Remote-Sensing Image Object Detection Based on Improved YOLOv8 Algorithm

Xiuzai Zhang; Tao Shen; Dai Xu

doi:10.3788/LOP231803

Laser & Optoelectronics Progress, Volume. 61, Issue 10, 1028001(2024)

Remote-Sensing Image Object Detection Based on Improved YOLOv8 Algorithm

Xiuzai Zhang^1,2、*, Tao Shen¹, and Dai Xu¹

¹School of Electronic and Information Engineering, Nanjing University of Information Science & Technology, Nanjing 210044, Jiangsu, China

²Jiangsu Province Atmospheric Environment and Equipment Technology Collaborative Innovation Center, Nanjing University of Information Science & Technology, Nanjing 210044, Jiangsu, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(13)

Fig. 1. YOLOv8 algorithm model structure

Download full size

Fig. 2. Improved YOLOv8 structure

Download full size

Fig. 3. Structure of GAM attention mechanism

Download full size

Fig. 4. Channel attention submodule

Download full size

Fig. 5. Spatial attention submodule

Download full size

Fig. 6. Residual attention module

Download full size

Fig. 7. Variable convolution process

Download full size

Fig. 8. Deformable RoI pooling process

Download full size

Fig. 9. Structure of C2f-DCN

Download full size

Fig. 10. Detection effects of different models. (a)‒(c) Original images; (d)‒(f) detection effects of improved YOLOv3; (g)‒(i) detection effects of YOLOv5; (j)‒(i) detection effects of YOLOv8; (m)‒(o) detection effects of improved YOLOv8

Download full size

Table 1. Ablation experiments

View table

Table 1. Ablation experiments

Method	Type	P /%	R /%	mAP@0.5 /%	FPS /（frame/s）
1	YOLOv8	75.5	66.1	69.8	98
2	YOLOv8+DWIoU	75.8	65.9	70.2	95
3	YOLOv8+ReGAM	74.9	66.7	70.3	92
4	YOLOv8+C2f-DCN	76.1	66.5	71.0	81
5	YOLOv8+DWIoU+ReGAM	75.8	66.6	70.9	89
6	YOLOv8+DWIoU+C2f-DCN	76.1	66.7	71.5	80
7	YOLOv8+ReGAM+C2f-DCN	75.9	66.6	71.3	78
8	YOLOV8+DWIoU+ReGAM+C2f-DCN	76.2	66.8	72.1	77

Table 2. Performance comparison of different network models on DOTA dataset
View table
Table 2. Performance comparison of different network models on DOTA dataset
Method FPS /（frame/s） mAP@0.5 /%
Faster R-CNN 8 42
SSD 38 59.6
YOLOv3 18 68.2
YOLOv4 17 68.4
YOLOv5 38 69.4
YOLOv8 98 69.8
Improved YOLOv8 77 72.1

Table 3. Performance comparison of different network models on RSOD dataset
View table
Table 3. Performance comparison of different network models on RSOD dataset
Method FPS /（frame/s） mAP@0.5 /%
Faster R-CNN 12 90.3
SSD 46 89.2
YOLOv3 33 84.7
YOLOv4 57 87.8
YOLOv5 62 91.8
YOLOv8 137 93.4
Improved YOLOv8 123 94.6

Tools

Get Citation

Copy Citation Text

Xiuzai Zhang, Tao Shen, Dai Xu. Remote-Sensing Image Object Detection Based on Improved YOLOv8 Algorithm[J]. Laser & Optoelectronics Progress, 2024, 61(10): 1028001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites