Abstract: In urban land object recognition and classification, environmental monitoring, and cultural heritage recording and protection, airborne LiDAR point cloud and multi-spectral images are two important remote sensing data. However, existing multi-modal data fusion methods have difficulty in effectively integrating the two cross-modal data. Therefore, this paper proposes a method for fusing LiDAR point clouds with multi-spectral images by combining Hough transform with semantic feature points. Line-CNN deep learning network is used to extract line features, and Hough transform is employed to detect rectangular boxes and generate corresponding semantic feature points, thus achieving precise matching fusion of multimodal data based on feature points. Experiments results show that the matching accuracy of this method reaches as high as 97.98% in four different scenarios, and the fusion correlation coefficient exceeds 90%, which proves that the method has excellent robustness and high precision, and provides a new solution for multi-source remote sensing data fusion.

Key words: Hough transform, semantic feature points, Line-CNN network, multi-modal data fusion