Research on an artificial intelligence (AI) cross-sectional quantitative analysis method for cotton and regenerated cellulose fibre blended fabrics

This study addresses the technical bottlenecks in separating components in cotton and regenerated cellulose fibre (such as viscose, modal, lyocell, etc.) blended fabrics using chemical methods. This study proposes a fibre cross-sectional image segmentation and blending-ratio quantitative analysis method based on the U-Net (a U-shaped convolutional neural network). By constructing a high-resolution image acquisition system, this study achieves automatic collection and preprocessing of fibre cross-sectional images. Combined with the U-Net architecture, this study performs semantic segmentation and contour extraction, allowing for the calculation of the cross-sectional area of individual fibres. Using a density model, this study derives the mass percentage of each component. Comparative experiments were conducted on 37 sets of cotton/regenerated cellulose fibre blended samples, showing that the average error of this AI system is less than 2% compared to traditional manual methods and is highly consistent with the chemical dissolution method, with a maximum error not exceeding 3%. Additionally, this method reduces the testing time for a single sample from the traditional 60 minutes to 5 minutes, demonstrating excellent detection accuracy, efficiency, and practicality. The research results provide a feasible path for rapid, non-destructive, and intelligent detection of fibre components, with potential for application in textile testing laboratories and production lines.