Research on identification of wool and cashmere by ANN based on hyperspectral imaging technology

Abstract

In the textile industry, distinguishing between wool and cashmere can be a challenging task. Extensive research based on the microscopic images of the two has achieved very good results. However, the microscopic slide preparation process required for this approach is time-consuming and labor-intensive, limiting its practical application. To address this challenge, this paper proposes a new method that integrates artificial neural networks and hyperspectral imaging technology. The novelty of this approach lies in the fact that it does not require sample preparation, and is more simple, fast, and nondestructive. Firstly, a total of 225 wool samples and 160 cashmere samples were selected from the acquired hyperspectral images. The spectral curves (range 900–2500 nm) of these samples were extracted using the Region of Interest (ROI) tool in the ENVI software, and their characteristics were analyzed. Subsequently, due to the similarities and strong correlation between their spectral curves, Principal Component Analysis (PCA) was employed to reduce the dimensionality of the data. A single-layer neural network and a multi-layer neural network were developed using the LR (Logistic Regression) and MLP (Multilayer Perceptron) models, respectively, with a training-to-validation set ratio of 7:3. The model trained with LR achieved an accuracy of 90.3% on the training set and 81.0% on the validation set, suggesting underfitting. The MLP model performed best with five principal components, attaining a training set accuracy of 94.1% and a validation set accuracy of 92.2%. Precision, recall, and F1-score were used to evaluate the two models, and comparison of the classification performance of the two models revealed that the MLP significantly outperformed the LR model. Therefore, the application of hyperspectral imaging technology enables rapid and non-destructive identification of wool and cashmere.