Improved YOLOv8s Object Detection Algorithm for Remote Sensing Image

Table 1. Experimental environment configuration
View table
Table 1. Experimental environment configuration
Parameter Configuration
Operating system Ubuntu
GPU NVIDIA GeForce RTX 3090
Python 3.8
CUDA 11.7

Table 2. Comparison of different feature fusion structures
View table
Table 2. Comparison of different feature fusion structures
Neck mAP50 mAP50∶95
FPN-PAN 70.0 47.2
BiFPN 70.9 47.9
EBiFPN 71.7 49.1

Table 3. Comparison of different loss functions
View table
Table 3. Comparison of different loss functions
Loss function mAP50 mAP50∶95
CIoU 70.0 47.2
WIoU 70.8 47.7
ShapeIoU 70.0 47.7
MPDIoU 69.9 47.6
SIoU 70.3 47.8
SWIoU 71.5 48.3

Table 4. Results of ablation experiments

View table

Table 4. Results of ablation experiments

Method	EBiFPN	C2f-EB	SWIoU	mAP50 /%	mAP50∶95 /%	Params /M	GFLOPs
1	—	—	—	70.0	47.2	11.1	28.5
2	√	—	—	71.7	49.1	12.2	31.8
3	—	√	—	71.4	48.4	11.1	28.7
4	—	—	√	71.5	48.3	11.1	28.5
5	√	√	—	71.9	49.2	12.3	32.1
6	√	√	√	73.0	49.5	12.3	32.1

Table 5. Detection results of different methods on the DOTA dataset

View table

Table 5. Detection results of different methods on the DOTA dataset

Method	mAP50 /%	Params /M	GFLOPs	Model size /MB
SSD	47.3	24.0	274.5	94.4
Faster R-CNN	50.4	136.8	401.8	534.2
CenterNet	49.8	32.7	70.2	127.9
RetinaNet	49.4	36.4	146.3	142.5
YOLOv5m	69.8	20.9	47.9	41.2
YOLOXs	67.6	9.0	26.8	35.2
YOLOv7-Tiny	66.5	6.0	13.0	12.0
YOLOv8s	70.0	11.1	28.5	22.0
YOLOv8m	72.2	25.9	79.3	50.8
Proposed	73.0	12.3	32.1	24.3

Table 6. Detection results of different methods on the RSOD dataset

View table

Table 6. Detection results of different methods on the RSOD dataset

Method	AP /%				mAP50 /%	Params /M	GFLOPs	Model size /MB
Method	Aircraft	Olitank	Overpass	Playground	mAP50 /%	Params /M	GFLOPs	Model size /MB
SSD	56.5	93.4	58.8	96.5	76.3	24.0	274.5	94.4
Faster R-CNN	66.2	96.7	66.2	98.5	81.9	136.8	401.8	534.2
Cenernnet	81.0	98.3	87.1	98.1	91.1	32.7	70.2	127.9
Retinanet	75.9	98.7	87.1	98.3	90.0	36.4	146.3	142.5
YOLOv5m	92.0	94.2	88.8	93.9	92.2	20.9	47.9	41.2
YOLOXs	89.7	99.5	81.0	98.5	92.2	9.0	26.8	35.2
YOLOv7-Tiny	90.3	96.0	60.9	96.4	85.9	6.0	13.0	12.0
YOLOv8s	92.7	97.7	88.8	98.1	94.3	11.1	28.5	22.0
YOLOv8m	93.6	98.3	90.2	98.4	95.1	25.9	79.3	50.8
Proposed	93.9	96.5	93.9	98.6	95.7	12.3	32.1	24.3

Tools

Get Citation

Copy Citation Text

Lunming Qin, Wenquan Mei, Haoyang Cui, Houqin Bian, Xi Wang. Improved YOLOv8s Object Detection Algorithm for Remote Sensing Image[J]. Laser & Optoelectronics Progress, 2025, 62(10): 1028003

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites