Surface Defect Detection of Fabric Based on Improved Faster R-CNN

Abstract

As an indispensable part of industrial production, industrial defect detection plays a vital role in ensuring product quality. In recent years, deep learning technology provides a new solution for industrial defect detection. More and more researchers apply deep learning technology to industrial defect detection tasks, and the existing methods have achieved good research results. Therefore, based on the classical deep learning detection algorithm Faster R-CNN, this paper conducted research on surface defects of fabric as the detection object. The main research contents of this paper are as follows:Firstly, fabric data sets containing 15 kinds of defects are selected as the research object of this paper, and the defect data format is processed and converted into the coco 2017 format. Then, according to the characteristics of fabric surface defect data set, an algorithm for detection of fabric surface defects based on improved Faster R-CNN was proposed.Based on the analysis of various convolutional neural networks, ResNet50 and ResNet101 are selected as the feature extraction backbone network of the algorithm in this paper, and the deformable convolution is embedded into ResNet, so that the model can adapt to the irregular shape of defects. In view of different surface defect scales, K-means clustering algorithm was used to customize the aspect ratio of anchor frame and optimize the original anchor frame parameters.ROI-Align is used to replace ROI Pooling, and blinear interpolation algorithm is used to accurately locate the anchor frame, which solves the position mismatch problem caused by two rounds of ROI Pooling. Finally, experiments were designed on the fabric dataset and aluminum profile data published on the Tianchi platform to demonstrate that the improved Faster R-CNN based fabric surface defect detection algorithm can achieve higher defect detection accuracy.