Surface Defect Detection Model of Motor Commutator Based on Semantic Segmentation

Abstract

The surface defect detection of industrial parts is very important in industrial automation production, but there are problems with the small number of defect samples and the small-scale defect. To solve the above problems, this paper proposes a surface defect detection model of motor commutator based on semantic segmentation. The model is divided into two parts: segmentation network and classification network. First, the segmentation network uses an encoder-decoder to better capture small targets. The encoder uses an improved lightweight network MobileNet V3 as a feature extractor. Effectively learn the optimal features from a small number of samples, and improve the segmentation accuracy of the network. Then the classification network uses the segmentation results to make predictions, and the segmentation results provide interpretability for the prediction of the classification network. Experiments show that the proposed model has good generalization ability on a small number of samples, can effectively detect small-scale defect, and has high accuracy.