FastSAM-based Change Detection Network for Remote Sensing Images

Remote sensing images change detection often faces the problem of false detections and missed detections due to various factors such as different sensors, seasons, and weather conditions when acquiring bi-temporal images. To tackle this challenge, we propose a remote sensing images change detection network based on FastSAM. First, a siamese network structure is employed as an encoder, utilizing the vision foundational model FastSAM to enhance the generalization ability. Then, in order to reinforce the semantic features within the regions of change, we propose a Differential Enhancement Adapter (DEA) module. Finally, a Full-Scale Skip Connections (FSC) is adopted to synergize the deep semantic features from varying scales with the superficial semantics, thus bolstering the model's capacity to discern finer details. This model has achieved significant results on the Jiashan County land use change detection datasets, and has also made significant improvements compared to other advanced models.