Research on the Identification and Optimization of CMF Design Styles for Woven Upholstery Fabrics Based on Sensory Imagery and CNN Model

Bridging subjective user perception with objective textile design parameters remains a fundamental challenge in intelligent material and surface engineering. This study presents a data-driven framework for the recognition and optimization of Color, Material, and Finish (CMF) design styles in woven upholstery fabrics by integrating Kansei Engineering with a convolutional neural network (CNN). A dataset containing 2,048 textile images was established, from which color descriptors, GLCM-based texture features, and gloss-related finish characteristics were extracted to characterize visual CMF attributes. Human perceptual evaluation involving 60 participants was employed to construct Kansei-based style labels, and a lightweight CNN achieved a classification accuracy of 95.8% for four representative design categories. The trained model was subsequently incorporated into a genetic algorithm to generate optimized CMF parameter combinations for target aesthetic preferences, and user verification demonstrated a significant increase in style consistency, with the mean opinion score improving from 2.05 to 6.20. By quantitatively linking visual feature representation with perceptual cognition, the proposed framework provides an effective computational strategy for intelligent textile design and offers methodological insights for perception-driven surface engineering, pattern recognition, and functional material optimization in multidisciplinary systems where image-based sensing and feature transmission play essential roles.