Leveraging The Unity In Variety Principle And Deep Learning Model To Enhance Aesthetic Appreciation Predictions For Ceramic Design

Abstract

Ceramic design evaluation faces the challenge of capturing consumer aesthetic preferences that balance traditional coherence with innovative variety. This study aims to address this problem by integrating deep learning models with the Unified Model of Aesthetics (UMA) to systematically predict aesthetic features of ceramic designs. Multiple models were compared, and an improved YOLOv11s architecture—incorporating MobileNetv4 as backbone, MPDIoU loss, and a Triple Attention mechanism—achieved the best performance. The proposed model reached 79.4% precision and a mean Average Precision at IoU threshold 0.5 (mAP@50) of 79.7%. Stability was confirmed through K-fold cross-validation, with accuracy fluctuating between 78% and 80%. Furthermore, the model’s predictions of unity, variety, and the principle of “Unity in Variety” showed strong alignment with participant evaluations on a 7-point Likert scale. These results demonstrate robust predictive capacity and reliable generalization, though challenges remain in classifying novel, unconventional forms. Overall, this research provides a systematic and objective framework for evaluating visual preferences by combining UMA principles with deep learning, offering both theoretical and practical significance for consumer-centered ceramic design.