Neural network classification of textiles by fiber features using microscopic images
Вантажиться...
Дата
2025
Автори
Назва журналу
Номер ISSN
Назва тому
Видавець
Анотація
This study presents a reproducible neural-network method for binary textile classification based on microscopic images in the visible spectrum, aimed at distinguishing natural and synthetic fibers for circular economy applications. An open corpus of 3,107 microscope images and a unified training protocol enable fair comparison of modern architectures (ViT-B/16, EfficientNet-B0, ConvNeXt-Tiny) and ensure stable validation accuracy under realistic shooting variations, fabric deformations, and local artifacts. The approach demonstrates high classification quality on laptop-level hardware and supports practical implementation in textile sorting, laboratory composition confirmation, and quality control. Openness of data and transparency of procedures facilitate technology transfer and industrial validation, contributing to reduced waste, improved purity of secondary fractions, and lower resource consumption in textile recycling chains. Future work will focus on corpus expansion, multi-scale feature modeling, and enhanced benchmarking protocols to increase stability in complex scenes and scale the solution toward full industrial sorting systems.
Опис
Ключові слова
Бібліографічний опис
Zalutska O.O. Neural network classification of textiles by fiber features using microscopic images. Resource-Saving Technologies of Apparel, Textile & Food Industry. Proceedings of International Scientific and Practical Conference. November 20, 2025. Khmelnytskyi, Ukraine. Pp. 309-313.