A Multi-Dimensional Vision-Based System for External Thread Defect Detection with Integrated Security Defense

This paper proposes a multi-dimensional vision-based system for external thread defect detection, aiming to overcome the limitations of conventional 2D inspection in geometric characterization. The proposed framework integrates 2D detection and 3D reconstruction to enable both accurate localization and quantitative analysis of defects. Specifically, a YOLOv13-based detector enhanced with data augmentation is employed to detect missing teeth, scratches, and corrosion defects, achieving average precisions of 95.3%, 96.7%, and 79.7%, respectively. To further capture geometric details, a Gaussian Splatting-based reconstruction method is introduced to recover high-fidelity 3D structures from multi-view images. Based on the reconstructed point cloud, dedicated 3D analysis methods are designed to enable defect size estimation with an error of less than 1 mm. Experimental results demonstrate that the proposed system achieves a favorable balance between detection accuracy and geometric measurement capability under complex industrial conditions. In addition, a robustness analysis under image perturbations is conducted to evaluate system reliability.