Broken Stitch Detection Method for Sewing Operation using Deep Learning

The garment industry in Bangladesh confronts multifaceted challenges, notably fabric defects and manpower shortages, which have the potential to impede quality assurance. Although classic automatic flaw detection applications have demonstrated good performance, these techniques are frequently set up using custom features created by a human operator. Recently many deep learning methods like Convolutional Neural Networks (CNN) have been proven to perform exceptionally well in a wide variety of computer vision applications. In this paper, we tested CNN, VGG-16, and VGG-19 to detect sewing defect images from correct stitching images and broken stitching images. This experiment was performed on a set of stitched images (500) consisting of correct stitching images (250) and broken stitched images (250). To increase the effectiveness of these models, we applied filters to remove noise from the train images. The final performance analysis demonstrated that the VGG-16 algorithm provides a higher accuracy of 86%. VGG-19 and CNN obtained 82% and 75% accuracy respectively.