Fig. 1. Structure of YOLOv5s network

Fig. 2. CSPNet modules. (a) CSP1_X component; (b) CSP2_X component

Fig. 3. Rotation box long side representation method

Fig. 4. Calculation of loss at the output of the rotating frame

Fig. 5. Background and target with similar texture and color

Fig. 6. SimAM module

Fig. 7. Head network structure with SimAM

Fig. 8. Diagram of SIoU regression

Fig. 9. CBM module replacement

Fig. 10. Schematic of angular periodic problem

Fig. 11. Training process of rotating box Head network

Fig. 12. Diagram of improved YOLOv5s network structure

Fig. 13. Data enhancement renderings

Fig. 14. Model training process

Fig. 15. Comparison of detection effects of different algorithms. (a) (b) (c) Base model; (d) (e) (f) proposed method