Fig. 1. Network structure of improved GB-FRN-YOLOv5

Fig. 2. Chart of target example angle change. (a) Image before scaling; (b) scaled image

Fig. 3. Schematic diagram of definition of rotating frame

Fig. 4. Schematic diagram of feature unaligned

Fig. 5. Schematic diagrams of feature refinement stage. (a) Schematic diagram of feature reconstruction; (b) bilinear interpolation

Fig. 6. Module of Ghost bottleneck

Fig. 7. Schematic diagram of ordinary convolutional layer and Ghost module calculation

Fig. 8. Feature map of attentional multiscale

Fig. 9. Training loss function of improved GB-FRN-YOLOv5 model

Fig. 10. Workpiece detection result graphs of different algorithms. (a) CAD-Net; (b) R³Det; (c) Gliding vertex; (d) GB-FRN-YOLOv5

Fig. 11. Comparison of two models. (a) Test plots of model D; (b) test plots of model E

Fig. 12. Cropping schematic of two detection methods. (a) (b) Detection map and image cropping of YOLOv5;(c) (d) detection map and image cropping of GB-FRN-YOLOv5

Fig. 13. Image processing to obtain workpiece centroid. (a) Original image; (b) grayscale; (c) median filtering; (d) binarization; (e) inversion of binarization; (f) centroid calculation