SFEARNet: A Network Combining Semantic Flow and Edge-Aware Refinement for Highly Efficient Remote Sensing Image Change Detection

In change detection, the pseudovariations in the visual features of remote sensing (RS) images are attributed to imaging conditions, lighting, seasonal changes, atmospheric interference, and other factors. These pseudovariations yield a great challenge to change detection. The traditional change detection network usually suffers from upsampling information loss and blurred edges. Aiming at resolving the above problems, a semantic flow and edge-aware refinement network (SFEARNet) for highly efficient RS image change detection has been proposed. The pyramid feature enhancement module (PFEM) has been designed for the enhancement of differential information. The introduction of the semantic flow information transmission module (SFITM) enables the effective transmission and retaining of key information through semantic flow. An edge-aware refinement module (EARM) has been developed, designed to extract change edge and enhance the refinement effect of the edge. The experiments have been conducted on the LEVIR building change detection dataset (LEVIR-CD), WHU building dataset (WHU-CD), Google dataset (GZ-CD), and cropland change detection dataset (CLCD). In comparison with the existing methodologies, the experimental results demonstrate that SFEARNet attains the highest change detection accuracy and the smallest floating-point operations per second (FLOPs) while maintaining a similar number of parameters (Params). This enables more efficient change detection. In particular, the proposed method can effectively refine the edges of the change region, reduce the loss of upsampling information, and enhance differential feature extraction. This brings a new solution to the field of RS image change detection. The code is available at https://github.com/miao-0417/SFEARNet.