Crater Obstacle Recognition and Detection of Lunar Landing Based on YOLO v4

During the landing process of an unmanned or manned lunar lander, the lunar image collected by the sensor has the characteristics of insufficient illumination, single surface texture, large scale changes and no prominent features, which makes it more difficult to detect landing obstacles in real time. Traditional image processing technology for obstacle recognition and detection is prone to missed and misdetected phenomena. Aiming at the problem of detecting crater obstacles during the descent of the lander, a lunar landing crater detection model based on YOLO v4 was proposed, and a crater data set was made, and the data set types were expanded through data enhancement algorithms, and the DenseNet network was used to enhance Feature extraction capabilities. Experiments on the crater data set show that the improved algorithm in this paper is better than YOLO v4 and YOLO v4 tiny in accuracy. Compared with YOLO v4, the average accuracy is increased from 72.65% to 77.24%, and good results are achieved.