To improve the accuracy of feature point localization in laser remote sensing images, a remote sensing laser image feature localization technology based on deep learning algorithm is proposed. Using hyperspectral model parameter fusion and spectral band image detection methods, multi frequency band detection and sparse representation are performed on remote sensing laser images, extracting multi-spectral and point cloud data, and recombining features from different information dimensions; Using deep learning algorithms, dynamically iteratively search for physical and geometric feature points of the target. Using image segmentation technology to cluster and process composite features of feature points; Using undirected graphs to model neighborhood relationships for feature localization, accurately locate feature points in remote sensing images based on feature clustering and neighborhood output expression results. The simulation results show that using this method for remote sensing laser image feature localization has a high resolution level and can achieve anti blurring target feature point localization, with a maximum positioning accuracy of 0.92. Under the interference of point cloud noise, the maximum offset is 6 * 10-3.