methods struggle to simultaneously achieve noise suppression, long-range dependence modeling, and computational efficiency. In response to the aforementioned challenges, this study proposes a spatial–frequency collaborative detection network (SFCNet) that combines VMamba’s linear-complexity global modeling with frequency-domain techniques. The designed network adopts a VMamba-based encoder to efficiently extract multiscale features and further integrates a frequency–spatial interaction module to perform content-aware noise suppression and detail enhancement through spatial-domain-guided adaptive frequency filtering. A progressive decoder, equipped with a hybrid pooling channel attention module, integrates multilevel features through hybrid pooling and progressive fusion strategies. To validate the effectiveness of the proposed SFCNet, extensive experiments were conducted on three widely used public datasets: LEVIR-CD, WHU-CD, and CLCD. The network achieved F1-scores of 91.17%, 94.29%, and 74.98%, along with intersection over union values of 83.78%, 89.21%, and 59.98%, respectively. These results clearly demonstrate that the proposed method enhances detection accuracy and robustness under complex scenarios.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"19 ","pages":"2485-2496"},"PeriodicalIF":5.3,"publicationDate":"2025-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11293753","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145929347","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}
Pub Date : 2025-12-09DOI: 10.1109/JSTARS.2025.3641788
Hao Yu;Gen Li;Haoyu Liu;Songyan Zhu;Jian Xu;Wenquan Dong;Changjian Li;Jiancheng Shi
Accurate, high-resolution global land use and land cover (LULC) mapping is crucial for environmental monitoring, but remains challenging when relying solely on multispectral data. Most existing global LULC mapping studies rely exclusively on multispectral observations, and even those that incorporate synthetic aperture radar (SAR) data often fail to fully exploit the information it provides. SAR provides an all-weather sensing capability and is uniquely sensitive to surface structure, texture, and moisture—critical information for LULC classes that are often spectrally ambiguous. To address this data gap, we introduce the Dynamic World+ dataset, a new global benchmark that expands the authoritative Dynamic World by aligning it with Sentinel-1 SAR data. In addition, to facilitate the combination of multispectral and SAR data, we propose a lightweight transformer architecture termed LULC-Former. It incorporates two innovative modules, the dual-modal enhancement module and mutual modal aggregation module, designed to exploit cross-information between the two modalities in a split-fusion manner. These modules enable spectral features to guide the interpretation of SAR texture, and vice versa, thereby improving the overall performance of global LULC semantic segmentation. Furthermore, we adopt an imbalanced parameter allocation strategy, which assigns parameters to different modalities based on the distinct physical information each provides for LULC characterization. Experiments demonstrate that our network outperforms existing transformer and CNN-based models, achieving a mean intersection over union of 59.58%, an overall accuracy of 79.48%, and an F1 score of 71.68% with only 26.70 M parameters. Furthermore, the generated national-scale LULC maps across diverse regions demonstrate the effectiveness of the proposed dataset and network.
{"title":"Synergistic Fusion of Sentinel-1 and Sentinel-2 for Global LULC Mapping: The Multimodal Network LULC-Former and Dynamic World+ Dataset","authors":"Hao Yu;Gen Li;Haoyu Liu;Songyan Zhu;Jian Xu;Wenquan Dong;Changjian Li;Jiancheng Shi","doi":"10.1109/JSTARS.2025.3641788","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3641788","url":null,"abstract":"Accurate, high-resolution global land use and land cover (LULC) mapping is crucial for environmental monitoring, but remains challenging when relying solely on multispectral data. Most existing global LULC mapping studies rely exclusively on multispectral observations, and even those that incorporate synthetic aperture radar (SAR) data often fail to fully exploit the information it provides. SAR provides an all-weather sensing capability and is uniquely sensitive to surface structure, texture, and moisture—critical information for LULC classes that are often spectrally ambiguous. To address this data gap, we introduce the <italic>Dynamic World+</i> dataset, a new global benchmark that expands the authoritative Dynamic World by aligning it with Sentinel-1 SAR data. In addition, to facilitate the combination of multispectral and SAR data, we propose a lightweight transformer architecture termed <italic>LULC-Former</i>. It incorporates two innovative modules, the dual-modal enhancement module and mutual modal aggregation module, designed to exploit cross-information between the two modalities in a split-fusion manner. These modules enable spectral features to guide the interpretation of SAR texture, and vice versa, thereby improving the overall performance of global LULC semantic segmentation. Furthermore, we adopt an imbalanced parameter allocation strategy, which assigns parameters to different modalities based on the distinct physical information each provides for LULC characterization. Experiments demonstrate that our network outperforms existing transformer and CNN-based models, achieving a mean intersection over union of 59.58%, an overall accuracy of 79.48%, and an F1 score of 71.68% with only 26.70 M parameters. Furthermore, the generated national-scale LULC maps across diverse regions demonstrate the effectiveness of the proposed dataset and network.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"19 ","pages":"2511-2524"},"PeriodicalIF":5.3,"publicationDate":"2025-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11286222","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145929431","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}
Pub Date : 2025-12-09DOI: 10.1109/JSTARS.2025.3642040
Masoud Mahdianpari;Ali Radman;Daniel J. Varon;Fariba Mohammadimanesh
Recent advances in foundation models, including large language models and advanced computer vision techniques, have opened new possibilities in remote sensing applications. One such model is the Segment Anything Model (SAM), which can perform image segmentation without task-specific training data. This is especially useful for detecting methane plumes in satellite imagery, where it is important to accurately separate methane column enhancements from complex background conditions. SAM’s prompt-based segmentation approach helps address these challenges and reduces the need for large annotated datasets. In this study, we introduce SAM4CH4, a zero-shot segmentation framework that applies SAM for methane plume detection using Sentinel-2 imagery, with segmentation prompts automatically generated by text encoder models including contrastive language-image pretraining (CLIP), CLIP Surgery, and Grounding DINO. We evaluate the approach on both a synthetically generated benchmark dataset and real Sentinel-2 images. Results show that bounding-box prompts from the Swin-L variant of Grounding DINO, combined with the latest version of SAM, consistently achieve high accuracy, exceeding 72% in F1-score and 95% in overall accuracy, and outperform a widely used statistical thresholding method by approximately 15% in F1-score. These results are also competitive with supervised deep learning methods, which typically require large labeled datasets and significant computational resources. By leveraging pretrained models and removing the need for manual annotation, the proposed SAM4CH4 framework offers a zero-shot scalable and efficient solution for operational methane plume detection and monitoring.
{"title":"SAM4CH4: Zero-Shot Methane Plume Mapping With Segment Anything and Vision-Language Models","authors":"Masoud Mahdianpari;Ali Radman;Daniel J. Varon;Fariba Mohammadimanesh","doi":"10.1109/JSTARS.2025.3642040","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3642040","url":null,"abstract":"Recent advances in foundation models, including large language models and advanced computer vision techniques, have opened new possibilities in remote sensing applications. One such model is the Segment Anything Model (SAM), which can perform image segmentation without task-specific training data. This is especially useful for detecting methane plumes in satellite imagery, where it is important to accurately separate methane column enhancements from complex background conditions. SAM’s prompt-based segmentation approach helps address these challenges and reduces the need for large annotated datasets. In this study, we introduce SAM4CH4, a zero-shot segmentation framework that applies SAM for methane plume detection using Sentinel-2 imagery, with segmentation prompts automatically generated by text encoder models including contrastive language-image pretraining (CLIP), CLIP Surgery, and Grounding DINO. We evaluate the approach on both a synthetically generated benchmark dataset and real Sentinel-2 images. Results show that bounding-box prompts from the Swin-L variant of Grounding DINO, combined with the latest version of SAM, consistently achieve high accuracy, exceeding 72% in F1-score and 95% in overall accuracy, and outperform a widely used statistical thresholding method by approximately 15% in F1-score. These results are also competitive with supervised deep learning methods, which typically require large labeled datasets and significant computational resources. By leveraging pretrained models and removing the need for manual annotation, the proposed SAM4CH4 framework offers a zero-shot scalable and efficient solution for operational methane plume detection and monitoring.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"19 ","pages":"2273-2284"},"PeriodicalIF":5.3,"publicationDate":"2025-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11286214","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145886537","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}
Pub Date : 2025-12-08DOI: 10.1109/JSTARS.2025.3641630
Yusheng Cui;Tao Xie;Jian Li;Xuehong Zhang;Shuying Bai;Chao Wang;Hui Liu
The inversion of chlorophyll-a (Chla) across water bodies is challenged by the diversity of optical water types (OWTs) and by the nonlinear superposition of optical signals from optically active constituents. To address these challenges, this study introduces an inversion method based on water classification that integrates dual-stream feature extraction (via a one-dimensional convolutional neural network (1-D CNN)-Transformer) and multimodel evaluation, termed OWT-CCINET. We first applied k-means clustering to the reflectance spectra, dividing the samples into four OWTs: eutrophic, turbid, clear, and moderately turbid. For each group, feature extraction was adjusted using either sequential or parallel branches. Several feature combinations were then tested with different regression models, and the most suitable model for each OWT was determined based on a comparative evaluation. Compared with conventional models such as the OCx and CI hybrid algorithms, the model with targeted feature extraction and model selection based on water classification significantly improved the accuracy and applicability for Chla inversion across various OWTs. By applying OWT-CCINET, the natural log (ln)-transformed Chla concentrations were estimated, and the model achieved outstanding results, with a mean absolute error of 0.410, a root mean square error of 0.549, and a coefficient of determination (R2) of 0.899. After minimally harmonizing MODIS and MERIS to SeaWiFS-equivalent bands, OWT-CCINET maintained stable cross-sensor performance, supporting long-term, multimission applications. This study provides novel methodologies and theoretical support for the development of cross-water type models, offering significant value for practical applications such as marine environmental monitoring.
{"title":"A Chlorophyll Concentration Inversion Method Based on OWTs and 1D CNN-Transformer Feature Extraction","authors":"Yusheng Cui;Tao Xie;Jian Li;Xuehong Zhang;Shuying Bai;Chao Wang;Hui Liu","doi":"10.1109/JSTARS.2025.3641630","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3641630","url":null,"abstract":"The inversion of chlorophyll-a (Chla) across water bodies is challenged by the diversity of optical water types (OWTs) and by the nonlinear superposition of optical signals from optically active constituents. To address these challenges, this study introduces an inversion method based on water classification that integrates dual-stream feature extraction (via a one-dimensional convolutional neural network (1-D CNN)-Transformer) and multimodel evaluation, termed OWT-CCINET. We first applied <italic>k</i>-means clustering to the reflectance spectra, dividing the samples into four OWTs: eutrophic, turbid, clear, and moderately turbid. For each group, feature extraction was adjusted using either sequential or parallel branches. Several feature combinations were then tested with different regression models, and the most suitable model for each OWT was determined based on a comparative evaluation. Compared with conventional models such as the OCx and CI hybrid algorithms, the model with targeted feature extraction and model selection based on water classification significantly improved the accuracy and applicability for Chla inversion across various OWTs. By applying OWT-CCINET, the natural log (ln)-transformed Chla concentrations were estimated, and the model achieved outstanding results, with a mean absolute error of 0.410, a root mean square error of 0.549, and a coefficient of determination (<italic>R</i><sup>2</sup>) of 0.899. After minimally harmonizing MODIS and MERIS to SeaWiFS-equivalent bands, OWT-CCINET maintained stable cross-sensor performance, supporting long-term, multimission applications. This study provides novel methodologies and theoretical support for the <italic>development</i> of cross-water type models, offering significant value for practical applications such as marine environmental monitoring.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"19 ","pages":"2006-2032"},"PeriodicalIF":5.3,"publicationDate":"2025-12-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11284877","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145886588","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}
Pub Date : 2025-12-05DOI: 10.1109/JSTARS.2025.3640403
Jun Hu;Hongjun Su;Yiping Chen;Yuanwei Qin;Zhaohui Xue;Qian Du
Water use efficiency (WUE) is a fundamental indicator of the balance between ecosystem carbon assimilation and water consumption. However, its spatial variability and dominant environmental drivers across China's river basins remain unclear, posing challenges for basin-scale management. In this study, a comprehensive WUE analysis framework was established through the integration of multisource remote sensing and auxiliary datasets. In this framework, multisource vegetation, climate, topography, and land-use data were integrated to estimate WUE from the GPP-to-ET ratio, and a novel basin-scale dataset covering 25 major river basins in China from 2002 to 2021 was generated (CBS-WUE, https://doi.org/10.5281/zenodo.17402779), which was validated against FLUXNET2015 observations. With this new dataset, inter-basin comparisons were conducted to characterize spatial heterogeneity and temporal dynamics, while multivariate statistical and machine learning analyses were employed to identify the relative contributions of climatic, biotic, and land-use drivers. Results indicated that elevation and vegetation structure were the primary factors influencing basin-scale WUE differences. The national average WUE was 1.13 g C kg−1 H2O, with basin-level values ranging from 0.11 to 1.80 g C kg−1 H2O. Among them, higher WUE was in basins of moderate elevation and dense vegetation, and lower WUE was in high-elevation or arid basins. This integrative analysis highlights the dominant role of topography and vegetation in shaping WUE patterns and provides a scientific basis for enhancing water resource efficiency and ecological sustainability under changing environmental conditions.
水分利用效率（WUE）是衡量生态系统碳同化与水分消耗平衡的基本指标。然而，中国河流流域的空间变异性和主导环境驱动因素尚不清楚，这给流域尺度管理带来了挑战。本研究通过多源遥感和辅助数据集的整合，建立了一个综合的WUE分析框架。在该框架下，综合多源植被、气候、地形和土地利用数据，从gpp - et比估算WUE，并生成了覆盖中国25个主要流域的2002 - 2021年流域尺度数据集（CBS-WUE, https://doi.org/10.5281/zenodo.17402779），并与FLUXNET2015观测数据进行了验证。利用这一新的数据集，进行了流域间的比较，以表征空间异质性和时间动态，同时采用多元统计和机器学习分析来确定气候、生物和土地利用驱动因素的相对贡献。结果表明，高程和植被结构是影响流域尺度水分利用效率差异的主要因素。全国平均水分利用效率为1.13 g C kg−1 H2O，流域平均水分利用效率为0.11 ~ 1.80 g C kg−1 H2O。其中，中等海拔和植被密集的流域WUE较高，而高海拔和干旱的流域WUE较低。这种综合分析强调了地形和植被在水利用效率模式形成中的主导作用，为在变化的环境条件下提高水资源效率和生态可持续性提供了科学依据。
{"title":"Elevation and Vegetation Cover Dominate Inter-Basin Water Use Efficiency Patterns in China","authors":"Jun Hu;Hongjun Su;Yiping Chen;Yuanwei Qin;Zhaohui Xue;Qian Du","doi":"10.1109/JSTARS.2025.3640403","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3640403","url":null,"abstract":"Water use efficiency (WUE) is a fundamental indicator of the balance between ecosystem carbon assimilation and water consumption. However, its spatial variability and dominant environmental drivers across China's river basins remain unclear, posing challenges for basin-scale management. In this study, a comprehensive WUE analysis framework was established through the integration of multisource remote sensing and auxiliary datasets. In this framework, multisource vegetation, climate, topography, and land-use data were integrated to estimate WUE from the GPP-to-ET ratio, and a novel basin-scale dataset covering 25 major river basins in China from 2002 to 2021 was generated (CBS-WUE, <uri>https://doi.org/10.5281/zenodo.17402779</uri>), which was validated against FLUXNET2015 observations. With this new dataset, inter-basin comparisons were conducted to characterize spatial heterogeneity and temporal dynamics, while multivariate statistical and machine learning analyses were employed to identify the relative contributions of climatic, biotic, and land-use drivers. Results indicated that elevation and vegetation structure were the primary factors influencing basin-scale WUE differences. The national average WUE was 1.13 g C kg<sup>−1</sup> H<sub>2</sub>O, with basin-level values ranging from 0.11 to 1.80 g C kg<sup>−1</sup> H<sub>2</sub>O. Among them, higher WUE was in basins of moderate elevation and dense vegetation, and lower WUE was in high-elevation or arid basins. This integrative analysis highlights the dominant role of topography and vegetation in shaping WUE patterns and provides a scientific basis for enhancing water resource efficiency and ecological sustainability under changing environmental conditions.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"19 ","pages":"1533-1548"},"PeriodicalIF":5.3,"publicationDate":"2025-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11278658","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145778394","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}
Pub Date : 2025-12-04DOI: 10.1109/JSTARS.2025.3639900
Yu Liu;Jing Bai;Zhu Xiao;Yuheng Lian;Licheng Jiao
Absolute visual localization based on onboard cameras and georeferenced data is the dominant solution for unmanned aerial vehicles (UAVs) when the global navigation satellite system is not available. However, current image-based matching methods are easily affected by significant feature differences between UAV images and georeferenced images, and map-based matching methods are affected by image rotation and resolution differences. To address the above-mentioned challenges, this article proposes a rotation and scale invariance map matching method (RSIM) to realize UAV visual localization in urban scenarios. Specifically, RSIM first extracts building contour information from UAV images. Then, based on the shape and spatial relationship features of the buildings, a scene matching method is designed to match the UAV image and the vector e-map. Finally, the centers of the matched building individuals are used as control points for UAV position solving. Extensive experiments conducted on the LoFTR and DKM datasets demonstrate that the proposed RSIM algorithm achieves superior localization performance compared to existing deep learning-based image matching methods. Moreover, our RSIM effectively enables UAV localization in urban scenarios even when the orientation and resolution of UAV images are unknown, achieving performance comparable to the SSRM map matching method, which relies on prior knowledge of these parameters.
{"title":"A Rotation and Scale Invariant Map Matching Method for UAV Visual Geolocalization","authors":"Yu Liu;Jing Bai;Zhu Xiao;Yuheng Lian;Licheng Jiao","doi":"10.1109/JSTARS.2025.3639900","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3639900","url":null,"abstract":"Absolute visual localization based on onboard cameras and georeferenced data is the dominant solution for unmanned aerial vehicles (UAVs) when the global navigation satellite system is not available. However, current image-based matching methods are easily affected by significant feature differences between UAV images and georeferenced images, and map-based matching methods are affected by image rotation and resolution differences. To address the above-mentioned challenges, this article proposes a rotation and scale invariance map matching method (RSIM) to realize UAV visual localization in urban scenarios. Specifically, RSIM first extracts building contour information from UAV images. Then, based on the shape and spatial relationship features of the buildings, a scene matching method is designed to match the UAV image and the vector e-map. Finally, the centers of the matched building individuals are used as control points for UAV position solving. Extensive experiments conducted on the LoFTR and DKM datasets demonstrate that the proposed RSIM algorithm achieves superior localization performance compared to existing deep learning-based image matching methods. Moreover, our RSIM effectively enables UAV localization in urban scenarios even when the orientation and resolution of UAV images are unknown, achieving performance comparable to the SSRM map matching method, which relies on prior knowledge of these parameters.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"19 ","pages":"1616-1627"},"PeriodicalIF":5.3,"publicationDate":"2025-12-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11278038","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145830804","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}
Target detection in Synthetic Aperture Radar (SAR) images is of great importance in civilian monitoring and military reconnaissance. However, the unique speckle noise inherent in SAR images leads to semantic information loss, while traditional convolutional neural network downsampling methods exacerbate this issue, impacting detection accuracy and robustness. Moreover, some dense target scenarios and weak scattering features of targets make it challenging to achieve sufficient feature discriminability, adding complexity to the detection task. In addition, the multiscale characteristic of SAR targets presents difficulties in balancing detection performance with computational efficiency in complex scenes. To tackle these difficulties, this article introduces a wavelet-driven transformer-based SAR target detection framework called TranSTD. Specifically, it incorporates the Haar wavelet dynamic downsampling and semantic preserving dynamic downsampling modules, which effectively suppress noise and preserve semantic information using techniques such as Haar wavelet denoise and input-driven dynamic pooling downsampling. Furthermore, the SAR adaptive convolution (SAC) bottleneck is proposed for enhancing the discrimination of features. To optimize performance and efficiency across varying scene complexities, a multiscale SAR attention fusion encoder is developed. Extensive experiments are carried out on three datasets, showing that our proposed algorithm outperforms the current state-of-the-art benchmarks in SAR target detection, offering a robust solution for the detection of targets in complex SAR scenes.
{"title":"TranSTD: A Wavelet-Driven Transformer-Based SAR Target Detection Framework With Adaptive Feature Enhancement and Fusion","authors":"Bobo Xi;Jiaqi Chen;Yan Huang;Jiaojiao Li;Yunsong Li;Zan Li;Xiang-Gen Xia","doi":"10.1109/JSTARS.2025.3639785","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3639785","url":null,"abstract":"Target detection in Synthetic Aperture Radar (SAR) images is of great importance in civilian monitoring and military reconnaissance. However, the unique speckle noise inherent in SAR images leads to semantic information loss, while traditional convolutional neural network downsampling methods exacerbate this issue, impacting detection accuracy and robustness. Moreover, some dense target scenarios and weak scattering features of targets make it challenging to achieve sufficient feature discriminability, adding complexity to the detection task. In addition, the multiscale characteristic of SAR targets presents difficulties in balancing detection performance with computational efficiency in complex scenes. To tackle these difficulties, this article introduces a wavelet-driven transformer-based SAR target detection framework called TranSTD. Specifically, it incorporates the Haar wavelet dynamic downsampling and semantic preserving dynamic downsampling modules, which effectively suppress noise and preserve semantic information using techniques such as Haar wavelet denoise and input-driven dynamic pooling downsampling. Furthermore, the SAR adaptive convolution (SAC) bottleneck is proposed for enhancing the discrimination of features. To optimize performance and efficiency across varying scene complexities, a multiscale SAR attention fusion encoder is developed. Extensive experiments are carried out on three datasets, showing that our proposed algorithm outperforms the current state-of-the-art benchmarks in SAR target detection, offering a robust solution for the detection of targets in complex SAR scenes.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"19 ","pages":"1197-1211"},"PeriodicalIF":5.3,"publicationDate":"2025-12-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11275702","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145778392","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}
Pub Date : 2025-12-03DOI: 10.1109/JSTARS.2025.3639503
Yong Wan;Liyan Peng;Rui Zhang;Ruyue Zhang;Haowen Wang
As one of the most severe forms of pollution, oil spills pose significant threats to the marine environment. Synthetic aperture radar (SAR), an active microwave remote sensing technology, enables sea surface monitoring under all weather and lighting conditions and provides high spatial resolution. It has been widely used in the field of marine oil spill detection. However, other natural phenomena, such as low wind regions and biogenic oil films, can also produce dark spot features in SAR imagery that resemble oil spills, leading to false alarms. Global navigation satellite system-reflectometry (GNSS-R), as an emerging remote sensing technique for ocean observation, offers distinct advantages, including high temporal resolution and multisource observation capabilities. By combining SAR backscattering coefficients with GNSS-R delay doppler map, it becomes possible to characterize the impact of oil spills on sea surface roughness from both backscattering and forward-scattering perspectives. This joint approach enables more accurate oil spill detection and has the potential to reduce the false alarms. Nevertheless, limited measured data for multisource remote sensing oil spill detection hinders robust multisensor fusion model development. To address this, this study proposes a synchronized data generation method, creating a joint SAR and GNSS-R oil spill dataset, and on this basis, a dual-branch EfficientNetV2-S architecture is adopted to build a multisource satellite oil spill data fusion model, which is applied to offshore oil spill detection. According to experimental results, the suggested model detects oil spills with an accuracy of 94.97%. Compared with SAR-only detection models, the false alarm rate is reduced by 3.6%, demonstrating that the dual-payload approach effectively lowers the rate of false detections in marine oil spill monitoring.
{"title":"A Dual-Branch EfficientNetV2-S-Based Method for Marine Oil Spill Detection Using Multisource Satellite Data Fusion","authors":"Yong Wan;Liyan Peng;Rui Zhang;Ruyue Zhang;Haowen Wang","doi":"10.1109/JSTARS.2025.3639503","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3639503","url":null,"abstract":"As one of the most severe forms of pollution, oil spills pose significant threats to the marine environment. Synthetic aperture radar (SAR), an active microwave remote sensing technology, enables sea surface monitoring under all weather and lighting conditions and provides high spatial resolution. It has been widely used in the field of marine oil spill detection. However, other natural phenomena, such as low wind regions and biogenic oil films, can also produce dark spot features in SAR imagery that resemble oil spills, leading to false alarms. Global navigation satellite system-reflectometry (GNSS-R), as an emerging remote sensing technique for ocean observation, offers distinct advantages, including high temporal resolution and multisource observation capabilities. By combining SAR backscattering coefficients with GNSS-R delay doppler map, it becomes possible to characterize the impact of oil spills on sea surface roughness from both backscattering and forward-scattering perspectives. This joint approach enables more accurate oil spill detection and has the potential to reduce the false alarms. Nevertheless, limited measured data for multisource remote sensing oil spill detection hinders robust multisensor fusion model development. To address this, this study proposes a synchronized data generation method, creating a joint SAR and GNSS-R oil spill dataset, and on this basis, a dual-branch EfficientNetV2-S architecture is adopted to build a multisource satellite oil spill data fusion model, which is applied to offshore oil spill detection. According to experimental results, the suggested model detects oil spills with an accuracy of 94.97%. Compared with SAR-only detection models, the false alarm rate is reduced by 3.6%, demonstrating that the dual-payload approach effectively lowers the rate of false detections in marine oil spill monitoring.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"19 ","pages":"1549-1566"},"PeriodicalIF":5.3,"publicationDate":"2025-12-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11275680","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145778390","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}
Pub Date : 2025-12-03DOI: 10.1109/JSTARS.2025.3639670
Cuiping Shi;Yimin Wang;Liguo Wang
With the rapid development of deep learning technology, significant progress has been made in the field of remote sensing (RS) scene image classification. However, the large intraclass distance and high interclass similarity still pose significant challenges for RS scene classification. In addition, there are multiscale targets in RS images, which make significant differences in target characteristics. To overcome the above limitations, this article proposes a novel attention-fusion focused multiscale architecture network (AF2-MSA Net). First, a multilevel feature extraction module (MFEM) was designed to extract semantic and detail information at different scales from RS images. Subsequently, an intricately designed global context recalibration module (GCRM) was embedded into MFEM, and the features at each level were enhanced through a global context recalibration mechanism, enabling the model to dynamically focus on key semantic regions and important contextual information. Next, an axis-aligned feature harmonization module (AAFHM) was constructed to fuse multiscale features from adjacent stages layer by layer. This module combines attention mechanisms from both channel and spatial branches to adaptively coordinate and fuse multiscale contextual information, achieving deep collaborative optimization of different scale features. Finally, the GCRM and AAFHM are integrated into a unified framework called AF2-MSA Net to achieve collaborative optimization of global semantics and multiscale discriminative features. Extensive experiments on three commonly used datasets have shown that the proposed AF2-MSA Net outperforms some state-of-the-art methods in RS image scene classification tasks.
{"title":"AF2-MSA Net: Attention-Fusion Focused Multiscale Architecture Network for Remote Sensing Scene Classification","authors":"Cuiping Shi;Yimin Wang;Liguo Wang","doi":"10.1109/JSTARS.2025.3639670","DOIUrl":"https://doi.org/10.1109/JSTARS.2025.3639670","url":null,"abstract":"With the rapid development of deep learning technology, significant progress has been made in the field of remote sensing (RS) scene image classification. However, the large intraclass distance and high interclass similarity still pose significant challenges for RS scene classification. In addition, there are multiscale targets in RS images, which make significant differences in target characteristics. To overcome the above limitations, this article proposes a novel attention-fusion focused multiscale architecture network (AF<sup>2</sup>-MSA Net). First, a multilevel feature extraction module (MFEM) was designed to extract semantic and detail information at different scales from RS images. Subsequently, an intricately designed global context recalibration module (GCRM) was embedded into MFEM, and the features at each level were enhanced through a global context recalibration mechanism, enabling the model to dynamically focus on key semantic regions and important contextual information. Next, an axis-aligned feature harmonization module (AAFHM) was constructed to fuse multiscale features from adjacent stages layer by layer. This module combines attention mechanisms from both channel and spatial branches to adaptively coordinate and fuse multiscale contextual information, achieving deep collaborative optimization of different scale features. Finally, the GCRM and AAFHM are integrated into a unified framework called AF<sup>2</sup>-MSA Net to achieve collaborative optimization of global semantics and multiscale discriminative features. Extensive experiments on three commonly used datasets have shown that the proposed AF<sup>2</sup>-MSA Net outperforms some state-of-the-art methods in RS image scene classification tasks.","PeriodicalId":13116,"journal":{"name":"IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing","volume":"19 ","pages":"1150-1164"},"PeriodicalIF":5.3,"publicationDate":"2025-12-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11275695","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145729382","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}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：