A color classification system for sunglass lenses based on YCrCb-MST hyperspectral reconstruction

Color deviation in sunglass lenses can cause various issues when pairing sunglasses, including visual fatigue, discomfort, and imbalances in pupil adjustment. The high color saturation of lenses, the minimal color variation between them, and the sensitivity of color measurement to environmental brightness restrict the effectiveness of traditional machine learning and deep learning methods in detecting and classifying lens colors. In this study, we developed a high signal-to-noise ratio image acquisition system that utilizes a point light source to illuminate the lens and capture the reflected light image, thereby minimizing the impact of ambient brightness on image acquisition. A YCrCb Multi-stage Spectral-wise Transformer (YCrCb-MST) is employed to reconstruct the hyperspectral image of the reflected light, addressing the limitations in classification ability caused by the stacking of the RGB spectrum and the minimal color differences. To improve the accuracy of color classification, this study utilizes the Particle Swarm Optimization-Extreme Learning Machine (PSO-ELM) algorithm. The PSO-ELM optimizes the input weights and hidden layer bias of the ELM through PSO. Through a specific learning and training process, the optimized ELM achieves satisfactory classification, and the reconstructed hyperspectral images of the lenses are efficiently categorized. The results confirm that the proposed YCrCb-MST-PSO-ELM method achieves an accuracy of 98.69 ± 0.59 % in the classification of sunglasses lens colors. We propose a low-cost, high-accuracy lens color classification system that offers a novel technical solution for classifying sunglasses lens colors. Additionally, it serves as a valuable reference for addressing color classification challenges in other fields.