Improved YOLOv3 Model for Workpiece Stud Leakage Detection

In this study, a deep convolutional neural network based on an improved You only look once version 3 (YOLOv3) is proposed to improve the accuracy and real-time detection of small targets in complex backgrounds when detecting leaky weld studs automotive workpiece. To predict stud locations, prediction layer model increases from three layers four layers. An image pyramid structure obtains feature maps at different scales, shallow fusion multiple scales contour details. Focal loss added function solve imbalanced sample problem. The reduced weight simple background classes allows algorithm focus foreground classes, reducing number missed studs. Moreover, K-medians replaces original K-means clustering robustness. Finally, dataset car body workpiece built for training testing. results reveal that average YOLOv3 80.42%, which higher than Faster R-CNN, single-shot multi-box detector (SSD), YOLOv3. time per just 0.32 s (62.8% 23.8% faster SSD respectively), fulfilling requirement leakage real-world working environments.