Method for Identifying Litchi Picking Position Based on YOLOv5 and PSPNet

China has the largest output of litchi in world. However, at present, is mainly picked manually, fruit farmers have high labor intensity and low efficiency. This means intelligent unmanned picking system broad prospects. The precise location main stem point very important for path planning an system. Some researchers identified branches litchi; however, there relatively little research on litchi. So, this paper presents a new open-access workflow detecting accurate locations stems data used case study. At same time, also compares several different network architectures detection segmentation selects YOLOv5 PSPNet as most promising models tasks, respectively. combines deep learning traditional image processing algorithms to calculate information points image. takes target model detect image, then extracts detected region interest (ROI) stem, uses semantic semantically segment ROI carries out post-processing operation after segmentation, obtains pixel coordinates stem. After coordinate conversion, original are obtained, drawn can obtain position recall precision method were 76.29% 92.50%, respectively, which lays foundation subsequent work obtaining three-dimensional according depth information, even though we not done paper.