EnCANet: A Network with Entropy-Guided Saliency and Cross-Spatial Attention for Urban Building Change Detection

View at publisher Summarise with VIRAT
Authors: Tiankai Ge, Xianglong Qiao, Yanyan Zhang
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Remote sensing change detection (RSCD) is essential for urban building monitoring, yet accurate building change extraction is often limited by the structural complexity of buildings and spatial heterogeneity. To address these challenges, we propose EnCANet, a supervised RSCD framework for urban building change detection that explicitly models spatial information density and cross-spatial dependencies. The framework comprises three key components: (1) an Entropy-guided Saliency Prior (ESP) module that computes local channel-wise entropy to quantify spatial information density, generating adaptive weights to guide early-stage feature modulation; (2) an Entropy-Focused Cross-Spatial Attention (EnFocSA) mechanism that integrates entropy priors into attention calculation to suppress pseudo-change responses while enhancing cross-region contextual discrimination; and (3) a Uni-FiRE decoder that employs a unified fusion-and-refinement strategy to restore multi-scale details and ensure structural consistency. Extensive experiments on the LEVIR-CD and WHU-CD building change detection datasets demonstrate that EnCANet achieves F1 scores of 92.25% and 94.27%, surpassing the closest state-of-the-art competitors by 0.46% and 1.01%, respectively. These results quantitatively demonstrate the effectiveness of integrating entropy-driven spatial priors and cross-spatial attention for accurate and robust urban building change detection in complex heterogeneous environments. Our code is available at https://github.com/SkyeGe04/EnCANet.

Fields of Study & Subjects

Remote-Sensing Image Classification Automated Road and Building Extraction Remote Sensing and Land Use

Publication details

Journal / source
IEEE Geoscience and Remote Sensing Letters
Pages
1–1
Publication date
1 January 2026
Language
EN