Remote Sensing Image Detection Algorithm Based on GhostNetv2 Improved YOLOv5s Algorithm

The imaging characteristics of remote sensing images make the detection of many targets and different target sizes, resulting in the difficulty of small target detection and insufficient detection accuracy. In this paper, we propose an optimization model for remote sensing image detection based on Y...

Full description

Saved in:
Bibliographic Details
Published in2023 8th International Conference on Information Systems Engineering (ICISE) pp. 193 - 196
Main Authors Li, Ruiyi, Huang, Wenxuan, Liu, Chunlin, Chen, Peizhi
Format Conference Proceeding
LanguageEnglish
Published IEEE 23.06.2023
Subjects
Computational modeling
DFC attention
GhostNetv2
Imaging
Optimization methods
Prediction algorithms
Predictive models
remote sensing image
Solid modeling
Training
YOLOv5s
Online AccessGet full text

Cover

More Information
Summary:The imaging characteristics of remote sensing images make the detection of many targets and different target sizes, resulting in the difficulty of small target detection and insufficient detection accuracy. In this paper, we propose an optimization model for remote sensing image detection based on YOLOv5s, using GhostNetv2 self-attention mechanism to capture global information, introducing GhostNetv2 module to improve YOLOv5s algorithm backbone network, and using EIOU border position regression loss function instead of CIOU loss function. Model training tests are conducted on the NWPU-VHR 10 public dataset, and the improvements improve the detection accuracy of remote sensing images, while making the model more lightweight. Compared with the YOLOv5s benchmark model, the map value of the algorithm in this paper is 94.9%, which is 2.2% better than YOLOv5s in terms of map, and the number of parameters and computation are reduced by 9.2% and 3.1%, respectively. It can effectively improve the accuracy and lighten the model at the same time.
ISSN:2643-7309
DOI:10.1109/ICISE60366.2023.00047