In order to automatically detect the corrosion on the piston rod of hydraulic hoist of hydropower station and find the location of corrosion conveniently and accurately, a piston rod defect detection robot based on machine vision was developed. Firstly, the structure of the robot is presented, the detection system introduces the data collection process in the process of image processing, using a laser camera to capture image information. After the image preprocessing, cascade filter and morphological images after operation, the image information of gradient data, the depth of the defect data is extracted. To acquire the information of image defects in concave and convex, after using the RGB color image information acquisition defects the camera, HSV space separation and median filter and morphological processing are used to extracti defects in the image color information. Finally for defect information coding for quad, through comprehensive information for defect recognition and using QT development the human-computer interaction interface is developed, which can realize the reality and statistics of the robot\' s motion control and defect information. Through the experimental verification of the robot\'s performance, it is proved that the reliability of the robot\'s motion and the accuracy of defect identification meet the design goal.