Fig 1.
Network structure of the YOLOv7 model.
Fig 2.
Schematic diagram of the transfer learning process.
Table 1.
Hyperparameters of the pre-trained model.
Fig 3.
SimAM structure.
Fig 4.
Improved ELAN structure.
Fig 5.
PConv module structure.
Fig 6.
Deep SORT module structure.
Fig 7.
IOU status.
Table 2.
Experimental environment.
Fig 8.
Curves of training and validation losses during model pre-training.
Fig 9.
mAP values of the transfer learning models and the un-transferred learning model.
Fig 10.
Results of different models for ship detection.
(a) Non-transfer learning model. (b) Transfer learning model 1. (c) Transfer learning model 2.
Table 3.
Comparison with different detection methods.
Table 4.
Comparison of tracking results with other mainstream methods.
Fig 11.
Results of different algorithms for ship detection and tracking.
(a) YOLOv7 + Deep SORT. (b) Ours.
Fig 12.
Improvement of ablation experiment results for YOLOv7 modules.
(a) The original YOLO model (b)YOLOv7+SimAM. (c)YOLOv7+SimAM+Pconv.
Table 5.
Ablation study of different improved methods.