J. Eur. Opt. Society-Rapid Publ. 22, 43( 2026) 425
Figure
9. Visual comparisons of some YOLO methods and our YOLO-KMM.
Table 2. Performance comparison of different models.
Model |
|
|
Performance metrics |
|
|
|
( lr) 2-7 |
Precision( P)(%) |
Recall( R)(%) |
mAP 50(%) |
FPS |
Param( M) |
GFLOPs |
YOLOv5 |
73.1 |
72.2 |
70.4 |
250.92 |
2.5 |
7.1 |
YOLOv8 |
85.7 |
75.3 |
74.9 |
231.83 |
3.0 |
8.1 |
YOLOv12 |
56.7 |
75.6 |
78.4 |
132.72 |
2.5 |
5.8 |
YOLOv11 |
86.9 |
71.4 |
80.4 |
186.31 |
2.5 |
6.3 |
YOLO-KMM |
94.0 |
74.3 |
88.2 |
246.18 |
2.3 |
5.4 |
Figure 10. FPS comparison of different models. Figure 11. Comparison of mean average precision at 50 % intersection over union( mAP50) among different models.
( mAP 50�95), model parameters, and model size. Among them, True Positive( TP) denotes the number of correctly detected targets, False Positive( FP) denotes the number of background regions mistakenly detected as targets, and False Negative( FN) denotes the number of targets incorrectly classified as background.
Precision P represents the proportion of correctly classified positive samples among all predicted positive samples, reflecting the model’ s accurate classification capability, and its calculation formula is as follows equation( 7).