Leather Defect Segmentation Using Semantic Segmentation Algorithms

Abstract

Leather is one of the essential materials in our life. It can be used widely to make different industrial products. Products made from leather are strong, expensive and durable which lasts for decades. So, It is very important for the industry to make a defect free product for their maximum profit and good customer feedback. Quality inspection is one of the important processes in the textile industry. It is done manually in most of the industry which is time taking, expensive, less accurate and requires lots of people. The main aim of our research work is to replace the manual process with automatic leather defect detection techniques which can save both time and money and increase the rate of production in the company. In this article, we proposed a deep learning-based semantic segmentation model that detects defects in leather images and highlights the defect with proper defect type. The experiments were carried out using the MVTEC leather dataset. The input images are changed into 256*256 pixels and then converted to gray-scale image and finally a semantic segmentation algorithm is applied to detect the leather defects. The experimental results are evaluated and compared using various semantic segmentation algorithms. We obtained the satisfactory result with evaluation metrics of 72.1% Intersection of Union (IOU) with 82.59% F1 Score on one of the semantic segmentation architectures Mobilenet_unet.