A novel drying-free identification method of cashmere textiles by NIR spectroscopy combined with an adaptive representation learning classification method

Rapid identification of cashmere textiles using near-infrared (NIR) spectroscopy chronically has been limited by the high spectral similarity of cashmere and wool, moisture interference, and the deficiency of existing spectral pattern recognition methods. A novel drying-free method of NIR spectroscopy combined with an adaptive representation learning-based classification algorithm and moisture-containing samples preparation was proposed to tackle the issue. The diffused reflectance NIR spectra of textile samples with different colors, different textures, and varying moisture contents were collected. The Soft Independent Modeling by Class Analogy (SIMCA), the Support Vector Machine (SVM) and the new method were compared to build the cashmere identification models of the dried samples and the moisture-containing samples, respectively. The effects of common spectral pretreating methods and moisture variation on the identification of cashmere textiles were studied in detail. The performance of the new approach is much superior to others when identifying moisture-containing samples, an accuracy for cashmere textiles, 93.33% and one for the cashmere-wool blend, 96.60% achieved, respectively. This fruit has been integrated with a portable NIR-based textile analyzer and successfully applied to identify the real-world textiles free of drying. This research is of great practical significance to the market supervision and the production of cashmere textiles.