Fig 1.
YOLOv5 network model.
Fig 2.
Flow chart of training and testing based on transfer learning.
Fig 3.
The connection between the smallest bounding box (green) and the center point (red), where the joint area is Su = ωh + ωgthgt − WiHi.
Fig 4.
The overview of CBAM.
Fig 5.
The principle and working mode of CoordConv.
Fig 6.
Introduce coordinate information and attention mechanism to YOLOv5 backbone network.
Fig 7.
CCW-YOLOv5 network architecture diagram.
Fig 8.
Target model improvement, training, and testing flowchart.
Table 1.
Parameter settings.
Fig 9.
Loss between prediction box and anchor box during training.
Fig 10.
Loss between prediction frame and anchor frame during verification.
Fig 11.
CCW-YOLOv5 model prediction effect diagram.
Fig 12.
Comparison of validation set mAP test curves.
Fig 13.
CCW-YOLOv5 P-R curve.
Fig 14.
Confusion matrix of test results for CCW-YOLOv5.
Table 2.
Target detection results of each model in forward looking sonar images.
Table 3.
Results of ablation research on various tricks.