A Novel Laser Spot Centroid Positioning Method in Satellite Communication Based on Deep Learning with Physical Information

In recent years, laser communication technology has been widely used in satellite communication. Because of the atmospheric turbulence and spatial interference in the remote link of satellite communication, laser spot in the laser sensor is often distorted. Adaptive optics technology can correct the distortion and locate the spot centroid accurately. By doing so, the beam pointing angle is calculated, and a stable communication link is established by rotating the laser terminal according to the pointing angle. The traditional data-driven deep learning methods can extract the centroid of the spot image by self-learning. In this paper, Convolution Neural Network (CNN) and Long Short-term Memory Network (LSTM) are used to extract the spatiotemporal features of the time-series spot images, and the distribution of the spot images and the physical knowledge information of the spot centroid calculation are introduced to assist the algorithm training and convergence, which improves the interpretability and generalization of the model. The model achieved higher accuracy and better results.