Key points are not available for this paper at this time.
In the weaving of patterned fabrics, surface defects are the main factor affecting quality of fabrics. Due to the low efficiency of manual detection, a deep learning methods for fabric defect detection is introduced. Faster-RCNN is a target detection algorithm that takes into account both accuracy and speed, but the original Faster-RCNN network is not suitable for fabric defect detection situation where there are large differences in the size and shape of the defects. Therefore, a double-branch parallel Faster-RCNN network for large and small target defect categories is proposed. The network structure of the two branches SF-Net(small Faster-RCNN network branch) and LF-Net(large Faster-RCNN network branch) are optimized and improved respectively. At the same time, two methods of weighted difference and post-processing are taken to effectively use the information of the template images. A series of experiments are conducted on real fabric datasets. In the experimental part, compared with baseline, our method improves mAP and AR by 4.3% and 6.5%, respectively, which proves the effectiveness of the proposed method.
Wang et al. (Fri,) studied this question.