Laser cleaning provides an optimal solution for industrial cleaning. However, the laser cleaning mechanism is a highly nonlinear physical process, which makes the detection of laser cleaning difficult. Through the process analysis and visual image analysis of laser paint removal process, a complete and standardized laser paint removal image data set is established. Convolution neural network framework is used to optimize the depth residual network, which is suitable for multi class paint removal detection tasks. The accuracy of 98.75% is achieved in the identification of test samples. It is proved that the deep residual network is widely used in the task of paint removal, and it has potential research significance and practical value.