Point target recognition technology based on optical images is widely used in space science, military security, automatic driving and other fields. Since the small targets in long-distance behave as point targets when being detected, the existing recognition algorithms based on spatial information such as target shape and contour will have difficulties. In order to address the problem of point target recognition, this article carries out the research of point target recognition technology based on motion features. The physical characteristic quantities of the temporal variation information of target intensity are extracted to reflect the motion state of the target. Combining with the wavelet analysis denoising to improve the accuracy of the statistical feature extraction, The recognition of point targets of different motion states is realized by using the machine learning method. This article builds an experimental system for point target recognition verification. The experimental results show that this method can realize high accuracy point target recognition based on small sample learning. The recognition accuracy based on all the extracted features and several key features is 87.8% and 84.5%, respectively, which verifies the validity of the point target recognition technology proposed in this article.