Leveraging Attention Mechanisms and Dual-Dimensional Convolution for Enhanced Remote Sensing Change Detection

Remote sensing change detection plays a vital role in various applications, including environmental monitoring, urban development, disaster management, and resource optimization. Traditional change detection methods, based on statistical and conventional approaches, face significant challenges in high-resolution remote sensing images. However, deep learning techniques have demonstrated remarkable success in improving the accuracy and efficiency of change detection by capturing complex patterns and subtle changes. Existing deep learning methods, while successful, often lack effective global-local feature understanding and bi-temporal interaction. This paper proposes a novel change detection method that leverages attention mechanisms and dual-dimensional cross convolution for bitemporal feature interactions. Specifically, two ResNet networks are employed: one with edge attention to enhance detailed local features and another with global attention to emphasize abstract, contextual features. This combination strengthens feature representation at different levels, enabling the model to focus on both precise localization and global relationships. Additionally, bitemporal interactions are achieved through dual-dimensional convolution, which improves the understanding of temporal changes across images. The method performance is validated on the real-world CLCD dataset, outperforming existing state-of-theart methods in change detection accuracy.