Combining the Hough Transform with MLP for Identifying Cashmere and Wool Fibers

Abstract

The identification of cashmere and wool fibers has always been a challenge in the textile research field. At present, identification methods are mainly based on physical or biological properties such as DNA composition, which are time-consuming and costly to address. With the development of deep learning and computer vision, many fiber identification methods based on image analysis have emerged that effectively solve these problems, but there is much room for improvement regarding accuracy and time requirements. In this paper, a new identification algorithm is proposed that extracts the outline information of the fibers using the adaptive threshold method, performs the Hough transform on the binarized images, obtains the one-dimensional features from the Hough transform accumulator using a new feature descriptor called "theta_max", and lastly designs a Multi-Layer Perceptron (MLP) for classification. Experiments show that, the new algorithm can attain 96% accuracy in our datasets. The work significantly improves identification accuracy, reduces the complexity and time requirements of the classification models, and provides an effective method for the identification of cashmere and wool fibers.