Classification and Identification of Natural Biodegradable Fabrics using Convolutional Neural Network
Jhamil G. Gutierrez, Jocelyn F. Villaverde, Jhamil G. Gutierrez, Jocelyn F. Villaverde
Abstract
Textile waste management faces persistent challenges particularly in the accurate classification of fabrics, which often relies on manual sorting methods based on fiber composition labels. These conventional practices are time consuming, error-prone, and frequently hindered by missing, faded, incorrect and damaged label tags. This paper presents an automated system for identifying natural fiber-based biodegradable fabrics using computer vision and machine learning. A custom device was developed, comprising a USB camera with a varifocal lens, a Raspberry Pi 5, and a 7-inch touchscreen display. The system employs a convolutional neural network (CNN) trained, validated and tested on 1,440 fabric images captured under varying focal lengths and zoom configurations. The final model achieved testing accuracy of 99.07% and reliably classified six biodegradable fabric types: Abaca, Cotton, Hessian, Linen, Silk, and Wool. This work highlights the feasibility of CNN-based fabric classification systems and offers a scalable, practical alternative to conventional sorting methods. The developed system holds significant potential for enhancing textile recycling, material recovery, and automated waste management processes within circular economy frameworks.