MULTIMODAL FEATURE FUSION FOR INTELLIGENT JADE CARVING DESIGN: A HUMAN-AI COLLABORATIVE APPROACH

Abstract

Multimodal feature fusion offers a transformative approach to intelligent jade carving design by integrating human expertise with artificial intelligence to enhance creativity in traditional crafts. As a traditional Chinese art form, jade carving faces modern challenges such as limited efficiency, restricted adaptability, and the difficulty of merging traditional workflows with digital innovation. This study introduces an intelligent design framework that employs human–AI collaboration and multimodal structured feature fusion to inspire creativity while maintaining cultural authenticity. The proposed system applies a three-stage reference image recommendation process involving material-feature extraction, thematic and compositional analysis, and design-decision support. Using deep learning segmentation and color-space thresholding, key geometric and color features of jade materials are extracted and structured as encoded vectors. These are then matched with a database of traditional compositions through multimodal models that generate semantic labels and apply weighted similarity scoring. The system integrates automatic screening with human evaluation to ensure accuracy, diversity, and cultural sensitivity. Experimental outcomes indicate improvements in design efficiency, precision, interpretability, and reduced cognitive load for designers. This research underscores the value of AI-assisted systems in preserving traditional craftsmanship while driving innovation, offering a framework that connects deep learning with human creativity for the sustainable transformation of cultural heritage design.