Yarn quality analysis by using computer vision and deep learning techniques

Yarn quality is important in the manufacture of fabrics and textile products. Yarn analysis ensures that final products meet desired quality standards. Furthermore, the quality of the yarn directly affects the characteristics of the fabric, such as strength, durability, texture, and appearance. Therefore, the textile industry uses yarn quality monitoring and control methods throughout the entire production process, from the selection of raw materials to the manufacture of finished products. This helps to prevent defects, minimize waste, and ensure customer satisfaction. There is currently a commercial equipment, USTER TESTER 6, that measures yarn quality in an industrial environment, which is made up of intelligent sensors of different types. This equipment, characterized by its high cost and size, collects some yarn parameters, such as mass, hairiness, spectrogram, and twist. However, there is a gap in the market for a low-cost system capable of obtaining more characteristics of the yarn, integrated into an industrial environment, and using nondestructive samples. This paper presents an innovative prototype for yarn analysis using computer vision and deep learning techniques, with remote access, able to respond to the needs of Industry 4.0 and industrial digitalization. The prototype demonstrates significant advancements in performance metrics, with improvements of 5–6% in mAP0.5 and 11–12% in mAP0.5:0.95 compared with the standard YOLOv5s6 model. Using a robust 10k-fold cross-validation, the system ensures reliable performance evaluation on unseen data. Comparisons with USTER TESTER 3 indicate a relative error below 4% for parameters such as diameter and linear mass, validating the prototype’s accuracy in key measurements.