Small Target Detection Algorithm Based on Improved YOLOv8n for Remote Sensing

Table 1. Experimental software and hardware configuration information
View table
Table 1. Experimental software and hardware configuration information
Configuration Parameter
GPU NVIDIA GeForce RTX 4090
Operating system Linux Ubuntu 20.04.1 LTS
Deep learning framework PyTorch 2.1.1+CUDA 12.1
Programming language Python 3.8.13

Table 2. Comparison of different target detection algorithms on NWPU VHR-10
View table
Table 2. Comparison of different target detection algorithms on NWPU VHR-10
Model mAP@0.5 mAP@0.5∶0.9
Faster R-CNN 0.806 0.547
SSD 0.789 0.538
YOLOv5n 0.838 0.537
YOLOv8n 0.872 0.545
REI-YOLOv8n 0.904 0.570

Table 3. Comparison of results of YOLOv8n and REI-YOLOv8n on NWPU VHR-10

View table

Table 3. Comparison of results of YOLOv8n and REI-YOLOv8n on NWPU VHR-10

Class	YOLOv8n				REI-YOLOv8n
Class	$P_{p r e c i s i o n}$	R	mAP@0.5	mAP@0.5∶0.9	$P_{p r e c i s i o n}$	R	mAP@0.5	mAP@0.5∶0.9
Airplane	0.974	0.985	0.994	0.666	0.936	1.000	0.995	0.670
Ship	0.870	0.861	0.891	0.536	0.768	0.778	0.878	0.522
Storage tank	0.985	0.965	0.991	0.607	0.985	0.975	0.992	0.600
Baseball diamond	0.973	0.959	0.990	0.703	0.933	0.987	0.990	0.717
Tennis court	0.922	0.678	0.792	0.486	0.896	0.757	0.872	0.520
Basketball court	0.757	0.750	0.783	0.492	0.760	0.793	0.880	0.594
Ground track field	0.973	1.000	0.995	0.826	0.949	0.964	0.987	0.840
Harbor	0.851	0.816	0.871	0.502	0.797	0.829	0.902	0.515
Bridge	0.745	0.708	0.796	0.325	0.952	0.831	0.849	0.343
Vehicle	0.950	0.353	0.617	0.301	0.882	0.438	0.693	0.382
Average	0.900	0.808	0.872	0.545	0.886	0.835	0.904	0.570

Table 4. Comparison of results of YOLOv8n and REI-YOLOv8n on Visdrone2019

View table

Table 4. Comparison of results of YOLOv8n and REI-YOLOv8n on Visdrone2019

Class	YOLOv8n				REI-YOLOv8n
Class	$P_{p r e c i s i o n}$	R	mAP@0.5	mAP@0.5∶0.9	$P_{p r e c i s i o n}$	R	mAP@0.5	mAP@0.5∶0.9
Pedestrian	0.445	0.361	0.365	0.158	0.423	0.400	0.385	0.174
People	0.506	0.245	0.283	0.103	0.515	0.272	0.305	0.116
Bicycle	0.259	0.122	0.094	0.038	0.272	0.132	0.105	0.044
Car	0.673	0.759	0.772	0.534	0.651	0.782	0.788	0.555
Van	0.481	0.401	0.397	0.272	0.476	0.419	0.409	0.291
Truck	0.474	0.308	0.306	0.201	0.478	0.331	0.337	0.232
Tricycle	0.359	0.278	0.234	0.128	0.400	0.252	0.244	0.138
Awning-tricycle	0.280	0.156	0.123	0.073	0.299	0.147	0.146	0.094
Bus	0.623	0.462	0.518	0.363	0.656	0.517	0.564	0.409
Motor	0.503	0.382	0.390	0.162	0.522	0.391	0.407	0.177
Average	0.460	0.347	0.348	0.203	0.469	0.364	0.369	0.223

Table 5. Comparison of ablation experiments of different modules on NWPU VHR-10

View table

Table 5. Comparison of ablation experiments of different modules on NWPU VHR-10

Module	Parameter /10⁶	Layer	FPS /（frame/s）	mAP@0.5	mAP@0.5∶0.9
Baseline	30.07	168	78.9	0.872	0.545
Baseline+RFE	30.09	176	70.2	0.884	0.556
Baseline+RFE+EiFPN	41.37	264	62.4	0.894	0.563
Baseline+RFE+EiFPN+Inner-PIoU	41.37	264	62.4	0.904	0.570

Table 6. Comparison of ablation experiments of different modules on Visdrone2019

View table

Table 6. Comparison of ablation experiments of different modules on Visdrone2019

Module	Parameter /10⁶	Layer	FPS /（frame/s）	mAP@0.5	mAP@0.5∶0.9
Baseline Baseline+RFE Baseline+RFE+EiFPN Baseline+RFE+EiFPN+Inner-PIoU	30.07	168	83.0	0.330	0.191
	30.09	176	76.7	0.344	0.191
	41.37	264	65.1	0.353	0.206
	41.37	264	65.1	0.369	0.223

Tools

Get Citation

Copy Citation Text

Qingjiang Cheng, Lu Li. Small Target Detection Algorithm Based on Improved YOLOv8n for Remote Sensing[J]. Laser & Optoelectronics Progress, 2025, 62(16): 1628003

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites