Abstract. Deep-learning methods dominate remote-sensing change detection (CD), yet state-of-the-art models remain parameter-heavy and struggle with crisp boundaries, limiting their use on edge devices. We present LEDGNet, a Lightweight, Edge-knowledge- Distillation-Guided CD Network, that reconciles accuracy, boundary fidelity, and efficiency. LEDGNet integrates three purpose-built components: 1) an Edge Distillation Module that mines multi-scale boundary cues from a high-capacity teacher and transfers them to a compact student through an edge-aware loss; 2) StarLite, a depth-wise separable encoder that preserves fine spatial detail while minimizing floating-point operations; and 3) LiteDecoder, an inexpensive feature-fusion head that restores full resolution without bulky up-sampling. This design halves the parameters and inference time of mainstream fine-grained CD networks while enhancing edge sharpness. On the CDD and LEVIR-CD benchmarks, LEDGNet achieves competitive F1 performance while maintaining a compact footprint of 20.58 M parameters and 35.18 G FLOPs. With an inference time of 255 ms, it strikes a balance between resource consumption and detection efficiency, making it well-suited for high-efficiency remote sensing monitoring.
Tingyu Ji, Yixin Chen, Ruiqian Zhang et al.· ISPRS Annals of the Photogra...· 0 citations
Effective management of industrial land is essential given its substantial environmental footprint, intensive resource consumption, and implications for urban planning and public health. Accurate visual grounding of industrial parcels in remote sensing imagery provides a critical basis for assessing associated environmental risks and supporting evidence-based resource allocation. However, existing approaches often rely primarily on visual cues and remain vulnerable to domain-specific ambiguity in complex urban environments. To address these limitations, we propose retrieval-augmented geospatial vision–language grounding (RAGVLG), a model-agnostic framework that integrates structured knowledge retrieval with vision–language models (VLMs) to improve industrial land grounding in remote sensing imagery. RAGVLG retrieves domain knowledge and contextual exemplars to guide grounding decisions, enabling flexible adaptation to different VLM backbones. We construct an industrial land-grounding dataset and systematically evaluate the proposed framework, demonstrating consistent improvements over baseline VLMs across multiple backbones. We further demonstrate its utility through city-scale applications in Shenzhen and Hong Kong, where industrial land is mapped, and the spatial distribution of potential environmental risks is assessed by integrating key environmental parameters. Local indicators of spatial association are used to quantify spatial clustering of risk. The results show that high–high hotspots account for 10.62% of all grids in Shenzhen, forming a continuous coastal corridor primarily across Nanshan and Baoan Districts. Hong Kong exhibits a higher proportion of high–high clusters, at 15.31%, with more fragmented high-risk areas concentrated in Kwai Tsing and Kowloon. These findings provide quantitative and spatially explicit evidence for regional industrial planning, environmental risk management, and sustainable industrial development.
Yuling Wu, Xiyu Jin, Ruiqian Zhang et al.· IEEE Journal of Selected Top...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
