最新刊期
卷 30 ， 期 3 ， 2026
- “介绍了其在矿区土壤监测领域的研究进展，专家们探索了可见—短波红外高光谱遥感技术在矿区土壤理化特性监测及问题诊断中的应用，为矿区土地精准复垦与生态系统可持续修复提供了技术支撑。”
摘要：Resource extraction activities fundamentally reshape soil systems through a series of processes such as surface stripping, underground excavation, waste dumping, and mineral processing. These operations disrupt natural soil by removing topsoil horizons, mixing soils with fragmented rock materials, and introducing exogenous substances associated with tailings and beneficiation residues. As a result, mining-affected soils commonly exhibit pronounced alterations in physical structure, chemical composition, and surface morphology, including increased coarse fragment content, reduced soil nutrients, and elevated concentrations of potentially toxic elements. Such changes often lead to a series of environmental problems, such as soil degradation, contamination, and desertification-like processes, ultimately impairing soil fertility and ecosystem functioning. Effective monitoring and assessment of these soil disturbances therefore constitute a critical prerequisite for guiding ecological restoration, evaluating reclamation effectiveness, and supporting sustainable management in mining areas.In recent years, visible-shortwave infrared (VNIR-SWIR) hyperspectral remote sensing technology has demonstrated significant potential in soil physicochemical properties estimation and environmental diagnosis in mining regions, owing to its non-destructive nature, high efficiency, and capability for large-scale, continuous monitoring. This potential arises from the sensitivity of VNIR-SWIR hyperspectral technology to diagnostic absorption features, reflectance magnitude, and spectral shape variations controlled by soil mineral composition, organic matter content, and moisture conditions. This capability is particularly valuable in mining landscapes, where soil properties exhibit pronounced spatial heterogeneity and rapid temporal changes driven by anthropogenic disturbance and reclamation interventions.Based on a comprehensive review of the literature published over the past three decades, this review systematically: (1) examines the formation and evolution processes of mining-affected soils, along with their key physical and chemical characteristics and corresponding spectral responses in the VNIR-SWIR domain; (2) summarizes the theoretical foundations and applicability of hyperspectral remote sensing estimation models, including empirical models, physical models, and machine learning approaches; and (3) highlights recent advances in the application across a wide range of mining environments, including open-pit and underground operations in both metal and coal mining areas, with representative case studies reported from major mining regions worldwide—such as China, Australia, and Europe—using laboratory spectroscopy, UAV-based imaging, and satellite hyperspectral techniques to estimate key soil quality indicators—such as organic matter, total nitrogen, and heavy metal concentrations.Research findings indicate that VNIR-SWIR hyperspectral technology, when integrated with appropriate estimation models, has notably improved the accuracy of estimating key soil physicochemical properties in mining areas. Nevertheless, significant challenges remain in terms of model stability, regional adaptability, and dynamic monitoring capabilities. Spectral interference caused by heterogeneous surface materials, bidirectional reflectance effects, scale mismatches among platforms, and uncertainties associated with model transfer across different mining regions continue to limit operational applications. Future research should focus on deepening the mechanistic understanding of soil–spectral interactions, promoting the fusion of multi-source, multi-scale, and multi-temporal data, and advancing the development of generalizable and transferable modeling frameworks. Moreover, multi-scale observations—from proximal sensing and UAV platforms to spaceborne hyperspectral missions—have expanded the spatial and temporal applicability of soil monitoring in complex mining settings. Collectively, these advances provide new opportunities for bridging the gap between experimental studies and regional-scale applications. Such efforts are essential for enabling robust, long-term monitoring of soil quality, supporting evidence-based land reclamation planning, and ultimately facilitating sustainable ecological restoration in mining landscapes.关键词：mining soil;VNIR-SWIR hyperspectral remote sensing;soil nutrients;heavy metals in soil0|0|0更新时间：2026-03-17 - “离子吸附型稀土矿开采致土壤污染，植被复垦效果差。专家提出YOLOv8-AS检测方法，改进YOLOv8n网络，提升无人机影像中矿区复垦植被识别定位能力，为生态恢复提供技术支持。”
摘要：The leaching mining of ion-adsorbed rare earth ore primarily employs in-situ leaching, pile leaching, and pool leaching methods, which result in significant soil pollution. This pollution presents serious environmental challenges, particularly affecting the growth and survival rates of reclaimed vegetation in rare earth mining areas. The restoration of reclaimed vegetation is crucial for mitigating environmental damage and restoring ecological balance. However, the application of intelligent technology to monitor and manage the health and growth of reclaimed vegetation in these mining areas encounters substantial challenges due to the complexities of the natural environment.Unmanned Aerial Vehicle (UAV) remote sensing image technology has emerged as a promising tool for monitoring and evaluating ecological restoration efforts in rare earth mining areas. The UAV can rapidly capture high-resolution images over large areas, facilitating efficient monitoring of reclaimed vegetation growth in these regions. However, the uneven spatial distribution, varying shapes, and diverse overall characteristics of reclaimed vegetation present significant challenges for achieving high-precision automatic recognition from UAV images. Consequently, relying solely on traditional image processing techniques for vegetation detection and classification is inadequate. To address these challenges and enhance the automatic recognition and localization capabilities of individual reclaimed vegetation in UAV images, this paper proposes a method for reclaimed vegetation detection in rare earth mining areas (YOLOv8n), which integrates the global feature YOLOv8-AS. This method represents an innovative improvement over YOLOv8n. First, the downsampling module ADown is introduced to optimize the feature convolution operation, thereby reducing the feature loss during the deep model training process. Second, the spatial pyramid pooling fast-global feature pool module is employed for feature extraction, significantly enhancing the detection capability of reclaimed vegetation with substantial variations in overall features.Results showed that in the self-constructed rare earth mining reclamation vegetation dataset, YOLOv8-AS outperforms YOLOv8n by 1.6% and 2.4% in terms of mAP@0.5 and mAP@0.5—0.95, respectively. Compared with those of YOLOv8n, the model size, number of parameters, and floating point computation of YOLOv8-AS decreased by 11%, 10%, and 9%, respectively. The mAP@0.5 and mAP@0.5—0.95 for the YOLOv8-AS algorithm are 91.1% and 46.8%, respectively. When compared with SSD, faster R-CNN, RT-DETR, YOLOv5, YOLOv7, and YOLOv7-TINY models in terms of mAP@0.5, YOLOv8-AS shows improvements of 14.07%, 23.32%, 1.2%, 2.3%, 3.3%, 2.9%, and 1.2%, respectively. According to the comparative experimental results of YOLOV8-AS and YOLOv8 across three scenarios—characterized by a predominance of small targets, simplicity, and complexity—the mAP@0.5—0.95 of YOLOV8-AS increased by 2.3%, 1.2%, and 3%, respectively, when compared with the baseline model YOLOv8. Furthermore, we applied YOLOv8-AS to the reclamation vegetation detection task in a larger scene within a rare earth mining area. The visualization results indicate that, regardless of the scenario—whether featuring numerous small targets, simple scenes, or complex environments—this method significantly enhanced its capacity to identify and accurately locate individual plants in the reclamation vegetation. This finding further substantiates its efficacy in accurately detecting reclaimed vegetation across various conditions. Such advancements are crucial for effectively monitoring the progress of ecological restoration in mining areas and provide essential support for achieving sustainable mining development.关键词：deep learning;object detection;YOLOv8n;UAV Imagery;Rare Earth Mining Area;Reclaimed Vegetation541|662|0更新时间：2026-03-17 - “新疆三道坝煤田火区研究取得新进展，专家构建基于STL时序分解的煤火监测方法，有效分离地表温度数据中的季节性和随机波动影响，精准反映地表温度演变趋势，为煤火监测和治理提供新思路。”
摘要：Coal fires are a major global environmental hazard characterized by long combustion cycles, strong concealment, and difficult mitigation. They pose severe risks to ecological security, human health, and energy resources. Coal-fire evolution is a continuous spatiotemporal process, which is why land surface temperature (LST) is a key indicator for identifying thermal anomalies and tracking fire development. With the rapid accumulation of multisource remote sensing data, time-series analysis has become a powerful tool for long-term coal-fire monitoring. However, LST data often exhibit strong randomness and complex variability, which introduces challenges for long-term sequence analysis. This study aims to develop a robust coal-fire monitoring method capable of accurately characterizing long-term LST variations and identifying coal-fire evolution patterns.The Sandaoba coalfield in Xinjiang, China, was selected as the study area. Using Landsat imagery and the Google Earth Engine platform, a continuous LST dataset from 1998 to 2023 was constructed. The seasonal-trend decomposition procedure based on loess (STL) was applied to separate the LST time series into trend, seasonal, and residual components, allowing the removal of strong seasonal fluctuations and stochastic disturbances. Spatiotemporal patterns of LST were analyzed using the extracted trend component. The random sample consensus (RANSAC) algorithm was used to fit long-term temperature trends at the pixel scale, ensuring further identification of coal-fire development stages and delineation of active fire zones. Field survey data collected in 2016 were employed to validate the identification results.We obtained the following resultsThe STL decomposition effectively separated the long-term LST series into stable trend components and variable seasonal and residual signals. The extracted trend item revealed long-term warming trajectories associated with coal-fire activity more accurately compared with the raw LST series. Among the 20 coal-fire points detected in the 2016 field survey, 16 fell within high-value areas of the mean and range of the trend component, indicating strong spatial consistency. The RANSAC-based trend fitting captured spatiotemporal LST evolution from 1998 to 2023 and demonstrated strong agreement with field observations, successfully identifying the initiation, expansion, and gradual stabilization stages of coal fires. These results validate the robustness and reliability of combining STL decomposition with RANSAC for long-term coal-fire monitoring.The proposed STL-based method significantly improves the capability to extract meaningful long-term LST trends and enhances the adaptability of coal-fire monitoring to complex spatiotemporal conditions. The method accurately identifies thermal anomalies associated with coal-fire development and shows strong consistency with field investigation results. This framework enables long-term, large-scale LST analysis and provides an effective tool for capturing the spatiotemporal characteristics of coal-fire evolution. The findings offer valuable technical support for future coal-fire surveillance, early warning, and mitigation planning.关键词：Coal fire identification;STL decomposition;Landsat;LST time series;thermal infrared remote sensing337|343|0更新时间：2026-03-17
MiningAreaRemoteSensing
- “针对复杂场景下SAR图像目标检测任务，专家提出Tri-GCA-FCOS网络，融合有监督注意力机制与SAR图像特性，有效提升目标检测性能，为复杂场景目标检测提供新方案。”
摘要：Target detection in Synthetic Aperture Radar (SAR) images presents significant challenges due to the unique electromagnetic scattering mechanisms and severe background clutter found in complex scenes. While traditional methods leverage physical properties and modern deep learning approaches utilize data-driven features, a gap remains in effectively combining these paradigms. Current anchor-free detectors often lack the specific adaptability required for the high-dynamic-range and speckle-noise characteristics of radar imagery. To address these issues, this study aims to enhance the capability of neural networks to perceive and capture targets by integrating domain-specific physical knowledge with advanced deep learning architectures. Specifically, the research seeks to overcome the limitations of the standard Fully Convolutional One-Stage Object Detection (FCOS) baseline by proposing a novel framework that explicitly exploits both the gradient edge information and the strong scattering characteristics of SAR targets to improve detection accuracy and robustness.Inspired by traditional interpretation techniques and deep learning advancements, this paper proposes a novel anchor-free network named Triple-Gradient-CFAR-Attention FCOS (Tri-GCA-FCOS). The methodology comprises three core innovations integrated into an end-to-end framework. First, recognizing the limitations of the generic baseline in handling radar data, a specialized attention branch supervised directly by Ground Truth (GT) labels is designed. This branch guides the network to focus on potential target regions during early feature extraction. Second, a multistream sub-network structure combined with a Tri-channel Interactive Channel-Spatial Attention Fusion (T-ICSAF) module is implemented. This module fuses original image features with pre-extracted gradient edge maps and strong scattering maps, allowing the network to learn rich and complementary features. Finally, a Combinedsupervised-Spatial and Squeeze-and-Excitation Channel Attention Mechanism (CSSCAM) module is proposed to address the issue where introducing physical characteristics might amplify background noise. Supervised by GT masks, the CSSCAM acts as a sophisticated filter that simultaneously enhances salient target features and suppresses background clutter before the final prediction stage.The proposed Tri-GCA-FCOS framework enables joint training of all components, ensuring optimal compatibility between deep features and handcrafted physical priors. The performance was evaluated using real-world data from a Miniature Synthetic Aperture Radar (MiniSAR). Extensive experiments demonstrate that the proposed method significantly outperforms the baseline and other state-of-the-art detection models. Visualization results confirm that the T-ICSAF module effectively aggregates gradient and scattering information, while the CSSCAM module successfully highlights target areas and attenuates complex background interference. Quantitative metrics indicate superior precision and recall rates, particularly in scenarios involving dense targets or high-clutter backgrounds, validating the rationale of the network design.This paper successfully presents a robust anchor-free SAR target detection method that bridges the gap between traditional physical feature extraction and data-driven deep learning. By explicitly incorporating gradient edges and strong scattering properties through a supervised attention mechanism, the Tri-GCA-FCOS network achieves a deeper understanding of targets. The proposed fusion and suppression modules effectively resolve the conflict between feature enhancement and noise amplification. The proven effectiveness on MiniSAR data suggests that this approach provides a valuable solution for high-precision SAR interpretation in complex monitoring scenarios, offering a promising direction for future research in physics-aware deep learning models.关键词：synthetic aperture radar;target detection;fully convolutional one-stage object detection;gradient magnitude image;constant false alarm rate0|0|0更新时间：2026-03-17 - “2023年5月31日天津八里台突发沉降，造成周边建筑和基础设施严重破坏。专家结合多波段SAR数据优势，提出复杂形变提取技术，利用C波段Sentinel - 1和L波段陆探1号SAR数据，获取并分析突发沉降区域形变信息，为八里台突发沉降产生及其动力学机制解析奠定技术和数据基础，有利于优化地热资源开发技术，促进城市安全及可持续发展。”
摘要：On May 31, 2023, a sudden serious ground subsidence event occurred in Balitai, Tianjin, China, causing severe damage to surrounding buildings and infrastructure. Conducting long-term, high-precision surface deformation monitoring in the affected area is of great significance for understanding the mechanisms of the event and for the prevention and control of related disasters. The Balitai subsidence event is mainly characterized by a large magnitude of deformation, a long duration, and significant nonlinearity. Traditional single-source Synthetic Aperture Radar (SAR) interferometry (InSAR) techniques for such deformation monitoring face challenges brought by the large gradient-induced decorrelation and insufficient spatiotemporal resolution. To address these issues, this study proposes a complex deformation extraction technique that combines multilevel variable-window multitemporal weighted-pixel offset tracking with distributed scatterer InSAR (DS-InSAR), leveraging the high-resolution intensity and high-precision phase information of multiband SAR data. This approach could overcome the problems of random errors and lack of continuity in traditional pixel offset tracking due to its low coherence. Long-term deformation monitoring results after the subsidence event was obtained by applying geometric normalization to the deformation monitoring results of Sentinel-1 and LuTan-1 data on the basis of the assumption that the surface deformation is mainly vertical after the subsidence event. With 57 C-band Sentinel-1 and 19 L-band LuTan-1 SAR imagery from January 2022 to December 2024, this study obtained and analyzed the deformation information of the sudden subsidence area in the pre-event, initial, and post-event phases. Results show that the Balitai subsidence event is highly sudden and phased. The study area obviously remained stable before the event (January 2022—May 31, 2023), with average surface deformation velocity within -3 to 3 mm/year. During the initial phase (May 31—June 7, 2023), the maximum deformation exceeded -4.5 m near the southern wellheads. In the post-event phase (June 7, 2023—December 28, 2024), the deformation rate decreased rapidly and showed significant consistency with the monitoring results of two GNSS stations during the same period, with a coefficient of determination of 0.98 and 0.97, and a root mean square error of 3.31 and 5.85 mm, respectively. The corresponding results thus validate the reliability of time-series InSAR monitoring outcomes. The cumulative deformation and the results obtained by fitting the exponential decay and polynomial function models indicate that most of the surface will have stabilized by 2025. The spatial correlation between the subsidence center and the geothermal well location, as well as the temporal consistency between the event timing and the drilling operation period, suggest that improper geothermal resource exploitation may be the main cause of the sudden subsidence. The proposed method could provide a powerful spatial monitoring approach for the Balitai subsidence event, fully demonstrating the high spatiotemporal resolution performance of China’s LuTan-1 satellite. It also lays a technical and data foundation for the analysis of the generation and dynamic mechanisms of the sudden subsidence in Balitai, which is conducive to optimizing geothermal resource development techniques and promoting urban safety and sustainable development.关键词：Balitai Sudden Subsidence;Sentinel-1;LuTan-1;pixel Ttracking;DS-InSAR;Spatiotemporal evolution109|465|0更新时间：2026-03-17 - “相关研究在合成孔径雷达（SAR）影像目标提取领域取得新进展，专家们构建了基于频域—空间协同的Transformer网络S3T-Net，为解决SAR影像中相干斑噪声干扰及复杂地物影响下目标提取精度不足问题提供了新方案。”
摘要：Synthetic Aperture Radar (SAR) image segmentation remains challenging due to inherent speckle noise, which obscures object boundaries and complicates the interpretation of scenes containing complex and heterogeneous land covers. Existing deep learning methods often suffer degraded accuracy under strong speckle interference and confusing background clutter.To address these issues, we propose a spectral-spatial synergetic transformer network (S3T-Net) for fine-grained target extraction in SAR imagery. S3T-Net adopts a dual-encoder design that couples a frequency-domain encoder with a vision transformer (ViT) encoder. The frequency-domain encoder uses Discrete Wavelet Transform (DWT) downsampling and a spectral-hierarchical dual-domain attention (SHDA) module to capture local texture details while reducing sensitivity to noise. In parallel, the ViT encoder leverages global self-attention to model long-range dependencies and overall structural context. The two feature streams are fused by a synergistic weighted feature confluence (SWFC) module. In addition, a recursive frequency-space refinement (RFSR) module is introduced to suppress upsampling noise and sharpen target boundaries.Experiments on three public SAR datasets—SARBuD (buildings), HRSID (ships), and FRBS (oil spills)—demonstrate that S3T-Net consistently outperforms multiple state-of-the-art methods in terms of Dice coefficient, with improvements of 0.52%, 0.62%, and 1.04%, respectively.By jointly exploiting complementary spectral and spatial cues, S3T-Net enhances feature characterization in highly noisy SAR scenes and improves segmentation accuracy across different target types. This spectral-spatial collaborative framework provides a practical and extensible solution for SAR target extraction. The source code is available at https://github.com/Yanghan37/S3T-Net.关键词：SAR imagery;object extraction;deep learning;noise suppression;frequency-spatial synergy0|0|0更新时间：2026-03-17
SARMethodsandApplications
- “介绍了其在城市蓝色空间监测领域的研究进展，专家们基于U-Net架构改进，提出了CRRU - Net深度学习模型，为科学协调城市发展与蓝色空间保护提供了数据支撑。”
摘要：Quantitative assessment of urban blue space distribution patterns and change dynamics holds significant scientific and practical value for coordinating urbanization processes with aquatic ecological protection. Current remote sensing studies on urban blue space monitoring are limited by image spatial resolution, posing challenges in accurately extracting small and micro water bodies and finely classifying water body types. Moreover, existing research lacks comprehensive, national-scale thematic reports on the distribution patterns and evolutionary features of urban blue spaces in China’s state-level new areas.To address these gaps, this study improves upon the U-Net architecture by introducing a channel recalibration mechanism and residual connections, proposing the CRRU-Net deep learning model for water body classification in high-resolution remote sensing imagery. The model was applied to monitor 19 state-level new areas across China. Additionally, the study generated a water body classification dataset using high-resolution (4 m) remote sensing imagery from 2014 and 2023, enabling differentiated change monitoring for four water body types: rivers, lakes, reservoirs, and ponds.Experimental results show that CRRU-Net achieves the highest accuracy in water body classification tasks, with an overall accuracy of 94.96%. Findings indicate that the total urban blue space area in China’s state-level new areas increased by 70.37 km² from 2014 to 2023. Coastal new areas (excluding Tianjin Binhai New Area) generally experienced reductions, while inland new areas (except Yunnan Dianzhong New Area) showed expansions; some regions appeared stable due to balanced gains and losses. Specifically, river and reservoir areas increased by 11.56% and 10.39%, respectively, since the 2014 initial construction phase, whereas lake and pond areas decreased by 15.35% and 10.26%, respectively.The results provide data support for scientifically coordinating urban development and blue space protection, offering significant practical implications for advancing sustainable urbanization.关键词：state-level new areas;urban blue space;hhigh-resolution remote sensing imagery;deep learning;semantic segmentation;U-Net;water body classification;Change Detection0|0|0更新时间：2026-03-17 - “随着海洋资源开发的深入，中国近海海域生态环境面临日益严峻的挑战。遥感技术因具备连续观测和广域覆盖的特点，已成为识别和分析海洋生态风险的重要技术手段。烟台南岸近海水域的研究表明，溢油和浒苔两类典型生态风险源的识别精度在多项评估指标中均超过了90%。”
摘要：With the deepening exploitation of marine resources, the ecological environment of China’s coastal waters is facing increasingly severe challenges. Remote sensing, with its capability for continuous observation and wide coverage, has become a crucial tool for identifying and analyzing marine ecological risks. This study focuses on the coastal waters of the southern coast of Yantai, targeting two typical ecological risk sources: oil spills and Ulva prolifera. By leveraging optical and SAR multi-source remote sensing imagery, deep learning models were developed to monitor these phenomena over time. Based on this monitoring, the spatiotemporal distribution characteristics of oil spills and Ulva prolifera from 2018 to 2021 were comprehensively analyzed. The results show that both models achieved accuracies exceeding 90% across multiple evaluation metrics. Oil spills occurred mainly beyond 30 nautical miles southeast of the coast, with higher frequencies in 2018, 2019, and 2021, primarily attributed to ship discharges. Ulva prolifera outbreaks predominantly occurred during the summer months, with large-scale events in 2019 and 2021 and higher intensities in the southwestern part of the study area.Further overlay analysis revealed a spatial gradient in ecological risks—dominated by Ulva prolifera in nearshore waters, increasing oil spill risks in midshore zones, and overlapping dual-risk hotspots in offshore areas. For reproducibility, the trained deep learning detection models for oil spills and Ulva prolifera are available at the following link:https://github.com/lile13955267907/DA-net.关键词：marine ecological environment;oil spills;Ulva prolifera;remote sensing;deep learning0|0|0更新时间：2026-03-17 - “介绍了其在遥感蒸散模型验证领域的研究进展，专家探索了光学—微波双波长光闪烁方法与遥感ET模型结果在时间尺度匹配方面的优势，为解决二者时间尺度不匹配问题提供了有效途径。”
摘要：Most current thermal-based remote sensing evapotranspiration (ET) models estimate sensible heat flux (H), latent heat flux (LE), and ET using surface temperature and other parameters derived from multispectral and thermal imagery acquired within tens of seconds during satellite overpasses. However, their instantaneous flux estimates are commonly validated against 30 min averaged flux data from the Eddy Covariance (EC) method, leading to a significant temporal scale mismatch. Developed over the past two decades, optical-microwave two-wavelength scintillometry measures H and LE at scales ranging from hundreds of meters to 10 km, is suitable for complex environments including mountainous terrain, and yields statistically stable flux results with a shorter averaging time of 1—2 min. As a complement to EC, this method provides a critical opportunity for validating remote sensing ET models and products. On the basis of observations from Optical-Microwave Scintillometer (OMS) and EC systems as well as meteorological profile towers in the alpine grasslands of the upper reaches and oasis croplands of the middle reaches of the Heihe River Basin, northwest China, this study highlights the outstanding advantages of OMS flux observations in temporal-scale matching with ET model estimates. Comparative analysis of OMS and EC flux across different averaging times (1 min, 2 min, 5 min, 10 min, 15 min, 30 min) confirms the theoretical feasibility that OMS can generate statistically stable fluxes with averaging periods as short as 1 min. This time scale is highly consistent with the scanning duration of single-satellite images (e.g., Landsat). On the basis of this finding, analysis of 1 min OMS data in alpine grassland shows that H and LE over the 30 min periods typically used for model validation often vary by 10%—30%. Using the 1 min flux data at the time of satellite overpass to validate ET model results introduces uncertainties of similar magnitude, which are exacerbated under atmospheric nonstationary conditions. Validation of instantaneous H and LE estimated by the two-source surface energy balance (TSEB) model and Landsat images using OMS and EC flux data from oasis croplands demonstrates better agreement between TSEB outputs and 1 min OMS flux observations. Optical-microwave scintillometry enables ground observations to match the spatiotemporal scales of satellites, providing an effective method for validating thermal-based ET models and products. Widespread application of this method will promote the optimization of ET retrieval algorithms and improve the accuracy of related products.关键词：Two-wavelength scintillometry;remote sensing evapotranspiration model;Heihe river basin;sensible and latent heat fluxes;temporal-scale matching105|104|0更新时间：2026-03-17 - “闭合林线动态变化反映高山生态系统与环境变化的相互作用，为揭示全球变暖下山地生态系统的响应机制提供关键范本。”
摘要：Closed-loop alpine treelines, characterized by distinct forest boundaries, are widely recognized as sensitive ecological indicators that reflect the complex interactions between high-mountain ecosystems and environmental changes. Understanding the spatiotemporal dynamics of treelines in China is particularly important, as these ecosystems are highly vulnerable to climatic variability and anthropogenic pressures. However, systematic insights remain limited due to the scarcity of long-term monitoring data and the multiscale complexity of mountain environmental influences. This study aims to overcome these challenges by integrating multi-source datasets and advanced analytical approaches to investigate the patterns, shifts, and driving mechanisms of treelines across China during the period 2000—2020.To achieve this objective, the study combined three complementary datasets—Global Forest Change (GFC), Global Land Cover and Land Use Change (GLCLUC), and the China Annual Tree Cover Dataset (CATCD). By employing the eight-connected component algorithm, this research systematically identified treeline locations and quantified their changes. Comparative analysis of the three datasets was conducted to assess spatial consistency and discrepancies, particularly in complex mountain regions such as the Hengduan Mountains. Furthermore, the study examined treeline shifts over two decades and applied XGBoost-based explanatory models to disentangle the relative contributions of topographic, climatic, and human factors to the observed treeline dynamics.Key findings reveal the followingFirst, spatial distributions derived from the three datasets are generally consistent, yet discernible discrepancies emerge in treeline quantity and positional accuracy (particularly in the Hengduan Mountains), highlighting dataset-specific detection sensitivity. Second, most treelines showed an upward shift during the study period: 92.5% (CATCD), 87% (GFC) and 51% (GLCLUC), showing evident regional variation, with steeper increases in the west and more gradual changes centrally and eastward. Third, across the 33 modeled treelines, the XGBoost models achieved a mean R2 of 0.771 (77.1%). In terms of dominant driver classification, topography served as the dominant controlling factor for the majority (60.6%) of treelines, followed by climatic drivers (36.4%), with human activities having a minimal contribution (3.0%).This study enhances the understanding of climate change impacts on China’s mountain ecosystems by revealing treelines’ spatiotemporal dynamics. The results provide a technical basis and tools for assessing ecological vulnerability and tracking treeline shifts. These insights further inform actionable frameworks for global mountain ecosystem conservation strategies.关键词：alpine treelines;treeline spatiotemporal dynamics;climate change;alpine ecosystem;remote sensing monitoring;disturbance mechanisms0|0|0更新时间：2026-03-17
Ecology and Environment
- “利用国产高光谱卫星数据，专家探测美国二叠纪盆地甲烷排放，分析其时空特征，为油气行业甲烷排放研究提供新思路。”
摘要：Methane emissions in the oil and gas industry manifest as both intermittent (highly variable and random) and persistent types. A single satellite data source may miss highly intermittent methane emission sources because of its limited effective data coverage. In this study, data from three domestically developed hyperspectral satellites, namely, the ZY-1E, GF-5B, and GF-5A carrying the Visible-shortwave Infrared Advanced Hyperspectral Imager, were used to detect methane emissions in the Permian Basin of the United States. The temporal variation characteristics of these emissions were analyzed by increasing the revisit observation frequency for methane point source emissions, aiming to provide a scientific basis for formulating methane emission reduction strategies.This study utilizes the L1 reweighted ISTA matched filter algorithm to detect methane emissions in the Permian Basin, USA. The temporal variation characteristics of specific point sources were analyzed by calculating the detected plume frequency (f) to analyze point source persistence, and combining it with methane point source data provided by Carbon Mapper. This work aims to provide more detailed information on the uneven distribution and dynamic changes of methane emissions across various segments of the oil and gas industry. (1) Simulation experiments using the L1 reweighted ISTA matched filter algorithm showed an accuracy range of 18% to -16%. Simultaneously, two point sources located in the overlapping area of a three-image mosaic were quantified, yielding an average observation error of 2.38%; (2) A total of 56 methane point sources were detected within the study area. Their quantity distribution was approximately 55.36% from production processes, 33.93% from gathering processes, and 10.71% from processing plants. The emission distribution was approximately 58.19% from production, 28.74% from gathering, and 13.07% from processing; (3) Among the detected point sources, highly intermittent sources (0 < f ≤ 0.25) accounted for 77.97% by count and 67.35% by estimated emissions. Sources with higher persistence (0.50 < f ≤ 1) accounted for 5.56% by count and 10.59% by estimated emissions; (4) The emission rates of the three selected point sources were all highest in December and lowest in August; (5) Furthermore, six methane emission point sources were detected across three typical oil/gas fields: Haynesville, Rumaila, and the Western Turkmenistan oil field. A persistent emission source was identified in the Western Turkmenistan oil field; despite having only two valid satellite overpasses, methane emissions were detected in both instances. (1) The L1 reweighted ISTA matched filter algorithm produced relatively accurate results for detecting methane emission point sources, and multisource satellites effectively enhanced the detection of intermittent point sources; (2) The majority of methane emission point sources originated from production processes, and most emission sources exhibited intermittent characteristics; (3) The emission rates of methane point sources in the Permian Basin are related to the seasonal fluctuations of U.S. oil and natural gas production.关键词：GF-5B;GF-5A;ZY-1E;methane emission source detection;detection plume frequency;Permian basin0|0|0更新时间：2026-03-17 - “介绍了其在海表温度遥感领域的研究进展，相关专家采用时空匹配方法将AHI近红外影像与ATMS微波辐射计波段数据融合，建立匹配数据集，并通过三种机器学习方法评估融合数据的SST反演精度，为解决云覆盖导致的SST预估异常及数据缺失问题提供了有效的技术参考。”
摘要：Sea Surface Temperature (SST) is crucial for managing marine ecosystems and mitigating oceanic disasters. For more than three decades, the scientific community has been dedicated to improving the precision of SST products through satellite remote sensing techniques. Despite the potential of thermal infrared sensors to yield SST estimations with high spatial and temporal precision, cloud cover undermines the accuracy of SST acquisition.This study presents a novel approach to improving SST retrieval accuracy by integrating multisource remote sensing data. The approach addresses the challenge of cloud cover by using the cloud-penetrating capabilities of microwave sensors in conjunction with infrared sensor data. This methodology involves generating level 1 sample datasets through spatiotemporal matching with in-situ SST data. After extensive pre-processing, the dataset is categorized into clear skies and cloud cover conditions. This paper employs three advanced machine learning algorithms—XGBoost, SVR, and RF—to conduct SST inversion with synergistic data from AHI/ATMS sensors. The performance of these algorithms is rigorously assessed through a comparative analysis of their inversion results and the Himawari-8 SST production. Moreover, the analysis meticulously examines SST inversion accuracy across diurnal and nocturnal conditions, effectively exploring between daytime and nighttime inversion accuracies.These findings demonstrate that integrating ATMS microwave data markedly improves the accuracy of SST inversion, particularly in cloudy conditions. The XGBoost algorithm exhibits exceptional performance, with an RMSE of 1.707 ℃ and an R² of 0.935. AHI/ATMS data effectively address data inconsistencies and cloud cover issues and highlights the importance of multiple sources of data to obtain a comprehensive and accurate SST dataset.This paper confirms the significant impact of multisource data on improving the accuracy and broadening the spatial coverage of SST inversions. The proposed approach effectively diminishes cloud interference, providing a compelling argument for the adoption of ATMS microwave sensing to overcome the challenges posed by cloud cover. Additionally, this study emphasizes the potential of machine learning algorithms to improve the resolution and accuracy of SST estimates, generating high-precision, wide-coverage SST distribution maps that provide important data for the effective management of marine ecosystems and proactive prevention of marine disasters.关键词：multi-source remote sensing data;SST;all-weather SST inversion;spatio-temporal matching;machine learning;thermal infrared and microwave remote sensing;AHI near-infrared imagery;ATMS microwave radiometer0|0|0更新时间：2026-03-17 - “专家利用SAR数据，引入极化特征参数，基于SVR和CNN算法构建中高海况下海浪风场参数联合反演模型，显著提高反演精度，为海洋环境研究提供新方法。”
摘要：Synthetic Aperture Radar (SAR), as an active microwave sensor, has high resolution and the ability to observe the sea in all-weather and all-time conditions, playing an important role in monitoring ocean wave and wind fields. However, existing wave and wind field parameter inversion methods still face low inversion accuracy in moderate to high sea conditions. In this work, we established two joint inversion models of wave and wind field parameters suitable for moderate to high sea conditions and verified their inversion performance, providing an important reference for subsequent marine environmental studies under extreme sea conditions.Addressing the problem of low inversion accuracy of existing wave wind field parameter inversion methods under medium to high sea conditions, this paper utilizes Sentinel-1 WV mode SAR data and introduces three approximate polarization feature parameters (polarization entropy, anisotropy, and correlation) as additions on the basis of traditional characteristic parameters (incident angle, normalized variance). Two joint inversion models for SAR wave wind field parameters that are suitable for medium to high sea conditions (wind speed of 10.8–28.8 m/s) were constructed on the basis of Support Vector Regression (SVR) and Convolutional Neural Network (CNN) algorithms.Results show that the root mean square errors (RMSE) of wind speed, wind direction, significant wave height, and average wave period inverted on the basis of SVR, compared with ERA5 and NDBC data, are 1.27 m/s, 23.46°, 0.22 m, and 0.62 s, and 1.10 m/s, 25.37°, 0.22 m, and 0.59 s. The RMSE for wind speed, wind direction, significant wave height, and average wave period inverted based on CNN compared with ERA5 and NDBC data are 1.15 m/s, 23.53°, 0.18 m, and 0.53 s, and 1.08 m/s, 25.85°, 0.19 m, and 0.54 s, demonstrating that both joint inversion models established in this paper can effectively invert wave wind field parameters under medium to high sea conditions. Compared with traditional theoretical methods, the joint inversion models based on SVR and CNN significantly reduced the RMSE of significant wave height, average wave period, and wind speed compared with ERA5 data by 0.67 m, 0.62 s, and 0.3 m/s, and 0.71 m, 0.71 s, and 0.42 m/s, respectively, proving that both joint inversion models significantly enhance inversion accuracy. Two scenes of hurricane data from the North Atlantic in 2020 were collected for case analysis to verify the inversion effect of the two joint inversion models under high sea conditions. Results indicate that both joint inversion models can also achieve relatively accurate wave wind field parameters under high sea conditions.In conclusion, the joint inversion models developed in this paper provide an effective solution for high-precision inversion of wave wind field parameters under medium to high sea conditions and offer important references for subsequent marine environment studies under extreme sea conditions.关键词：synthetic aperture radar;joint inversion method;CNN;SVR;wave wind field parameters;moderate to high sea conditions;polarization characteristic parameters142|262|0更新时间：2026-03-17
Atmosphere and Ocean
-
Quality control methods and framework for
in situ observation data in remote sensing validation AI导读“遥感真实性检验领域迎来新突破，专家们深入研究地面观测数据误差来源，构建了全面的质量控制体系，为提升遥感像元真值准确性与可靠性提供有力支持。”
摘要：In situ observation data serve as the primary source of pixel-scale reference truth for the validation of remote sensing products and play a crucial role in remote sensing modeling, parameter inversion, and products quality assessment. Reliable validation depends on the availability of in situ observations that are not only accurate but also representative of the spatial and temporal characteristics of coarse scale remote sensing pixels. Over recent decades, numerous in situ observation networks and datasets have been established worldwide. Despite these advances, the quality of in situ observation data remains a major limiting factor in remote sensing validation. Various sources of uncertainty introduced during data acquisition, transmission, and processing may degrade data reliability and propagate errors into truth value and products validation results. Therefore, Quality Control (QC) is essential to ensure the accuracy, reliability, and applicability of in situ observation data. The objective of this study is to systematically review the major sources of error affecting in situ observation data and to summarize existing quality control and evaluation methods for in situ observation data in remote sensing validation.This study systematically reviews the major sources of error and existing problems in in situ observation data for remote sensing validation. On this basis, existing quality control and evaluation methods are reorganized and summarized from six key dimensions: data standardization, completeness, timeliness, accuracy, consistency, and spatiotemporal representativeness, forming a structured quality control framework oriented toward remote sensing validation. To demonstrate the practical application of the reviewed methods, land surface albedo is selected as a representative variable, and typical validation studies are examined to illustrate how quality control and evaluation procedures are applied to mitigate instrumental, measurement, and representative errors.The study indicates that quality control of in situ observation data for remote sensing validation currently faces two major challenges. First, the multisource and heterogeneous characteristics of in situ observation datasets, including differences in instruments, observation protocols, data formats, and sampling strategies, complicate unified processing and automated quality assessment. Second, a unified and systematic quality control framework specifically designed for remote sensing validation is still lacking, resulting in fragmented methods and inconsistent evaluation criteria across different datasets and applications. In addition, representative errors caused by scale mismatches between in situ observations and satellite pixels remain a major source of uncertainty. By synthesizing existing studies, this paper provides a comprehensive overview of quality control and evaluation methods for in situ observation data, clarifying their roles in mitigating instrumental errors, measurement errors, and representative errors, as well as their applicability and limitations in pixel-scale validation.For remote sensing validation, quality evaluation should place particular emphasis on the spatiotemporal representativeness of in situ observations at the pixel scale, taking into account in situ sites distribution, surface heterogeneity, variable characteristics, and temporal coverage. By systematically summarizing quality control methods and framework for in situ observation data, this review provides a reference for standardized data processing and quality evaluation, thereby supporting more reliable and scientifically robust pixel-scale validation of remote sensing products.关键词：in-situ observation data;remote sensing products;validation;Error;quality control;land surface albedo57|52|0更新时间：2026-03-17 -
LiDAR ground point cloud simplification balancing terrain feature and point distribution uniformity AI导读
“针对地面点云简化难题，专家提出新方法，通过加权耦合地形因子、选取特征点集及构建回归预测模型，有效提升地形特征点捕捉精度，为地面点云简化提供新思路。”
摘要：Aiming at the existing ground point cloud simplification algorithms, which have the problems of easy loss of point cloud boundary features, inaccurate identification of feature points and uneven spatial distribution of points, this paper proposes a novel LiDAR ground point cloud simplification method that balances terrain feature and point distribution uniformity.The method begins by weighting and coupling multiple terrain factors to generate a comprehensive factor that thoroughly describes the complexity of terrain changes. Subsequently, an initial feature point set is selected based on the boundary feature distance and feature significance of each point. Finally, a regression prediction model incorporating comprehensive terrain feature constraints is constructed to iteratively predict and capture accurate and uniformly distributed terrain feature points.Six sets of ground point cloud data with different terrain features are selected as the study objects, and the proposed method is compared with seven representative methods. The results demonstrate that the proposed method achieves the highest accuracy, with the average root mean square error and average mean absolute error of the generated digital elevation models reduced by 2.7% to 61.2% and 2.0% to 61.9%, respectively. Additionally, the derivatives (average slope and terrain roughness) are closer to the true values.The proposed method has the best simplification results for ground point clouds, but in terms of computational efficiency, it still needs to be improved compared with advanced algorithms.关键词：remote sensing;LiDAR;point cloud simplification;terrain feature;point distribution uniformity;DEM0|0|0更新时间：2026-03-17 - “遥感图像语义分割领域迎来新突破，相关专家基于UNet网络提出FLGF-UNet，有效解决遥感建筑物提取中的漏检、误检等问题，为城市变化检测等提供有力技术支持。”
摘要：Semantic segmentation of remote sensing images plays a crucial role in fields such as urban change detection, environmental protection, and geological disaster identification. Addressing issues in current remote sensing building extraction, such as missed detections, false positives, and incomplete extractions due to tree obstruction or similar object interference, this paper proposes an improved building extraction network based on the U-Net architecture—the Fusion of Local Global Features Network (FLGF-UNet). This model integrates local and global information, acquiring features in parallel through multiscale local and global modeling. This step ensures that the feature representation at each stage of the network incorporates both local and global information, enabling a more natural and comprehensive integration of information across different scales and levels during the feature extraction process.The parallel feature fusion method of FLGF-UNet ensures that the features of each stage contain fine-grained local information and global dependencies so that the network has both local and global information in the feature representation of each stage. As a result, it effectively overcomes the shortcomings of transformer in local information exchange and at the same time outperforms traditional CNN in global information modeling. The LF module is introduced to extract local information and global context information of different scales to ensure the integrity of feature dependencies, thereby making up for the shortcomings of a single module in feature extraction. In addition, the semantic gap between the encoder and decoder is bridged by adding an interactive fusion module between the encoder and decoder to enhance the fusion effect of spatial details, global context, and semantic features. The superiority and versatility of FLGF-UNet are verified by comparing it with networks such as U2Net, Swin transformer, MA-Net, HD-Net, and RS-Mamba on the WHU dataset, the Massachusetts dataset, and typical urban building instance datasets in China. Results show that FLGF-UNet outperforms other SOTA networks in performance and has high practical application value.In conclusion, FLGF-UNet is an innovative network for extracting buildings from high-resolution remote sensing images. It takes parallel multiscale local-global modeling as its core, so that features at all levels can simultaneously have local details and long-range semantics, effectively bridging the gap between spatial dependence and local details and significantly improving extraction accuracy. Extensive experiments across datasets verified that its performance is significantly better than that of existing methods, providing a reliable solution for high-precision building extraction from high-resolution remote sensing images. FLGF-UNet brings together cutting-edge methods and technological innovations, marking an important leap forward in remote sensing image analysis, and will continue to drive the in-depth development of this field in terms of accuracy improvement, scene expansion, and practical applications.In the future, the breakthrough results of FLGF-UNet will inspire the academic community to continue to delve deeper into the local-global structure. Migrating and expanding this fusion strategy to more remote sensing interpretation tasks has great potential and promising prospects. At the same time, in view of the diverse scale and complexity of urban scenes, further optimizing the model architecture and improving adaptive capabilities are natural evolutionary paths. Subsequent research should focus on the implementation of technology and effectively transform the advantages of the algorithm into perceptible, quantifiable, and sustainable socioeconomic benefits for urban planners and disaster emergency decision-makers.关键词：remote sensing images;building extraction;fusion of local-global feature networks;Feature fusion;interactive fusion module258|204|0更新时间：2026-03-17 - “遥感技术领域迎来新突破，相关专家构建了基于全局感知卷积与 Transformer 的 GCTNet 融合网络，有效解决全色图像与多光谱图像融合时计算复杂度高、特征提取不充分等问题，显著提升融合图像质量并降低计算复杂度，为遥感图像全色锐化应用开辟新路径。”
摘要：Pansharpening plays a crucial role in remote sensing applications such as mineral exploration, urban planning, agricultural monitoring, and geological hazard analysis. High-resolution remote sensing imagery is often obtained by combining a high-resolution panchromatic (PAN) image with a low-resolution multispectral (MS) image, making fusion quality essential for subsequent interpretation tasks. However, existing pansharpening methods often suffer from high computational complexity, limited representation capability, and insufficient extraction of both local spatial details and global contextual dependencies. To address these limitations, this study proposes a novel global-aware convolution and transformer-based fusion network termed GCTNet, which is designed to enhance feature extraction, improve spatial-spectral consistency, and maintain computational efficiency.GCTNet is constructed using a dual-branch multiscale architecture that independently extracts spatial and spectral features from the PAN and MS inputs. A global-aware convolution (GAConv) module is introduced to replace conventional convolution operations through a learnable linear combination of two reference kernels, significantly reducing redundant parameters while preserving detailed spatial modeling capability. A global feature harmonization mechanism further injects global contextual cues into the convolution process. A lightweight transformer module is embedded into the network to capture long-range dependencies that traditional convolutions struggle to represent. This module includes a multi-DConv head transposed attention mechanism for efficient global context modeling and a gated depthwise convolutional feed-forward network for adaptive local feature refinement. Subsequently, an attention-based feature fusion module is employed to adaptively integrate spatial and spectral information from the two branches, ensuring balanced and effective fusion. Multiscale encoders and decoders process hierarchical representations at different resolutions, and a reconstruction module produces the final high-resolution MS images.Comprehensive experiments were conducted on three widely used remote sensing datasets—WorldView-3, WorldView-2, and QuickBird—under reduced-resolution and full-resolution evaluation protocols. Experimental results show that GCTNet achieves leading performance among 13 recent pansharpening methods, including both traditional models and deep learning-based approaches. Quantitative evaluations across multiple metrics, including PSNR, SAM, ERGAS, and SCC, demonstrate significant improvements in spatial detail preservation and spectral fidelity. Visual assessments further confirm that GCTNet produces fused images with sharper structural edges, clearer textures, and reduced spectral distortion. Full-resolution experiments validate the robustness and practical applicability of the proposed approach, while cross-sensor generalization tests—trained on WV3 and tested on WV2—demonstrate strong transferability across different satellite sensors.This study presents GCTNet, a hybrid pansharpening framework that integrates GAConv and transformer modules to effectively capture local spatial details and global contextual dependencies. By reducing redundant parameters, enhancing global feature perception, and incorporating adaptive multiscale fusion, GCTNet achieves state-of-the-art pansharpening performance with high robustness and strong cross-sensor generalization capability. The proposed method provides an effective and practical solution for high-quality remote sensing image fusion and holds great potential for real-world applications in Earth observation, environmental monitoring, and geospatial analysis.关键词：image fusion;remote sensing image processing;pansharpening;deep learning;Transformer;global-aware convolution;multi-scale feature representation;feature extraction0|0|0更新时间：2026-03-17 - “针对地面点云提取地形断裂线难题，专家提出新算法，通过多尺度特征训练与区域增长聚类等技术，有效提升提取完整度与正确性，为高精度DEM建模提供有力支持。”
摘要：Topographic break lines are key structural lines that describe complex terrain features and optimize terrain representation. They can precisely depict the geometric shapes and structural details of the terrain. Accurate extraction of these lines is of great significance for enhancing the authenticity and reliability of digital terrain models. However, accurately and completely extracting topographic break lines from point cloud data has always been a core technical challenge in the field of digital terrain modeling. In response to common problems such as low completeness, insufficient accuracy, and severe over- and under-extraction in current methods for extracting topographic break lines based on ground point clouds, this paper proposes a topographic break line extraction algorithm that takes into account multiscale characteristics and anti-shrinkage fracture. The aim is to significantly improve the accuracy and robustness of topographic break line extraction.This method first addresses the issue of single-scale characteristics being insufficient for comprehensively capturing the complex terrain changes by using a random forest model based on multiple scales and features to capture potential terrain feature points. Then, potential feature points are classified into potential ridge and valley feature points by analyzing the point cloud contraction trend, and regional growth clustering with principal direction consistency constraints is used for denoising. Subsequently, the problem of feature lines breaking and endpoint contraction caused by traditional Laplacian smoothing is addressed by using the Laplacian with vertical constraints to smooth and refine the denoised potential ridge and valley feature points. Finally, under the constraint of connecting edge clipping rules to eliminate false connections, a minimum spanning tree is constructed to obtain high-quality topographic break lines.The validity and superiority of the method were verified by conducting experiments in two typical terrainsmining areas and mountainous regions. Quantitative and qualitative comparisons were made with three mainstream methods (LapS, D8, and PIM). Results showed that the proposed method was significantly superior to the comparison methods in terms of completeness, accuracy, and comprehensive quality indicators. In each complex terrain area, the completeness, accuracy, and quality of topographic break line extraction improved by at least 10.4%, 5.8%, and 11.8%, respectively. The study also evaluated the accuracy of constructing a digital elevation model by using the proposed topographic break lines as constraints. The experiments found that, under different point cloud densities, especially when the data density was lower than 10%, the DEM accuracy constrained by the topographic break lines was significantly better than that without constraints. The root mean square error and mean absolute error of the DEM increased by 55.9% to 82.9%, effectively maintaining the terrain details and overall shape.In conclusion, the proposed method, taking into account multiscale characteristics and anti-shrinkage fracture strategies, significantly improves the accuracy and completeness of topographic break line extraction. On this basis, it effectively supports the construction of high-fidelity digital elevation models, thereby providing a reliable technical solution for terrain feature analysis and high-precision digital terrain modeling in complex environments.关键词：airborne LiDAR;Topographic break lines;multiscale terrain characteristics;Laplacian smoothing with constraints;digital elevation model101|186|0更新时间：2026-03-17 - “遥感影像全色锐化领域迎来新突破，专家提出基于预融合策略结合Mamba与CNN的全色锐化方法，有效解决传统方法融合结果模糊或光谱失真问题，为提升全色锐化性能提供有效方案。”
摘要：Pansharpening of remote sensing images aims to produce a high-resolution multispectral (HRMS) image by integrating a low-resolution multispectral (LRMS) image with a high-resolution panchromatic (PAN) image. Most existing methods for pansharpening usually follow this pipeline: First, the LRMS image directly undergoes upsampling. Then, feature extraction and fusion are conducted to reconstruct the final HRMS image. However, this fusion strategy will lead to problems such as spectral distortion or blur in the fusion result. Furthermore, due to the lack of systematic feature interaction mechanism, these methods struggle to fully exploit the deep feature relationships between the PAN and LRMS images and between the global contextual information and local details. To address these issues, this paper proposes a pansharpening method based on pre-fusion strategy combined with Mamba and Convolutional Neural Network (CNN).To effectively integrate the merits of LRMR and PAN images, this paper proposes an approach that leverages the strong local feature capturing ability of CNN and the powerful long-range dependency representation capability of Mamba. Specifically, the Gaussian difference is first utilized to extract the high-frequency information of PAN image, enhancing the texture information of the LRMS image. Second, CNN and Mamba are employed to capture the local and global features from the PAN and enhanced LRMS images, respectively. Then, the CNN-based middle fusion branch is exploited to realize the full interaction and fusion of the extracted local and global information of the two modality images. Finally, the fused features are reconstructed to produce the high-quality HRMS image.The proposed method performs well in pansharpening. Experimental results on the QuickBird and IKONOS datasets show that our method qualitatively and quantitatively outperforms comparative methods such as the traditional GS, PCA, and advanced deep learning-based approaches, such as PanFormer and Pan-Mamba. Specifically, compared with the average values of the metrics of these competitors, the PSNR and UIQI of the proposed method are improved by 10.91% and 5.62%, respectively; the RMSE, ERGAS, and SAM of our method are reduced by 30.52%, 18.56%, and 60.41%, respectively. Furthermore, ablation studies verify the effectiveness of the pre-fusion strategy and the Mamba module designed in our proposed method.To address the challenges in fusing PAN and LRMS images, this paper proposes an improved deep fusion network called PMC-Net. Guided by a detail-enhanced pre-fusion network, the proposed method leverages the respective advantages of Mamba and CNN to collaboratively extract global and local features from input images, respectively. Experimental results on two remote sensing image datasets—QB and IK—demonstrate that the proposed PMC-Net significantly enhances multispectral image quality, outperforming the comparative methods in subjective visual evaluation and objective metric assessment. Although the proposed method demonstrates promising fusion performance, certain limitations remain, which can be addressed in future research. First, because of the lack of large-scale diversified remote sensing datasets, the generalization capability of the proposed model across different scenarios has not been fully validated. Second, the impact of imbalanced distribution of samples on fusion performance requires further investigation. Furthermore, the algorithm's computational efficiency and fusion effectiveness could still be improved.关键词：remote sensing image;pansharpening;pre-fusion;Gaussian difference;Mamb;state space model;convolutional neural network;multispectral;Panchromatic0|0|0更新时间：2026-03-17
Models and Methods
- Postal code：100190
- Tel：010-58887052 Email：nrsb@aircas.ac.cn
- Technical support is provided by Beijing Founder electronics co., LTD 京ICP备20021838号-9
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
0