Ying Wang

Building damage change detection (BDCD) is crucial for emergency assessment and post-disaster reconstruction following natural disasters. While Visual Foundation Models, such as the Segment Anything Model (SAM) and its variants, have shown promise in remote sensing tasks, their heavy reliance on large-scale annotated data hinders rapid response in emergency scenarios and significantly limits their performance in damage change detection. To address these issues, this paper proposes Enhancing SAM with Wavelet Domain Adapter Using Extremely Limited Data (ELDSAM) for BDCD that fine-tunes SAM using a wavelet domain adapter under extremely limited annotated samples. Specifically, a Wavelet Transform based Multi-scale Frequency Domain Adapter (WTAdapter) is first constructed. By freezing SAM's original parameters to preserve its foundational visual capabilities, only the adapter is fine-tuned with minimal data for task adaptation by its robust frequency feature capturing ability. Secondly, a Decoder Feedback Optimization Module (DFOM) is designed to dynamically generate attention maps for damaged areas, guiding the adapter to focus on potential damage features while suppressing irrelevant background interference. Finally, Cross-Sample Contrastive Learning (CSCL) is introduced to further mine discriminative information for damage changes from the limited data, thereby enhancing the model's focus on damaged regions. Experimental results demonstrate the superior performance of ELDSAM in scenarios with extremely limited data. Using only 16 annotated samples, it outperforms mainstream approaches in damage change detection performance, showing strong potential for practical emergency applications. Our code will be available at http://github.com/youngwill/ELDSAM.

High-resolution optical remote sensing change detection (CD) is of great significance in urban evolution monitoring, disaster assessment, and land management. Traditional deep models often face computational, memory, and inference latency bottlenecks when processing large high resolution imagery. To address this, we propose EfficientViM CD: a Hidden-State Mixer based efficient remote sensing change detection network. The approach builds upon the EfficientViM backbone, migrating global interaction operations into a compact hidden state space and leveraging Hidden State Mixer based on state space duality (HSM-SSD) to fuse global context while reducing computational complexity. We employ a Siamese encoding architecture to extract multi-scale features and hidden states from paired temporal images, and utilize a Cross-Hidden Fusion module to integrate hidden semantic interactions between time points. At each scale, local difference features are computed and enhanced in hidden state space, and a multi-scale decoder reconstructs a pixel-level change probability map. We conducted experiments on four public datasets (LEVIR-CD, WHU-CD, S2Looking, SVCD) and compared against nine state-of-the-art methods. Results demonstrate that EfficientViM-CD achieves competitive accuracy while delivering significant advantages in inference speed and memory efficiency. This method offers a lightweight, efficient, and scalable solution for high-resolution remote sensing change detection, with potential for real-time monitoring and emergency response systems.

Salient object detection plays a crucial role in intelligent interpretation of remote sensing tasks. Significant advancements have been made with SOD methods based on convolutional neural networks (CNNs) and Transformers. However, the structural differences and complex topological relationships presented by diverse remote sensing scenes often limit SOD performance. To address these challenges, this study proposes a novel method for SOD in remote sensing images called Topology-Aware Hierarchical Mamba Network (THMNet), which fully utilizes the Mamba’s advantages in global modeling and linear computation. Specifically, a grouped Mamba topology enhancement module is designed to extract high-level semantic features and improve global representation of structural information for salient objects. Furthermore, a Hierarchical Mamba Segmentation Optimization (HMSO) module is introduced, which employs prior guidance to adaptively focus on the foreground and background at different levels. This module reduces processing of non-critical information within complex topological relationships, and enhances the capture and depiction of intricate objects. Finally, experimental results demonstrate that our method outperforms mainstream CNNs- and Transformers-based SOD techniques, effectively detecting salient objects in complex backgrounds. The code is available at: https://github.com/LazyShark2001/THMNet.

Remote sensing (RS) image change detection (CD) aims to identify and localize the differences in the same geographical area observed across bi-temporal RS images. Existing CD methods typically focus on characterizing and distinguishing change regions, while often overlooking the non-change areas (background information) in RS images. Therefore, the influence of subtle changes in non-change regions may be amplified, leading to false negatives or false positives in the change regions. To address this issue, we propose a multi-view complementary learning network (MVCL) for RS image CD, which considers change detection from both background view and change view perspectives, while focusing on the extraction and aggregation of both change and non-change (background) features. Specifically, we introduce a background branch during the encoder stage dedicated to extracting background information. This background information is then injected into the change regions through the background-dependent feature fusion module (BFFM), serving as prior knowledge. By leveraging the stable background information, irrelevant and noisy changes are suppressed, guiding the generation of change regions and reducing the interference of background noise. Furthermore, to mitigate the noise introduced during multi-scale fusion in the non-change (background) areas, we propose a prior-guided spatial background enhancement module (PGBM), which consists of a prior semantic guidance strategy and a spatial background enhancement module. According to the results of experiments conducted on three public datasets, the proposed method outperforms several SOTA methods in both quantitative metrics and visual effectiveness. The source code is available at https://github.com/AdKingH/MVCL.

The safety and management efficiency of urban infrastructure are crucial in the urbanization process, and the rapid, precise identification of road manhole covers is essential for ensuring public safety and optimizing maintenance operations. However, the diverse shapes, materials, complex backgrounds, and visual similarities of road manhole cover defects pose significant challenges for object detection. Methods based on deep learning, while employing multi-scale pyramid structures for feature extraction, often overlook the visual similarity among different defect types and the subtle differences in edge features, leading to limited detection performance. This paper introduces an enhanced method, EEFA-YOLO, for defect detection in road manhole covers, incorporating two novel modules: the Multi-Scale Edge Enhancement (MSEE) and the Feature Aggregation Pyramid (FAP). The MSEE utilizes multi-scale feature extraction and edge information enhancement to improve the model’s sensitivity to subtle objects and edge details. Meanwhile, the FAP leverages a feature aggregation and diffusion mechanism to ensure uniform contextual information across scales, effectively addressing issues related to scale variance and background interference. Additionally, we constructed a diverse dataset of road manhole covers across various scenarios and defect types, encompassing four categories: good, broken, lost, and misaligned, providing high-quality data support for algorithm training and validation. Experimental results indicate that the proposed method outperforms existing approaches across most evaluation metrics, achieving an mAP at least 1.5% higher than the baseline YOLOv11 and 5.3% higher than the representative two-stage method Faster-RCNN. Additionally, it demonstrates excellent adaptability to medium- and large-scale targets, providing an effective solution for intelligent city management and road maintenance.

Small object detection in aerial imagery remains challenging due to sparse feature representation, limited spatial resolution, and complex background interference. Current deep learning approaches enhance detection performance through multi-scale feature fusion, leveraging convolutional operations to expand the receptive field or self-attention mechanisms for global context modeling. However, these methods primarily rely on spatial-domain features, while self-attention introduces high computational costs, and conventional fusion strategies (e.g., concatenation or addition) often result in weak feature correlation or boundary misalignment. To address these challenges, we propose a unified spatial-frequency modeling and multi-scale alignment fusion framework, termed USF-DETR, for small object detection. The framework comprises three key modules: the Spatial-Frequency Interaction Backbone (SFIB), the Dual Alignment and Balance Fusion FPN (DABF-FPN), and the Efficient Attention-AIFI (EA-AIFI). The SFIB integrates the Scharr operator for spatial edge and detail extraction and FFT/IFFT for capturing frequency-domain patterns, achieving a balanced fusion of global semantics and local details. The DABF-FPN employs bidirectional geometric alignment and adaptive attention to enhance the significance expression of the target area, suppress background noise, and improve feature asymmetry across scales. The EA-AIFI streamlines the Transformer attention mechanism by removing key-value interactions and encoding query relationships via linear projections, significantly boosting inference speed and contextual modeling. Experiments on the VisDrone and TinyPerson datasets demonstrate the effectiveness of USF-DETR, achieving improvements of 2.3% and 1.4% mAP over baselines, respectively, while balancing accuracy and computational efficiency. The framework outperforms state-of-the-art methods in small object detection.

Change detection (CD) signifies a pivotal domain within remote sensing image processing. The transformer has been introduced in the field of CD for its global perception capabilities. However, existing transformer-based methodologies serve primarily as mere direct feature extractors, rendering the attention mechanism within the decoder underutilized. To address this, we propose a symmetric transformer learning network (STLNet) specifically tailored for remote sensing image CD tasks, constructed entirely using transformers. STLNet is designed to leverage the intrinsic capability of transformers to model extensive long-range dependencies effectively. This approach significantly bolsters the extraction of distinctive globallevel features, thereby facilitating the accurate delineation of CD regions. Initially, we utilize an adaptive multigrain encoder to extract feature information from bitemporal images, thereby honing the focus on changing targets and providing deeper and more comprehensive information. Subsequently, we adopt an effective decoder architecture comprised of transformer structures, namely the local gather decoder (LGD). The LGD employs a multi-level semantic feature integration from the encoder to augment feature representation and inter dependencies, crucial for detailing small changed areas effectively via a hierarchical attentional fusion (HAF) block. Ultimately, the detection of changes is based on the rich semantic information provided by the LGD, enabling us to achieve enhanced precision in our remote sensing CD efforts. Results show that the proposed STLNet achieved F1 scores of 92.32% on LEVIR-CD, 90.01% on WHU-CD, and 82.15% on SYSU-CD, surpassing mainstream CD methods.