Attention-Enhanced YOLO for Real-Time Defect Detection in 3D-Printed Dental Prostheses

Abstract

The proliferation of additive manufacturing in dental prosthesis fabrication necessitates robust quality assurance mechanisms to ensure patient safety and regulatory compliance. This paper introduces an attention-enhanced YOLO architecture specifically designed for real-time defect detection in 3D-printed dental devices. The proposed approach integrates coordinate attention modules into the backbone network to enhance feature extraction while maintaining computational efficiency suitable for production-line deployment. The methodology addresses three critical defect categories: surface roughness anomalies, dimensional deviations, and internal void formations. Through comprehensive experiments on a dataset comprising 4,195 annotated dental prosthesis images spanning multiple materials and geometries, the proposed architecture achieves 93.6% mean average precision at 67.3 frames per second on edge computing hardware. Ablation studies demonstrate the effectiveness of integrating an attention mechanism and of multi-scale feature fusion strategies. The detection framework reduces false positive rates by 31.2% compared to baseline YOLO implementations, meeting stringent medical device manufacturing standards while enabling cost-effective automated inspection workflows for dental laboratories.