Figure 1.
YOLO algorithm implementation flow
Figure 2.
Improved YOLOv8 network structure diagram
Figure 3.
Partial data display of self-built night vision pedestrian data set
Figure 4.
LabelImg User Interface
Figure 5.
Content of the YOLO-formatted annotation txt file
Figure 6.
TXT Format Structure of YOLOv8
TABLE I.
EXPERIMENTAL PLATFORM HARDWARE CONFIGURATION
| Name | Related Configuration |
|---|---|
| Operating System | Windows11 |
| Processor | 11th Gen Intel(R) Core(TM) i7-11800H @ 2.30GHz 2.30 GHz |
| Memory | 16G |
| GPU | NVIDIA GeForce RTX 3050 |
| Deep Learning | Pytorch |
| Framework |
TABLE II.
EXPERIMENTAL PARAMETER SETTINGS
| Parameter | Parameter Setting |
|---|---|
| Number of training iterations | 50 |
| Learning rate | 0.001 |
| Batch size | 16 |
| Image size | 640 |
| Optimizer | Lion |
TABLE III.
COMPARISON OF ORIGINAL AND IMPROVED ALGORITHMS
| Algorithm | Precision | Recall | mAP@0.5:0.95 | mAP@0.5 | F1 Score |
|---|---|---|---|---|---|
| YOLOv8 | 85.3 | 89.3 | 88.9 | 57.8 | 87.2 |
| YOLOv8+CBAM | 87.3 | 84.4 | 88.94 | 58.2 | 87.84 |
Figure 7.
Training results prior to model optimization
Figure 8.
Training results after model optimization
Figure 9.
PR curve before algorithm improvement
Figure 10.
PR curve after algorithm improvement
Figure 11.
Image Detection Interface
Figure 12.
Image Detection Interface
Figure 13.
Video Detection Interface