Multifeature identification method of cashmere and wool with hybrid feature selection

Abstract

Wool and cashmere are very similar in their morphological characteristics, physical and chemical properties, and composition, which brings great difficulties and challenges to their identification. The existing methods have two shortcomings: they are insensitive to local texture features when extracting global features of sample images and ignore the correlation between features in feature selection. Therefore, this paper proposes a new cashmere wool recognition method. First, Tamura features are combined with the gray-level co-occurrence matrix (GLCM) to extract texture features, which can add statistical characteristics of local regions. Second, a feature selection method based on LGB-RFECV is proposed, which use LightGBM’s leaf-wise growth strategy to choose the different feature subsets and then use RFECV to evaluate their performance and find the best feature subset. Finally, a LightGBM classification model optimized by Bayesian optimization is used to classify cashmere and wool. The simulation results show that the proposed method achieves a recognition accuracy of 98.61%, validating that the multifeature recognition method based on hybrid feature selection is effective for the identification of cashmere and wool.