最新刊期
卷 29 ， 期 8 ， 2025
- “在计算机视觉领域，航空遥感图像目标检测取得显著进展，专家系统性剖析旋转目标检测挑战，为技术发展提供新方向。”
摘要：Object detection mainly involves classification and regression. It is a basic task in computer vision and has been widely studied. Early research mainly focused on horizontal object detection in natural image scenes. In recent years, the development of Convolutional Neural Networks (CNNs) has facilitated the establishment of many general object detection methods. These methods have achieved good results on natural images, which greatly promoted the advancement of object detection tasks in aerial remote sensing images. Compared with natural images, aerial remote sensing images have complex backgrounds and generate objects that are densely distributed in arbitrary orientation. Therefore, traditional horizontal object detection is no longer suitable for object detection in aerial remote sensing images. According to the requirements of object detection tasks in aerial remote sensing images, the task of oriented object detection that relies on Oriented Bounding boxes (OBBs) to detect oriented objects accurately has gradually emerged. Most oriented object detection relies on the horizontal proposal/anchor of mainstream horizontal object detection frameworks to predict the OBBs. Although substantial progress has been made in performance, some fundamental flaws still exist, including the introduction of object-irrelevant information in the region features, as well as more complex rotation information calculation. At the same time, traditional CNNs cannot explicitly model the orientation variation of objects, which seriously affects the detection performance. Most existing oriented object detectors are based on horizontal object detectors by introducing an additional channel in the regression branch to predict the angle parameters. Angle regression-based methods have shown promising results, but they still encounter certain fundamental limitations. Compared with horizontal object detectors, angle regression detectors introduce new problems, mainly including 1) inconsistency between metrics and losses, 2) boundary discontinuity, and 3) square-like problems. At the same time, small objects in arbitrary orientation present great challenges to existing detectors, especially small oriented objects with extreme geometric shapes and limited features, which can lead to serious feature mismatch. These challenges have attracted widespread attention and prompted in-depth consideration from researchers in the relevant fields. To further promote the development of oriented object detection, this study mainly summarizes and analyzes the research status of oriented object detection in aerial remote sensing images. Currently, various pipelines exist for object detection, with the most common approach adding an angle output channel to the regression branch to predict the OBBs. As a result, many oriented object detectors are built upon the horizontal object detection framework. This study begins by reviewing key representative horizontal object detectors. With the development of object detection, many well-designed oriented object detection methods have been proposed and exhibit good performance. The main challenges encountered in current oriented object detection research are summarized in this study into five aspects: 1) feature alignment in object detection, 2) inconsistency between metric and loss, 3) boundary discontinuity and square-like problems, 4) low recognizability of small objects, and 5) label annotation problem. This study provides a comprehensive analysis of these challenges, introduces methods to address them, and discusses representative solutions in detail. The limitations and shortcomings of existing methods are examined, with potential directions for future exploration being considered. In the field of oriented object detection in aerial remote sensing images, several widely used and representative public benchmark datasets with OBB annotations are comprehensively summarized and introduced. The experimental results and visualizations of representative state-of-the-art detectors are compared and analyzed using the commonly employed datasets, namely, DOTA, HRSC2016, DIOR-R, and STAR. Current one-stage detectors are simple and effective, with anchor-free object detectors demonstrating comparable detection accuracy to traditional two-stage detectors. On this basis, this study integrates the current research status of object detection in aerial remote sensing images and anticipates future research trends. Future research could enhance the detection accuracy of oriented object detection while optimizing model complexity by developing anchor-free oriented object detectors, extracting rotation-invariant features, and minimizing annotation costs through weak supervision. Through this review, we aim to provide a valuable reference for researchers interested in exploring oriented object detection in aerial remote sensing images.关键词：object detection;aerial remote sensing images;Oriented Object Detection;convolutional neural networks;oriented bounding box165|385|0更新时间：2025-08-29 - “跨视角图像地理定位研究取得新进展，专家构建分类体系，为提升定位精度提供数据基础。”
摘要：Cross-view image geo-localization aims to retrieve the most similar image from a reference database through matching images captured from different viewpoints, which subsequently leverages the GPS tag of images to fulfill localization tasks. Traditional single-view image geo-localization is limited by factors such as dataset quality, scale, and positioning accuracy. Therefore, in recent years, numerous researchers and institutions have released multiple cross-view geo-localization datasets, which lay the data foundation for improving geo-localization accuracy. However, systematic analysis of these cross-view image geo-localization datasets is still lacking. Therefore, this study aimed to provide a comprehensive review of published cross-view image geo-localization datasets. Based on the literature review, we collect and organize 32 cross-view image geo-localization datasets spanning from 2011 to 2024. We review 32 classic datasets that have emerged since the development of cross-view image geo-localization, which constructs a classification system from four dimensions: viewpoint information, construction type, authenticity, and temporal information. We summarize the basic information of these datasets in a tabular form, including the name, image resolution, data scale, and encompassed scenes of the dataset. Fully expressing the fundamental attributive characteristics of cross-view geolocation datasets from multiple perspectives. Then, we delve into these cross-view image geo-localization datasets from five aspects: metadata, influence, keywords, acquisition sources, and application fields. In addition, we collate and summarize the mainstream algorithms for cross-view image geo-localization (e.g., network structure optimization, loss function optimization, and attention mechanism). Finally, we discuss the future development directions of cross-view localization datasets from four perspectives: the trend of multimodal datasets, the approach of large language models, image distraction handling, and model optimization. In summary, we offer a comprehensive review of cross-view image geo-localization datasets from various perspectives. To the best of our knowledge, this study is the first to review such datasets in the field, which can provide a reference for researchers in related fields. However, the current datasets still involve issues such as low data quality, single source of data, and weak generalization ability. Thus, further research is needed.关键词：cross-view;image geo-localization;datasets;deep learning;unmanned aerial vehicle;image retrieval;image matching;computer vision80|527|0更新时间：2025-08-29
Reviews
- “在云—气溶胶研究领域，新提出的2D-SMA算法通过统计概率模型和二维层次检测窗口，显著提高了层次检测的准确性和可靠性。”
摘要：Spaceborne lidar is a unique approach for the research and monitoring of clouds and aerosols due to its ability to observe their vertical properties. The Cloud–Aerosol Lidar with Orthogonal Polarization (CALIOP) onboard the CALIPSO satellite has been operational in orbit for many years and has provided a large amount of profiling measurements of clouds and aerosols. Detecting the spatial locations of cloud and aerosol layers from lidar data is a prerequisite for accurately retrieving and extracting layer information. While the official CALIOP algorithm detects layers using empirical thresholds, many layers are missed in its output, which may explain the systematic underestimation of aerosol optical depth offered by CALIPSO compared with that by the Moderate Resolution Imaging Spectroradiometer. Current hypothesis testing methods represented by the 1D-Simple Multiscale Algorithm (1D-SMA) determine whether a given signal belongs to a layer by verifying whether it conforms to the distribution hypothesis of background air. These methods eliminate the traditional empirical threshold array and improve the accuracy of layer detection. However, none of the methods above have considered the spatial continuity of the layer signals in the 2D vertical profiling scene, and missed layers still occur in their cases.This study further proposes a 2D-Simple Multiscale Algorithm based on Bernoulli probability distribution, which replaces the empirical threshold with a statistical probability model. For a background atmosphere bin, the probabilities that its signal intensity is greater or less than the ideal value are both 1/2. If the signals at each bin are independent, then the distribution of the number of signal bins within the detection window that are greater or less than ideal value follows a multiple Bernoulli experiment. We design a probability of belonging to a layer based on the signal intensity of all the bins in the detection window, which reflects their overall deviation from the ideal background atmosphere bins. The new algorithm marks that the center bin of the detection window as layer bin when the probability is small to a certain extent, which means it deviates far from background atmosphere. We use 2D layer detection windows, which cover multiple profiles at different horizontal resolutions, to utilize the spatial correlation of signals on adjacent profiles.Statistical comparing experiment based on the continuous global observations of CALIOP in December, 2017 shows that, at full horizontal resolution (5—80 km), the new algorithm detects 50.45% and 32.45% more layer area than the official CALIOP algorithm and 1D-SMA, respectively. When the new algorithm is applied at a horizontal resolution of 5—20 km, it achieves a comparable or greater area of detected layers than that of the official algorithm at 5–80 km horizontal resolutions. Moreover, this study demonstrates the reliability of the layers identified by the new algorithm by evaluating the depolarization ratio of ice clouds.In general, the new algorithm effectively reduces the missing of weak layers compared with the official CALIOP algorithm. The new algorithm is also simple and easy to implement, with certain research potential and application prospects.关键词：remote sensing;spaceborne LiDAR;CALIOP;cloud and aerosol;layer detection;Multiscale64|190|0更新时间：2025-08-29 - “星载GNSS-R与散射计联合探测能力研究取得新进展，为海陆参量反演研究提供新思路。”
摘要：This study aims to explore the combined detection capabilities of spaceborne GNSS-R (Global Navigation Satellite System Reflectometry) and scatterometers for remote sensing applications. Specifically, we focus on the enhancement effect of integrating the two technologies, which differ in their scattering mechanisms and observation modes, on the accuracy of ocean and land surface parameter retrieval.Two experiments were conducted in two typical scenariosocean surface wind speed and soil moisture retrieval. GNSS-R, which provides unique forward-scattered signals with high temporal and spatial resolution, and scatterometer, which is an active microwave sensor that receives backward-scattered signals, were compared individually and in combination. The evaluation was based on their operational characteristics, particularly their complementary features, including forward versus backward scattering and active versus passive observation modes.The results show significant improvements when combining the two technologies. In near-real-time applications, the combination reduced the Root Mean Square Error (RMSE) of sea surface wind speed retrieval by at least 13% compared with single-source sensors. In large-scale applications, the high-precision benefits of the combination extended to a broader coverage area, which reduced the RMSE for soil moisture retrieval by at least 6%.This study demonstrates that integrating spaceborne GNSS-R and scatterometer data can substantially improve the accuracy of remote sensing measurements, specifically in near-real-time and large-scale scenarios. Future research should further optimize observation geometry, polarization combinations, and band selections to enhance retrieval methods for ocean and land surface parameters.关键词：spaceborne GNSS-R;scatterometer;forward scattering;backward scattering;combined retrieval;ocean surface wind speed;soil moisture;DDM;reflectivity;NRCS0|0|0更新时间：2025-08-29 - “记者从最新研究中获悉，一种基于BPNN的水深反演模型在中国涠洲岛和美国莫洛凯岛海域验证了其高精度和可靠性，有效支持浅海测深。”
摘要：Bathymetric maps with high spatial resolution can display topographic details and provide data support for maritime navigation, coastline management, and marine resource utilization and development.This study conducted experiments in the sea areas of Weizhou Island in China and Molokai Island in USA. With the support of Sentinel-2 and Landsat-9 images, a water depth inversion method was proposed first, which incorporates geographic location features as modeling elements. Then, an optimal water depth inversion model based on a Back Propagation Neural Network (BPNN) was constructed. Finally, different remote sensing data were used to perform accuracy tests of the inversion method proposed in this paper in various sea areas.During model selection, machine learning models consistently outperformed empirical models across all accuracy metrics. The BPNN model exhibits the highest modeling accuracy in machine learning models. In addition, the machine learning model is more stable, its inversion of the water depth map can better invert the actual water depth change in the experimental area, and its inversion image is smoother. The introduction of geographic location features can significantly improve the accuracy of water depth inversion. Experimental results have shown that the inversion accuracy in the Weizhou Island was improved from an R2 value of 0.7666 to 0.9952, and the RMSE was reduced from 2.5016 m to 0.3578 m. As a validation experiment, the R2 value in the Molokai Island area was 0.9939, and the RMSE decreased from 3.0165 m to 1.0189 m. At the same time, the introduction of geographic location features can also eliminate the influence of some clouds and fog on remote sensing images, with more accurate water depth inversion results being generated.The conclusion demonstrated that, using all bands of the image for modeling, the inversion of the water depth map is smoother. This method can better invert regional bathymetry trends, with fewer outliers and more accurate inversion results. After geographic location features were incorporated, the addition of vegetation index features did not yield better results. Instead, it slightly decreased the modeling accuracy of the model. Therefore, analyzing the autocorrelation between each element and making comprehensive decisions on modeling factors are important. In summary, the water depth inversion model constructed in this study has high accuracy, strong reliability, and good portability. It can be effectively used to measure shallow sea depth.关键词：optical remote sensing;Offshore waters;Geographic location characterization;BPNN model;Weizhou Island;Molokai Island;accuracy validation167|1228|0更新时间：2025-08-29 - “记者从新疆阜康煤田火区报道，研究人员开发了新算法，有效检测煤火甲烷排放，为煤火自燃早期识别提供新思路。”
摘要：Coal-rich regions such as Xinjiang, Ningxia, and Inner Mongolia frequently experience spontaneous combustion of coal seams. This phenomenon releases large quantities of methane, which is a high-impact greenhouse gas. The unorganized and diffuse nature of these emissions introduces great challenges for detection and quantification, which often contribute to the so-called “missing carbon” sector in greenhouse gas inventories. Owing to the limitations of satellite resolution and the inadequate adaptability of existing methane detection technologies in rugged terrains, this study focuses on coal fire areas characterized by unorganized emissions. Using wide-swath ground-based hyperspectral imagery, we developed a novel detection methodology for methane emissions from individual coal fire sources in diverse terrains. The goal is to analyze methane escape patterns and assess latent risks of underground coal fires, which offer a new framework for early-warning systems.In this work, we investigated methane escape in two representative mountain coal fire areas located in Fukang, Xinjiang, using ground-based hyperspectral wide-swath imagery collected in June 2023. Seven distinct algorithms were applied: first, we proposed a Modified Least Squares Image Enhancement (MLSIE) algorithm by applying the L2-norm least squares regression principle to methane-sensitive spectral windows (1.66 and 2.3 μm) in the SWIR range. Second, we developed the Methane Ratio Derivative Spectral Unmixing (RDSU) algorithm, which is represented by RCH4I1 and RCH4I3. These algorithms incorporate the spectral ratio derivative unmixing technique and the spectral sensitivity of methane to suppress background interference from pseudo-coal fire regions while enhancing the spectral signature of methane. Third, by integrating factors such as cloud shadow detection, mineral content, combustion characteristics, and building-related features, we developed two methane ratio indices: one is 1DSRCH4I3, which enhances the contrast between methane and artifacts; the other is 2DSRCH4I3, which mitigates artifact interference and highlights methane-enriched areas.Our evaluation demonstrated the following results. (1) The proposed MLSIE, RDSU, and DSRCH4I algorithms significantly improved methane detection accuracy compared with the existing CH4I algorithm. (2) The 2DSRCH4I3, MLSIE (2.3 μm), and RCH4I1 algorithms exhibited superior performance in complex terrains, while 2DSRCH4I and MLSIE (2.3 μm) algorithms were also effective in relatively simple mountainous coal fire areas. (3) MLSIE (2.3 μm) displayed robust generalization ability. Moreover, 2DSRCH4I3 effectively minimized artifacts and false positives, which led to superior detection accuracy. RCH4I1 also demonstrated clear detection efficacy in coal fire areas with significant methane leakage. (4) Methane plumes were detected in two distinct forms: one coexisting with combustion flames, and the other freely escaping into the surrounding atmosphere.This work presents a novel methodology for detecting methane emissions from high-temperature and panoramic coal fire sources with diverse geomorphic characteristics. The findings provide valuable technical support for the identification and assessment of underground coal fire risks. Furthermore, this work introduces a new approach to early-warning systems for coal fire disasters, with the use of methane escape as a diagnostic indicator of potential fire formation. However, we recognize that the influence of black carbon aerosols, which may interfere with mixed spectral signals, was not addressed. Future research could explore the dynamic quantification of methane plumes by integrating hyperspectral image spectral enhancement techniques to improve detection accuracy.关键词：coal fire;methane;Hyperspectral imaging;SWIR;Unorganized emissions;Plume detection;Artifact suppression;Greenhouse gas;climate change96|290|0更新时间：2025-08-29 - “最新研究进展表明，改进的卫星遥感反演算法在云微物理特性研究领域具有重要应用价值，为相关应用提供可靠数据基础。”
摘要：The retrieval of cloud particle effective radius (re) from satellite remote sensing is a critical technique for studying cloud microphysical properties and precipitation processes. It has significant applications in aerosol-cloud interactions, severe convective weather monitoring and early warning, and weather modification. Accuracy validation of the retrieved re is essential for these applications.Based on improved satellite retrieval algorithm for cloud particle effective radius using 3.7 μm channel data, this study derives cloud particle effective radius (re_o) from MODIS and AVHRR observations. The retrieved re_o are systematically compared with in-situ re measurements by aircraft from 22 cases of continental cumulus cloud, and the algorithm’s reliability and accuracy is evaluated.The comparisons show that the error of particle effective radius between the retrieval and airborne measurements is less than 2.4 μm, which is very close to 2 μm of international verification results within marine stratus. The distribution of re with temperature/height (vertical structure) is quite consistent with that detected by aircraft measurement. The retrieved re from the 3.7 μm has a high correlation with the airborne measurement, with a correlation coefficient of 0.79 and a linear fitting slope of 0.81. However, the re from MODIS cloud product has a low correlation with the airborne measurement, and the correlation coefficient and linear fitting slope are 0.43 and 0.32, respectively. All these results suggest high accuracy of the retrieved particle effective radius of clouds and the high reliability of the retrieved methodologies which demonstrate that improved algorithm can provide a reliable data foundation for applications.关键词：remote sensing;cloud particle effective radius;vertical structure;satellite retrieval;aircraft measurements;validation35|137|0更新时间：2025-08-29
Atmosphere and Ocean
- “在作物生长监测领域，专家提出了STFSR时空融合方法，有效提高了NDVI数据精度，为高时空分辨率植被指数数据生成提供新思路。”
摘要：Normalized Difference Vegetation Index (NDVI) images with fine spatial and temporal resolutions are important data for real-time precise vegetation monitoring. Remote sensing image, as an important data source for producing NDVI data, however, always present a trade-off between the spatial and temporal resolutions due to the limitation in the power of satellites. Generally, sensors with fine spatial resolution always have a long revisit time (e.g., Landsat images), while sensors with a short revisit period always have coarse spatial resolution (e.g., Moderate-resolution Imaging Spectroradiometer (MODIS) images). Spatio-temporal fusion technique can be applied to generate NDVI images with both fine spatial and temporal resolutions, by fusing NDVI images acquired from these two categories of sensors. The existing spatio-temporal fusion methods, however, suffer from a long-standing challenge, that is, the NDVI change between the images at the known and prediction times, which restricts the accuracy of spatio-temporal fusion prediction greatly. In this paper, a Spatio-Temporal Fusion Then Spatial Reconstruction (STFSR) method was proposed to cope with the NDVI change issue in predicting the 30 m Landsat NDVI images.Generally, when predicting the missing Landsat NDVI image by spatio-temporal fusion, a pair of spatially complete fine and coarse spatial resolution NDVI images is also required (probably temporally far from the prediction time). Except for the original auxiliary images, the proposed STFSR method also included the fine spatial resolution image temporally closer to the prediction time, but with different degrees of data loss caused by cloud cover (hereafter, simplified as auxiliary cloudy NDVI image) in prediction. The implementation of STFSR is divided into two steps: (1) Reconstructing the non-cloud area (the corresponding non-cloud area in the auxiliary cloudy image, but in the image to be predicted) using the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM). (2) Reconstructing the cloud area (the corresponding cloud area in the auxiliary cloudy image, but in the image to be predicted) by a Spatial-Temporal Random Forest (STRF) algorithm, a spatial reconstruction method integrating the information from both fine and coarse spatial resolution NDVI images.In the experiments in three regions, the effectiveness of the proposed STFSR method was evaluated, by comparing with two commonly used spatio-temporal fusion methods, the STARFM and the Spatial Weighting-based Virtual Image Pair-based Spatio-Temporal Fusion (VIPSTF-SW) algorithms. The results demonstrate that the proposed STFSR can produce greater accuracy than the other two methods for all three regions. Furthermore, when the cloud coverage increases to a certain percentage (e.g., 80%) in the auxiliary cloudy image, the STFSR method can still provide a more satisfactory prediction than two benchmark methods. Specifically, the average Root Mean Square Error (RMSE) of STFSR is 0.0217 and 0.0188 smaller than that of STARFM and VIPSTF-SW, respectively. The corresponding average Correlation Coefficient (CC) is 0.0820 and 0.0742 larger, and the corresponding average Relative Global-dimensional Synthesis Error (ERGAS) is 4.3170 and 3.8535 smaller.The proposed STFSR method takes full advantage of the important information in the cloudy, but temporally closer NDVI image, which fails to be utilized in existing spatio-temporal fusion methods. Generally, the proposed STFSR method provides a flexible solution to deal with the NDVI change in spatio-temporal fusion. Moreover, this model has great potential for the generation of other vegetation index data with fine spatial and temporal resolutions, such as the Enhanced Vegetation Index (EVI) and the Leaf Area Index (LAI).关键词：remote sensing;Landsat;MODIS;NDVI;spatio-temporal fusion;spatial reconstruction159|297|0更新时间：2025-08-29 - “在遥感影像领域，研究者提出了一种光学和SAR图像鲁棒匹配方法，通过深度特征重构增强，有效提高了正确匹配率。”
摘要：The automatic precise registration between optical and Synthetic Aperture Radar (SAR) imagery remains a major challenge in remote sensing due to fundamental differences in their imaging mechanisms. The inherent modality gap manifests as substantial radiometric discrepancies (speckle noise vs. photometric consistency) and geometric distortions (side-looking geometry vs. nadir projection), which introduce critical obstacles for conventional feature matching approaches. While existing deep learning-based methods have progressed in extracting deep features, most architectures inadequately address two crucial aspects: multiscale feature fusion across different imaging characteristics and cross-modal invariant feature representation. This insufficiency leads to compromised robustness in complex geographical scenarios. To address this concern, we propose a robust matching method based on deep feature reconstruction for optical and SAR images. Our method features a pseudo-Siamese network that integrates multiscale deep features and image reconstruction. First, a multiscale feature extraction architecture efficiently obtains multiscale deep features at the pixel level. This architecture allows the network to capture detailed information from various scales, which is crucial for understanding the complex patterns present in remote sensing images. Second, a pseudo-SAR translation branch for optical images is designed to reconstruct images from deep features, which enhances the ability of the network to learn robust common features. This branch mimics the characteristics of SAR images. This feature enables the network to find shared features between the two image types more effectively. Through this translation process, the network learns to focus on the essential elements that are common to optical and SAR images, which improves the matching accuracy. Finally, a joint loss function based on multilayer feature matching similarity and reconstructed image average error is constructed for robust matching. This loss function ensures that the network not only matches features accurately across different layers but also maintains a high degree of fidelity in reconstructing the original images. By combining the two aspects, the network can achieve a balance between feature similarity and image reconstruction quality, which leads to more reliable matching results. Experiments on two remote sensing image datasets with different resolutions and diverse terrain scenes (urban, suburban, desert, mountain, and water) show that our method outperforms several state-of-the-art matching methods in correct matching rate. The proposed method demonstrates superior performance in various environments, which indicates its versatility and effectiveness in real-world applications. The ability to handle different resolutions and terrain types is particularly important for practical remote sensing tasks, where conditions can vary widely.This research advances cross-modal image analysis in two aspects: (1) a new paradigm combining feature learning with cross-domain translation is provided, and (2) practical solutions for SAR-optical registration in challenging environments are proposed.关键词：remote sensing;optical image;SAR image;image matching;deep learning104|368|0更新时间：2025-08-29 - “在遥感领域，对比学习算法挖掘高光谱图像关键特征，本文提出多尺度监督对比学习网络，实现精准分类。”
摘要：Hyperspectral image classification, which aims to assign a belonging category to each pixel in a hyperspectral image, is an important application in remote sensing. In recent years, contrastive learning has been widely used in hyperspectral image classification tasks due to its excellent capability to mine key features of data. However, most of the current self-supervised contrastive learning paradigms use a two-stage scheme to train the network. Moreover, they often define objects in the same class as negative samples in the pre-training stage, which usually leads to a wider intra-class gap. In addition, contrastive learning algorithms generally use data enhancement methods such as cropping and rotating to generate positive samples, and the diversity of generated positive samples is more limited. In this study, a hyperspectral image classification network based on multiscale supervised contrastive learning is proposed to solve the abovementioned problems. The method aims to extract multiscale spatial and spectral features level by level, along with adaptive fusing of the features to generate the final results.This study proposes a Multiscale Supervised Contrastive Learning Network (MSCLN) for hyperspectral image classification, which includes two parts: a multiscale contrastive feature learning network and a spatialspectral hybrid probability-directed fusion classification network. In the multiscale contrast feature learning network, a spectral-guided branch and a spatial feature extraction branch that introduces an attention mechanism are designed to extract spectral–spatial features level by level. Then, two multiscale spatial features of the same object are constructed as positive samples by introducing label information. Specifically, 2n views can be generated for n objects, in which all views of the same type of objects are positive samples of each other and the rest are negative samples. Finally, in the spatial–spectral hybrid probability-directed fusion classification network, the learnable parameters are set to integrate the spectral–spatial features to obtain the final classification probability. By co-training the two networks, more accurate classification results can be obtained.In three public hyperspectral datasets, Houston 2013, WHU-Hi-LongKou, and Pavia University, 50, 80, 50 labeled samples were randomly selected from each category as training sets, respectively. The overall classification accuracy of the proposed algorithm reached 96.20%, 99.20%, and 98.96% for the three datasets, respectively. The classification performance was also better than that of the other comparison methods.The method extracts the discriminative spectral–spatial features hierarchically by MSCLN. Then, the labeling information is introduced to construct the two multiscale spatial features of the same object as positive samples. This way makes the same type of sample distance more aggregated while pushing away the inter-class distance. Finally, adaptive fusion of spectral–spatial features is realized to obtain an excellent classification map.关键词：remote sensing;hyperspectral images;image classification;contrastive learning;spatial spectral feature fusion;attention mechanism336|661|0更新时间：2025-08-29 - “最新研究提出了一种无监督的基于双支路生成对抗网络与Transformer的多光谱与全色遥感图像融合方法，有效提升了融合图像的光谱信息和空间分辨率。”
摘要：Multispectral remote sensing images contain rich spectral information that can reflect ground features. However, the spatial resolution of these images is low, and their texture information is relatively insufficient. By contrast, panchromatic remote sensing images have high spatial resolution and rich texture information, but they lack rich spectral information that can reflect ground features. In practice, two types of images can be integrated into a single one to obtain the complementary advantages from the different images. Accordingly, the fused image can better meet the needs of downstream tasks. To this end, this study proposes an unsupervised method for fusing panchromatic and multispectral images using a dual-branch generative adversarial network combined with Transformer.Specifically, the source images (panchromatic and multispectral images) are first decomposed into base and detail components using guided filtering, where the base component mainly focuses on the main body of the source image. Moreover, the detail component mainly represents the texture and detail information of the source image. Next, the decomposed base components of the panchromatic and multispectral images, along with the decomposed detail components of the two types of source images, are concatenated. Then, the concatenated base and detail components are input into the base and detail branches of the dual-branch generator, respectively. Thereafter, according to the different characteristics of the base and detail components, the Transformer and CNN are utilized to extract the global spectral information from the base branch and the local texture information from the detail branch. Subsequently, the model is continuously trained in an adversarial manner between the generator and the dual discriminators (base layer discriminator and detail layer discriminator). Finally, the fused image with rich spectral information and high spatial resolution is obtained. Extensive experiments on the public dataset demonstrate that the proposed method outperforms the state-of-the-art methods in terms of qualitatively visual effects and quantitatively evaluated metrics.This study proposes an unsupervised method for fusing panchromatic and multispectral remote sensing images using a dual branch generative adversarial network combined with Transformer. The superiority of the proposed method was verified through qualitative and quantitative comparisons with multiple representative methods. The ablation studies confirm the effectiveness of the network structure designed in this study.关键词：remote sensing image fusion;Guided filtering;convolutional neural network;Generate adversarial network;transformer network;Basic layer;Detail layer;Panchromatic;multispectral540|1329|0更新时间：2025-08-29 - “在建筑物提取领域，研究者提出了一种双分支网络，有效解决了形状尺度多变和边界提取不准确的问题，为建筑物提取提供了新方法。”
摘要：Building extraction in Remote Sensing (RS) images is a key research focus in RS, which plays a vital role in various tasks, such as urban planning, illegal building detection, and natural disaster assessment. The rapid development of deep learning technology has facilitated its use in building extraction in RS images, which achieved significant extraction effect. However, building extraction in RS images has two main challenges: one is the difference in scales and shapes of buildings in RS images, and the other is the difficulty in accurately extracting building boundaries. This study proposes a two-branch building extraction network, which integrates main body and edge separation and multiscale information extraction, to address the aforementioned challenges. Both branches employ the U-Net architecture. First, a Main Body and Edge Separation Branch (MBESB) is designed for feature decomposition based on the decoupling idea and optical flow estimation technique. MBESB generates the main body and edge features of buildings separately, which enhances the ability of representing building boundaries. Then, a Lightweight Multiscale Information Extraction Branch (LMIEB) is constructed based on dilated convolution, depth-separable convolution, and attention mechanism to fully extract the differently scaled features of buildings. LMIEB introduces an Attention-enhanced Multi-scale Convolutional Module (AMCM) and integrates this module into the last three stages of the U-shaped network to harvest multi-scale information from feature maps at different levels. Finally, a loss function enhanced by main body and edge features is presented with the help of the features generated by MBESB and LMIEB to improve the training process of the building extraction network. The loss function consists of three components: a global loss based on overall building features, a main body loss focused on the building’s body features, and an edge loss emphasizing boundary details. Experiments were conducted using two datasets for public building extraction, namely, the Inria and WHU datasets, to evaluate the performance of the proposed MMT-Net method. Moreover, to further validate the effectiveness of the proposed method, extensive comparative experiments were carried out on subregions Tyrol, Austin, and Chicago of the Inria dataset. Five deep learning methods (ENet, SegNet, MMB-Net, Refine-UNet, and MAP-Net ) were used as the comparative methods. Quantitative analysis of building extraction results was conducted in terms of four evaluation metrics, namely, precision, recall, F1, and IoU. For the Inria and WHU datasets, the F1/IoU values of the proposed MMT-Net are 0.8894/0.8008 and 0.9567/0.9170, respectively, which are superior to those of the five comparative methods. Additionally, visualization results are shown on both Inria and WHU datasets. The findings indicate that the proposed method outperforms other five comparative methods in terms of both missed building detections and false building detections. Experimental results on two commonly used datasets for public building extraction show that the proposed building extraction network is feasible and effective. In addition, the results of the ablation experiments indicate that MBESB, LMIEB, and the loss function with auxiliary enhancement of the main body and edge features proposed in this work can effectively enhance the performance of building extraction.关键词：remote sensing image;building extraction;deep learning;U-Net;main body and edge separation;two-branch;multi-scale;lightweight241|336|0更新时间：2025-08-29
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