A multi-scale deep learning framework for change detection in high-resolution remote sensing images

• A novel deep learning model for accurate change detection in high-resolution remote sensing imagery. • HRNet backbone maintains fine spatial details via multi-resolution feature fusion. • Residual Feature Refinement with SE attention minimizes noise and enhances features. • Pyramid Pooling Module captures multi-scale context to handle varied change sizes. • The proposed method outperforms state-of-the-art models on benchmarks with higher F1-score and recall. Change detection (CD) in high-resolution remote sensing images (HRSIs) is crucial for applications such as urbanization monitoring, environmental assessment, and disaster management. However, it remains challenging due to contextual complexity, varied change scales, and complex scene structures. To address these challenges, a novel deep learning-based CD model was proposed. The proposed model contains three key components: (1) a High-Resolution Network (HRNet) backbone to maintain fine spatial details through multi-resolution feature fusion, (2) a Residual Feature Refinement Module with Squeeze-and-Excitation attention to minimize irrelevant features and improve discriminative representations, and (3) a Pyramid Pooling Module (PPM) to obtain multi-scale contextual information. Experiments on two benchmark datasets demonstrated that the proposed model surpasses state-of-the-art models, achieving higher F1-score, recall, and IoU while decreasing false alarms. These results indicate that the proposed model is suitable for various real-world CD applications in remote sensing, also showing its capacity to handle key challenges in HRSI.