Yarn-dyed Fabric Defect Detection with YOLOV2 Based on Deep Convolution Neural Networks

Abstract

To reduce labor costs for manual extract image features of yarn-dyed fabric defects, a method based on YOLOV2 is proposed for yarn-dyed fabric defect automatic localization and classification. First, 276 yarn-dyed fabric defect images are collected, preprocessed and labelled. Then, YOLO9000, YOLO-VOC and Tiny YOLO are used to construct fabric defect detection models. Through comparative study, YOLO-VOC is selected to further model improvement by optimize super-parameters of deep convolutional neural network. Finally, the improved deep convolutional neural network is tested for yarn-dyed fabric defect detection on practical fabric images. The experimental results indicate the proposed method is effective and low labor cost for yarn-dyed fabric defect detection.