By 2020, it is estimated that there are more than 200,000 space targets larger than 1 cm in orbit around the Earth, and there may be millions of fragments smaller than 1 cm. As the space activities of various countries and organizations become more frequent, the number of space targets will increase, and the amount of debris in space is threatening manned and unmanned spaceflight activities. Therefore, the tracking and detection of space targets is particularly important. The main research content of this paper is the application research of machine learning in space target detection, including the application research of satellite laser ranging (including satellite debris ranging) and optical target observation.Satellite laser ranging technology is currently one of the space geodetic technologies with the highest distance accuracy, and its highest measurement accuracy can reach millimeter level. The data obtained by using satellite laser ranging technology can further carry out many kinds of scientific research, including precise orbit determination, establishment of an accurate earth reference system, determination of the autobiographical parameters of the earth, and verification of general relativity. These data need to be preprocessed before scientific application, including main echo matching and noise removal signal recognition. Chapter 2 of this article mainly introduces two new methods of satellite laser ranging signal recognition, one is based on image edge detection , One is the recognition method based on deep neural network. The newly proposed method was used to process and identify the measured data of satellites and space debris, and compared with the processing results of the current conventional methods, the accuracy of the new method was basically verified, and it can be used as a satellite laser ranging signal identification method. References and supplements.Optical observation of space targets is one of the most basic observation methods in astronomy. The third chapter of this article mainly introduces the application of deep learning in the preprocessing of optical target observation data, mainly removing contaminated observation images, including observations contaminated by clouds. The data processing classification accuracy rate reaches 99%, and the classification accuracy rate of observation images polluted by stars reaches 80%. There is still much room for improvement in this part in the future.
修改评论