Leather Defect Detection Method in Clothing Design Based on TDENet

Abstract

Defect detection models generally require a large number of samples to learn the features of defects, but in practical scenarios, it is difficult to collect samples of important defects. How to use a small number of samples to learn the features of rare defects has become a challenging problem. In order to promote research on defect detection with few samples, a new leather surface defect dataset was constructed, including defect samples and defect free samples. A two-stage defect enhancement network was proposed to improve the performance of leather defect detection in small sample scenarios. It utilized defect free samples and divided the entire training process into two stages. The first stage of training requires a large number of defective samples, while the second stage of training only requires a small number of defective and flawless samples. In addition, a defect highlighting module has been proposed to better utilize defect free samples to enhance the features of leather defect areas. The experimental results show that compared with existing leather defect detection methods, this algorithm can effectively detect various defects in different materials of leather, with high accuracy and fast speed.

Keywords

• Surface defects
• machine vision
• Gabor filter
• gray level co-occurrence matrix energy