Leather Defect Detection Using Semantic Segmentation: A Hardware platform and software prototype

Abstract

Leather is a textile material made from the animal skins created through a process of tanning of hides. It is a durable material, and the price is higher compared to other types of textiles. The leather is highly sensitive to its quality and surface defect condition as it is expensive. The manual defect inspection process is tedious, labor intensive, time consuming, and often prone to human error. The aim of this research is to replace the manual process of leather inspection using fully automatic defect detection based on cutting-as machine vision techniques. The laboratorial platform consists of some mechanical components (conveyer or camera moving system), camera, lighting system, computing device (computer), and display system. In the proposed laboratorial platform, a conveyor system is used which is a fast and efficient mechanical handling apparatus for automatically transporting leather pieces during inspection. A camera is fitted above the surface of conveyor so that it can detect leather and capture and send to the computing devices. Then, a series of image processing will be carried out to detect defect detection which consist image pre-processing, training the deep learning models, and testing. The proposed semantic segmentation deep learning model was experimented using MVTEC leather dataset. We obtain 94% of Intersection of Union (IOU) in the experiments.