In view of the unsatisfactory photo quality obtained by UAV during autonomous inspection, this project studies the specific factors that affect the image acquisition quality, and carries out the feasibility study of object tracking and defect intelligent identification based on edge computing. This paper mainly solves the problems of positioning error, environmental interference and defect identification after imaging. At the same time, the intelligent image recognition algorithm is designed and optimized based on the computing power of the airborne end. This paper calculates the mapping relationship between the coordinates of two-dimensional imaging points and their spatial three-dimensional coordinates based on the principle of optical imaging. Then the mathematical model of measuring target orientation is established based on computer vision theory. The R-CNN system algorithm based on deep learning can be used to measure the orientation information of the target's relative vision with high precision.