In order to address the poor flexibility of traditional rust recognition methods when applied to different images of rust in the process of rust removal by laser, a rust detection method based on a multi-layer perception (MLP) neural network is proposed. A machine vision inspection system for laser cleaning is built, and the MLP neural network is used to identify rust images. The results show that the MLP neural network model has a coverage rate of more than 95% and a false recognition rate of less than 6% for identifying rust from images with different degrees of rust captured under different lighting conditions. The open operation of the image eliminates small misidentified areas, and the minimum external rectangle, which is used as the region of interest for laser cleaning, is generated according to the visual inspection results. The coverage rate of the final region of interest is close to 100%. This method can improve the detection efficiency of rust removal by laser and promote the automation of this process.