A Lightweight Knowledge Distillation-Based Network for Efficient Remote Sensing Change Detection

The rapid development of remote sensing technologies has driven demand for lightweight networks capable of efficient change detection on resource-constrained devices. This paper proposes a novel architecture combining knowledge distillation and an improved EfficientNet structure. Key innovations include the Feature Mask Distillation (FMD) and Multi-Kernel Fusion (MKF) modules to enhance feature extraction and multi-scale representation. Experiments on LEVIR-CD and DSIFN datasets demonstrate superior performance with F1-scores of 82.97% and 66.18%, respectively, offering an efficient solution for change detection in resource-limited scenarios.