Fig 1.
YOLOv5s network structure.
Fig 2.
MC-YOLOv5s network structure.
Fig 3.
The structure of bneck in MobileNetV3.
Table 1.
The detailed structure of the backbone.
Where the expansion factor represents the dimensionality of the first convolutional ascending dimension, SE represents the squeeze-and-excitation module, and NL represents the activation function type.
Fig 4.
The structure of the C3 module and CNeB module.
(a) structure of the C3 module (b) structure of the CNeB module.
Fig 5.
CNeB-block structure.
Fig 6.
Comparison of standard convolution with depthwise separable convolution.
(a) Standard convolution (b) Depthwise separable convolution.
Fig 7.
Structure of Inverted bottleneck before and after improvement.
Fig 8.
GELU function.
Table 2.
Camera parameters.
Fig 9.
The first row is an example of an image from the SeaShips dataset. The second row is an example of the images collected for this experiment.
Fig 10.
Statistical analysis of ship dataset labels.
Table 3.
Experimental platform configuration.
Fig 11.
MC-YOLOv5s training results.
Table 4.
Performance comparison between YOLOv5s and MC-YOLOv5s.
Fig 12.
Test results for different types of ships.
Fig 13.
P-R curves for different types of ships.
(a) YOLOv5s model P-R curves (b) MC-YOLOv5s model P-R curves.
Fig 14.
F1 fraction curves for different types of ships.
(a) YOLOv5s (b) MC-YOLOv5s.
Fig 15.
Confusion matrix for different types of ship test sets.
(a) YOLOv5s (b) MC-YOLOv5s.
Fig 16.
Comparison of YOLOv5s and MC-YOLOv5s in the test set visuals.
(a) YOLOv5s (b) MC-YOLOv5s.
Table 5.
Comparison of the performance of the models on the validation set.
Fig 17.
Inspection results of different models on different types of ships.
Table 6.
Comparison of performance indicators for ablation experiments.