Image Deep Feature Fusion Matching Algorithm

This study significantly improved the feature extraction network structure by incorporating Depthwise Convolution, Pointwise Convolution, and Squeeze-and-Excitation Blocks. These innovations optimize the feature processing flow, enhancing feature representation and computational efficiency of the network. Based on this, we developed the Dynamic Feature Enhancement Matching Algorithm (DFEMA), which effectively improves matching in feature-sparse areas by enhancing feature expressions and dynamically adjusting matching thresholds. Additionally, we optimized the PROSAC algorithm, enhancing its robustness in handling mismatches. These advancements collectively led to approximately 5-10% improvement in overall matching accuracy and efficiency, offering new solutions for the accurate processing and application of complex remote sensing imagery.