Collaborative Optimization of Superpixel-Based Graph Cuts for Land Use Change Detection

To address the challenges of noise interference and boundary ambiguity in land-use change detection for highresolution remote sensing images, this study proposes a detection method based on superpixel-graph cut collaborative optimization. By fusing spectral differences between coregistered images from two periods, superpixel units are generated using the SLIC algorithm. Multi-dimensional change descriptors are constructed by integrating spectral and LBP texture features, with change intensity quantified via cosine similarity. A graph-cut energy function is further designed to incorporate local consistency of superpixels and spatial continuity constraints, and the global optimal solution is achieved using the min-cut/max-flow algorithm. Experimental results demonstrate that the proposed method effectively eliminates “salt-and-pepper” noise while accurately identifying complex land cover boundaries, providing a highly robust technical solution for dynamic natural resource monitoring.