Fabric Hairiness Analysis for Quality Inspection of Pile Fabric Products Using Computer Vision Technology

Abstract

The textile industry is one of the important industries which can enhance their productivity using Artificial Intelligence technology. Most of the textile industries inspect the product manually, however, the development of AI-driven technology pushes manufacturing industries into a new era. In this paper, a computer vision-based approach for hairiness analysis in pile textile is proposed. The proposed system detects and highlights the hairy area of the fabric where the images were acquired by using an RGB camera. The experimental images were collected in our laboratory using a stereo ZED camera, but only the RGB images were processed to detect the hairiness of the fabric. The proposed hardware architecture consists mainly of camera, processing device, robotics, camera control system, while the software prototype consists of image processing techniques. In software architecture, image acquisition systems, image pre-processing, image classification, and image visualization are the major steps. A CNN-based deep learning architecture was used to classify the hairier textile and the test was carried out in big images using sliding window techniques. The classification accuracy was 99% during the training and test accuracy was 92%.