Unsupervised Change Detection (UCD) in Very High Resolution (VHR) Remote Sensing (RS) images remains to be a difficult challenge due to the inherent spatio-temporal complexity within data. Inspired by recent advancements in Visual Foundation Models (VFMs) and Contrastive Learning (CL), this research aims to develop CL methodologies to translate implicit knowledge in VFM into change representations, thus eliminating the need for explicit supervision. To this end, we introduce a Semantic-to-Change (S2C) learning framework for UCD in VHR RS images. Differently from existing CL methodologies that typically focus on learning multi-temporal similarities, we introduce a novel triplet learning strategy that explicitly models temporal differences, which are crucial to the CD task. Furthermore, random spatial and spectral perturbations are introduced during training to enhance robustness to temporal noise. In addition, a grid sparsity regularization is defined to suppress insignificant changes, and an IoU-matching algorithm is developed to refine the CD results. Experiments on three benchmark CD datasets demonstrate that the proposed S2C learning framework achieves significant improvements in accuracy, surpassing current state-of-the-art by over 31%, 9% and 23%, respectively. It also demonstrates robustness and sample efficiency, suitable for training and adaptation of various VFMs or backbone neural networks.

Cross-View Geo-Localization (CVGL) in remote sensing aims to locate a drone-view query by matching it to geo-tagged satellite images. Although supervised methods have achieved strong results on close-set benchmarks, they often fail to generalize to unconstrained, real-world scenarios due to severe viewpoint differences and dataset bias. To overcome these limitations, we present VFM-Loc, a training-free CVGL framework that leverages the generalizable visual representations from vision foundational models (VFMs). VFM-Loc identifies and matches discriminative visual clues across different viewpoints through a progressive alignment strategy. First, we design a hierarchical clue extraction mechanism using Generalized Mean pooling and Scale-Weighted R-MAC to preserve distinctive visual clues across scales while maintaining hierarchical confidence. Second, we introduce a statistical manifold alignment pipeline based on domain-wise PCA and Orthogonal Procrustes analysis, linearly aligning heterogeneous feature distributions in a shared metric space. Experiments demonstrate that VFM-Loc exhibits high accuracy on standard benchmarks and surpasses supervised methods by over 20% in Recall@1 on the challenging LO-UCV dataset with large oblique angles. This work highlights that principled alignment of pre-trained features can effectively bridge the cross-view gap, establishing a robust and training-free paradigm for real-world CVGL. The relevant code is made available at: github.com/DingLei14/VFM-Loc.

Remote sensing image object detection is currently a research hotspot, but its performance still faces fundamental challenges: existing methods have not fully exploited the unique prior knowledge of remote sensing scenes, such as the structure and distribution patterns of ground objects. At the same time, multiscale objects and high-resolution data in images also bring computational burdens. To this end, this article selects YOLOv12-N, which achieves an excellent balance between detection performance and computational efficiency, as the baseline model and proposes a new object detector for RSI—Multicore Convolutional Intersection of Union (IoU) Network (MCI-Net), which includes multiple components, such as Multicore Attention Framework (MAF), convolutional attention synthesis (CAS) and the Shape_IoU. MAF enhances feature representation and improves detail capture ability by introducing multiscale and large kernel attention. CAS optimizes feature processing, reduces computational and memory requirements, and improves model efficiency. Shape_IoU loss function not only considers the overlapping area between the predicted box and the true box but also pays extra attention to the shape and scale characteristics of the bounding box itself, thereby optimizing detection performance. MCI-Net performs better than the baseline model on the Satellite Imagery Multivehicles Dataset (SIMD), Dense Image Overlapping Regions (DIOR), VHR-10 and DOTA datasets, with improvements of 6.2%, 2.2%, 3.5%, 2.9% on the SIMD, 1.7%, 0.6%, 1.3%, 1.3% on the DIOR dataset, and 0.3%, 0.1%, 0.1%, and 0.1% on the VHR-10 dataset, and 1.9%, 2.7%, 2.2%, 2.7% on the DOTA dataset.

High-resolution video satellites enable large-view dynamic monitoring for earth observation. Among satellite video interpretation techniques, multi-object tracking (MOT) receives growing attention for its foundational role. Rigid targets in satellite videos exhibit strong inter-frame appearance and posture consistency, showing quasi-linear trajectories with constrained displacements. Inspired by these kinematic characteristics, this paper introduces an online end-to-end Propagating Spatio-temporal State and Progressively Associating Trajectory MOT (PS2PAT-MOT) framework. It consists of a detection branch for locating multi-category and multi-target objects in current frame, and a correlation branch using an inter-frame spatio-temporal state propagation (STSP) module to propagate location and appearance information and encode same-target correlations between adjacent frames. Detection and correlation outputs from both branches undergo affinity calculation, while the Progressively Associating Trajectory (PAT) strategy generates continuous tracklets using differentiated association thresholds for distinct trajectory segments. Experimental results on two publicly available AIR-MOT, SAT-MTB, and a self-built LV-SatMOT (Large-View Satellite video MOT) datasets demonstrate the effectiveness of the proposed PS2PAT-MOT framework. Codes are available at: https://github.com/HELOBILLY/PS2PAT-MOT.

The task of visual geo-localization based on street-view images estimates the geographical location of a query image by recognizing the nearest reference image in a geo-tagged database. This task holds considerable practical significance in domains such as autonomous driving and outdoor navigation. Current approaches typically use perspective street-view images as reference images. However, the lack of scene content resulting from the restricted field of view (FOV) in such images is the main cause of inaccuracies in matching and localizing the query and reference images with the same global positioning system (GPS) labels. To address this issue, we propose a perspective-to-panoramic image visual geo-localization framework. This framework employs 360° panoramic images as references, thereby eliminating the issue of scene content mismatch due to the restricted FOV. Moreover, we propose the structural feature enhancement (SFE) module and integrate it into LskNet to enhance the ability of the feature extraction network to capture and extract long-term stable structural features. Furthermore, we propose the adaptive scene alignment (ASA) strategy to address the issue of data capacity and information content asymmetry between perspective and panoramic images, thereby facilitating initial scene alignment. In addition, a lightweight feature aggregation module, MixVPR, which considers spatial structure relationships, is introduced to aggregate the scene-aligned region features into robust global feature descriptors for matching and localization. Experimental results demonstrate that the proposed model outperforms current state-of-the-art methods and achieves R@1 scores of 72.5% on the Pitts250k-P2E dataset and 58.4% on the YQ360 dataset, indicating the efficacy of this approach in practical visual geo-localization applications.

Synthetic aperture radar images have all-weather and all-time capabilities and are widely used in the field of ship target surveillance at sea. However, its detection accuracy is often limited by factors such as complex sea conditions, diverse ship scales, and image noise. Aiming at the problems such as inconsistent scale of ship target detection in SAR images, difficulty in detecting small targets, and interference from complex backgrounds, this paper proposes a ship detection method for SAR images based on band recombination and multi-scale feature enhancement. Firstly, aiming at the problem that the single-channel replication mode adopted by the deep neural network cannot fully extract the ship target information in SAR images, a band recombination method was designed to enhance the ship information in the images. Furthermore, the coordinate channel attention and bottleneck Transformer attention mechanisms are introduced in the backbone part of the network to enhance the network's representation ability of the target spatial distribution and maintain the global feature modeling ability. Finally, a multi-scale feature enhancement and multi-scale effective feature aggregation module was designed to improve the detection accuracy of multi-scale ships in wide-format images. The experimental results on the LS-SSDD and HRSID datasets show that the average accuracies of the method proposed in this paper reach 78.1% and 94.5% respectively, which are improved by 6.9% and 0.8% compared with the baseline model, and are superior to other advanced algorithms, verifying the effectiveness of the method proposed in this paper. Meanwhile, the algorithm proposed in this paper has also demonstrated good performance in wide-format SAR images of actual large scenes. The method proposed in this paper effectively improves the problems of missed detection and false detection of small-target ships in SAR images of large scenes. At the same time, it enhances the efficiency of rapid and accurate detection in large scenes and can provide reliable technical support for the field of maritime target surveillance.

Change detection (CD) in multitemporal remote sensing imagery presents significant challenges for fine-grained recognition, owing to heterogeneity and spatiotemporal misalignment. However, existing methodologies based on vision transformers or state-space models typically disrupt local structural consistency during temporal serialization, obscuring discriminative cues under misalignment and hindering reliable change localization. To address this, we introduce ChessMamba, a structure-aware framework leveraging interleaved state-space modeling for robust CD with multi-temporal inputs. ChessMamba integrates a SpatialMamba encoder with a lightweight cross-source interaction module, featuring two key innovations: (i) Chessboard interleaving with snake scanning order, which serializes multi-temporal features into a unified sequence within a single forward pass, thereby shortening interaction paths and enabling direct comparison for accurate change localization; and (ii) Structure-aware fusion via multi-dilated convolutions, selectively capturing center-and-corner neighborhood contexts within each mono-temporal. Comprehensive evaluations on three CD tasks, including binary CD, semantic CD and multimodal building damage assessment, demonstrate that ChessMamba effectively fuses heterogeneous features and achieves substantial accuracy improvements over state-of-the-art methods.The relevant code will be available at: github.com/DingLei14/ChessMamba.

Unsupervised Change Detection (UCD) in multimodal Remote Sensing (RS) images remains a difficult challenge due to the inherent spatio-temporal complexity within data, and the heterogeneity arising from different imaging sensors. Inspired by recent advancements in Visual Foundation Models (VFMs) and Contrastive Learning (CL) methodologies, this research aims to develop CL methodologies to translate implicit knowledge in VFM into change representations, thus eliminating the need for explicit supervision. To this end, we introduce a Semantic-to-Change (S2C) learning framework for UCD in both homogeneous and multimodal RS images. Differently from existing CL methodologies that typically focus on learning multi-temporal similarities, we introduce a novel triplet learning strategy that explicitly models temporal differences, which are crucial to the CD task. Furthermore, random spatial and spectral perturbations are introduced during the training to enhance robustness to temporal noise. In addition, a grid sparsity regularization is defined to suppress insignificant changes, and an IoU-matching algorithm is developed to refine the CD results. Experiments on four benchmark CD datasets demonstrate that the proposed S2C learning framework achieves significant improvements in accuracy, surpassing current state-of-the-art by over 31%, 9%, 23%, and 15%, respectively. It also demonstrates robustness and sample efficiency, suitable for training and adaptation of various Visual Foundation Models (VFMs) or backbone neural networks. The relevant code will be available at: github.com/DingLei14/S2C.

Objective With the continuous growth of the population and the rapid development of the global economy, increasing human activities are driving land- cover utilization changes. Timely and accurate understanding of these changes is crucial for national economic construction, social development, and ecological protection. The use of multi- temporal remote sensing images to detect land cover changes, continuously update national land survey results, and maintain the accuracy and current status of basic geographic information is essential for intelligent change detection methods. However, existing land cover change detection is susceptible to the influence of light and seasonal variations, leading to pseudo- changes and misdetection or omission in change detection results. To address this, we design a remote sensing image change detection method based on adaptive boundary sensing. Convolutional neural networks (CNNs) excel at extracting local features, while Transformer is more advantageous in global feature extraction. Our method adopts a hybrid CNN and Transformer structure for feature extraction, combining edge information to enhance change detection sensitivity, providing more accurate results and improving the model resistance to external conditions such as light and seasonal interference. Methods During the encoding stage, res2net is employed as an encoder to extract multiscale features and enhance variation features through a difference enhancement module, reducing redundant feature interference. In the decoding stage, a boundary extractor using deformable convolution obtains precise semantic boundary features. These edge features guide the Transformer for contextual information aggregation. Finally, a multi- scale fusion output strategy integrates different scale feature maps, adding multiple connections between decoders of varying levels to fuse low-level spatial information with high-level semantic information, achieving contextual information aggregation, generating the predicted change map, and completing the change detection task. Results and Discussions To validate our method's effectiveness, experiments are conducted on two public datasets: (1) the CLCD dataset, comprising 600 cropland change sample image pairs collected by Gaofen-2 satellites over Guangdong Province in 2017 and 2019, with resolutions ranging from 0.5 to 2 m; (2)the RSCD dataset is publicly from the 2022 Aerospace Hongtu Cup Remote Sensing Image Intelligent Processing Algorithm Competition, consisting of 3000 image pairs from Gaofen-1 and Gaofen-2 with 0.8 m to 2 m resolution. On these two datasets, our method achieves F1 scores of 72.82 degrees o and 58.96 degrees o, respectively. Meanwhile, visualization results also indicate better performance in recognizing both small and large area changes, with continuous boundaries and complete detection areas. Our method's change maps closely match actual outcomes, accurately detecting changing areas' spatial locations. This demonstrates that the edge- guided context aggregation proposed herein enhances the interaction between local detail and global semantic features during Transformer coding and decoding, improving detection efficacy. Compared with seven classical change detection methods on two datasets, our method outperforms the selected comparison methods. Ablation studies on the CLCD dataset further confirm the effectiveness of each module in enhancing overall performance. Conclusions Addressing boundary discontinuity and misdetection issues in land cover change detection of high- resolution remote sensing images, we design an adaptive boundary sensing method, which adopts a hybrid structure of CNN and Transformer. Selecting res2net as the encoder for multiscale feature extraction and differential enhancement, and leveraging edge features to guide Transformer for contextual information aggregation, we adopt a multi- scale output fusion strategy to combine global semantic and local detail features across layers. This approach yields more precise change detection results compared to other traditional methods, enhancing the model's resilience to external condition interferences.

Remote sensing object detection has important application value in fields such as environmental monitoring and resource detection and analysis. However, the current universal object detectors are not very effective in detecting remote sensing objects. To this end, this paper proposes an efficient, low-complexity and anchor-free remote sensing object detection framework YOLO-RSOD based on YOLOv7. First, an additional Tiny Object Head is proposed for better detection of micro-remote sensing objects. The original Head is then replaced with Decoupled Head (DH) to explore the detection potential of the decoupled detection head structure. Then the Explicit Vision Center (EVC) in the Centralized Feature Pyramid Network (CFP) is added to further improve the detection ability of remote sensing objects. Finally, this article also integrates a global attention module (GAM) to find attention areas in dense object scenes. Ablation experiments on the general remote sensing target detection dataset VisDrone2021 demonstrate the effectiveness of several modules introduced in this paper in remote sensing target detection. On the VisDrone2021 data set, YOLO-RSOD can achieve accuracy rates of 30.7

Recently the analysis of remotely sensed images has played a vital role in various aspects of research. The current researches ignore the unique prior knowledge in remote sensing images and do not consider exploring the contextual information of the object, while the existence of multi-scale and high image resolution of objects in remote sensing images also affects the accuracy of the object detection task. Based on the above problems, this paper proposes a object detector DCI-Net (Dynamic Context-Aware IoU Network) based on remote sensing images, in which the proposed CASK (Context-Aware Selective Kernel) module can explicitly model the interdependence between the convolutional feature channels. A loss function Pi_IoU is proposed, which adaptively adjusts the penalty factor in combination with the size of the detected object. A DySample module is introduced, which is able to effectively extract and utilize the spatial structure features. The model in this paper improves the detection accuracy of complex objects in remote sensing images. On the DIOR dataset, compared with the baseline model YOLOV9, the accuracy is improved by 0.6

Small object detection has important application value in the fields of environmental monitoring, resource detection and analysis, etc. However, the current general object detectors are not very ideal for the detection of small objects. To this end, this paper proposes an efficient, low-complexity, anchor-free small object detection framework YOLO-SOD based on YOLOv8. First, a content-aware feature recombination upsampling operator (CARAFE) is integrated in the upsampling operation part of the framework to achieve more accurate and efficient feature reconstruction. Then, the spatial and channel reconstruction convolutional block (SCConv) is widely integrated in BackBone and Neck to reduce the computational cost caused by redundant feature extraction in visual tasks. Finally, the occlusion-aware attention module (SEAM) is introduced at the end of the detection framework to help the model more accurately identify occluded objects. The ablation experiment on the general small object detection dataset VisDrone2021 proves the effectiveness of several modules introduced in this paper for small object detection. On the VisDrone2021 dataset, YOLO-SOD can achieve an accuracy of 31.1

目前遥感影像跨视角匹配技术无法直接使用大幅卫星影像进行匹配,难以满足大范围复杂场景匹配的任务需求,且依赖大规模数据集,不具备良好的泛化能力.针对上述问题,本文在质量感知模板匹配方法的基础上结合多尺度特征融合算法,提出一种基于视角转换的跨视角遥感影像匹配方法.该方法首先利用手持摄影设备采集地面多视影像,经密集匹配生成点云数据,利用主成分分析法拟合最佳地平面并进行投影变换,以实现地面侧视视角到空视视角的转换;然后设计了特征融合模块对VGG19网络从遥感影像中提取的低、中、高尺度特征进行融合,以获取遥感影像丰富的空间信息和语义信息;最后利用质量感知模板匹配方法将从视角转换后的地面影像上提取的特征与遥感影像的融合特征进行匹配,获取匹配的软排名结果,并采用非极大值抑制算法从中筛选出高质量的匹配结果.实验结果表明,在不需要大规模数据集的情况下本文方法具有较高的准确性和较强的泛化能力,平均匹配成功率为64.6％,平均中心点偏移量为5.9像素,匹配结果准确完整,可为大场景跨视角影像匹配任务提供一种新的解决方案.

Compared with general object detection, the scale variations, arbitrary orientations, and complex backgrounds of objects in remote sensing images make it more challenging to detect oriented objects. Especially for oriented objects that have large aspect ratios, it is more difficult to accurately detect their boundary. Many methods show excellent performance on oriented object detection, most of which are anchor-based algorithms. To mitigate the performance gap between anchor-free algorithms and anchor-based algorithms, this article proposes an anchor-free algorithm called dual-resolution and deformable multihead network (DDMNet) for oriented object detection. Specifically, the dual-resolution network with bilateral fusion is adopted to extract high-resolution feature maps which contain both spatial details and multiscale contextual information. Then, the deformable convolution is incorporated into the network to alleviate the misalignment problem of oriented object detection. And a dilated feature fusion module is performed on the deformable feature maps to expand their receptive fields. Finally, box boundary-aware vectors instead of the angle are leveraged to represent the oriented bounding box and the multihead network is designed to get robust predictions. DDMNet is a single-stage oriented object detection method without using anchors and exhibits promising performance on the public challenging benchmarks. DDMNet obtains 90.49%, 93.25%, and 78.66% mean average precision on the HRSC2016, FGSD2021, and DOTA datasets. In particular, DDMNet achieves 79.86% at mAP 75 and 53.85% at mAP 85 on the HRSC2016 dataset, respectively, outperforming the current state-of-the-art methods.

Aiming at the difficulties of high-precision remote sensing image classification caused by scale variations, inter-class similarity and intra-class difference, a method with attention-guided feature fusion and joint learning is proposed for remote sensing image scene classification to make full use of multi-scale features extracted from the images. First, the deep convolutional neural network is used to extract three levels of feature maps from the images. Then, the residual attention mechanism is designed to enhance the semantic information and suppress the noise information of the feature maps. Finally, global average pooling is used to obtain the global information of the feature maps and to construct the feature vectors. Then the three levels of feature vectors are fused by connection.The three levels of feature vectors and the fusion result are classified in independent fully connected layers, respectively.During the training process, the joint loss is calculated to optimize the model ’s parameters. And multi-classifier decision-level fusion is adopted to improve the robustness of prediction. Experimental results on the UC Merced, AID and NWPU-RESISC45 datasets show that the proposed method can significantly improve the discrimination on similar scenes and scenes with intra-class difference. And compared with the similar method using multi-scale features, the overall accuracies are improved by 0.84%, 4.04% and 4.43%, respectively.

The goal of cross-view image based geo-localization is to determine the location of a given street-view image by matching it with a collection of geo-tagged aerial images, which has important applications in the fields of remote sensing information utilization and augmented reality. Most current cross-view image based geo-localization methods focus on the image content and ignore the relations between feature nodes, resulting in insufficient mining of effective information. To address this problem, this study proposes feature relation guided cross-view image based geo-localization. This method first processes aerial remote sensing images using a polar transform to achieve the geometric coarse alignment of ground-to-aerial images, and then realizes local contextual feature concern and global feature correlation modeling of the images through the feature relation guided attention generation module designed in this study. Specifically, the module includes two branches of deformable convolution based multiscale contextual feature extraction and global spatial relations mining, which effectively capture global structural information between feature nodes at different locations while correlating contextual features and guiding global feature attention generation. Finally, a novel feature aggregation module, MixVPR, is introduced to aggregate global feature descriptors to accomplish image matching and localization. After experimental validation, the cross-view image based geo-localization algorithm proposed in this study yields results of 92.08%, 97.70%, and 98.66% for the top 1, top 5, and top 10 metrics, respectively, in CVUSA, a popular public cross-view dataset, and exhibits superior performance compared to algorithms of the same type.

Most cross-view image matching algorithms focus on designing network structures with excellent performance, ignoring the content information of the image. At the same time, there are non-fixed targets such as cars, ships, and pedestrians in ground perspective images and aerial perspective images. Differences in perspective, direction, and scale cause serious interference with the cross-view matching process. This paper proposes a cross-view image matching method with feature enhancement, which first transforms the empty image to generate a transformation image aligned with the ground–aerial image domain to establish a preliminary geometric correspondence between the ground-space image. Then, the rich feature information of the deep network and the edge information of the cross-convolution layer are used to establish the feature correspondence between the ground-space images. The feature fusion module enhances the tolerance of the network model to scale differences, improving the interference problem of transient non-fixed targets on the matching performance in the images. Finally, the maximum pooling and feature aggregation strategies are adopted to aggregate local features with obvious distinguishability into global features to complete the accurate matching between ground images. The experimental results show that the proposed method has good advance and high accuracy on CVUSA, which is commonly used in public datasets, reaching 92.23%, 98.47%, and 99.74% on the top 1, top 5 and top 10 indicators, respectively, outperforming the original method in the dataset with a limited field of view and image center, better completing the cross-perspective image matching task.

Deep learning has achieved great success in remote sensing image change detection (CD). However, most methods focus only on the changed regions of images and cannot accurately identify their detailed semantic categories. In addition, most CD methods using convolutional neural networks (CNN) have difficulty capturing sufficient global information from images. To address the above issues, we propose a novel symmetric multi-task network (SMNet) that integrates global and local information for semantic change detection (SCD) in this paper. Specifically, we employ a hybrid unit consisting of pre-activated residual blocks (PR) and transformation blocks (TB) to construct the (PRTB) backbone, which obtains more abundant semantic features with local and global information from bi-temporal images. To accurately capture fine-grained changes, the multi-content fusion module (MCFM) is introduced, which effectively enhances change features by distinguishing foreground and background information in complex scenes. In the meantime, the multi-task prediction branches are adopted, and the multi-task loss function is used to jointly supervise model training to improve the performance of the network. Extensive experimental results on the challenging SECOND dataset demonstrate that our SMNet obtains 71.95% and 20.29% at mean Intersection over Union (mIoU) and Separated Kappa coefficient (Sek), respectively, which proves the effectiveness and superiority of the proposed method.