Object Recognition of Three-dimensional SAR Based on PointNet

Object recognition of three-dimensional (3-D) SAR is one of the hot areas in the SAR application field. In order to improve the performance of 3-D SAR target recognition, this paper proposes a novel deep learning method for 3-D point-cloud recognition based on PointNet classification network. In the scheme, a point cloud extraction model is used to 3-D SAR image, then a PointNet network is trained with the extracted point cloud data. The effectiveness of the approach is verified by 3-D SAR simulation images with three different airplane targets. The results demonstrate that the method can obtain a high recognition accuracy in the case of a low 3-D SAR resolution.