SCMM-Net: A novel deep learning-based UAV multi-source image matching algorithm
Wenyu Li, Pengze Li, Ya Guo, Xiaomeng Liang, Ruijie Jia · 2025
Multi-sensor images captured by unmanned aerial vehicles (UAVs) during flight operations often exhibit significant variations in perspective and imaging modalities, making the research on multi-source image matching particularly crucial. To address the challenge of insufficient feature point extraction in UAV multi-source image matching, we propose SCMM-Net, a novel deep learning-based UAV multi-source image matching algorithm. The proposed framework incorporates an enhanced scale enhancement module following the convolutional neural network (CNN) feature extraction based on an image pyramid structure. This innovative module expands the feature extraction receptive field, thereby significantly increasing the quantity of detectable feature points. Consequently, it improves both the initial feature matches and the final inlier counts after match filtering. Experimental results demonstrate that our method substantially enhances the number of extracted feature points, feature matches, and inliers for UAV multi-source image pairs, showing superior performance in multi-modal image matching tasks for UAV applications.