Improved YOLOv8 Object Detection Method for Drone Aerial Images
A new improved YOLOv8 drone aerial image object detection method, referred to as the BDI-YOLO model, is proposed to address the problems of small target object size and blurry feature information in drone aerial images, which can lead to missed and false detections. Firstly, the Bidirectional Featur...
Saved in:
| Main Author: | |
|---|---|
| Format: | Article |
| Language: | zho |
| Published: |
Editorial Office of Aero Weaponry
2025-06-01
|
| Series: | Hangkong bingqi |
| Subjects: | |
| Online Access: | https://www.aeroweaponry.avic.com/fileup/1673-5048/PDF/2024-0163.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | A new improved YOLOv8 drone aerial image object detection method, referred to as the BDI-YOLO model, is proposed to address the problems of small target object size and blurry feature information in drone aerial images, which can lead to missed and false detections. Firstly, the Bidirectional Feature Pyramid Network (BiFPN) is introduced to improve the neck structure, utilizing a bidirectional information transmission mechanism and an adaptive feature selection mechanism to enhance the model's ability to extract features of different scales in aerial images. Secondly, replace the detection head with a Dynamic Head (Dyhead) to enhance the model's receptive field for distant small targets, thereby reducing the missed rate and false detection rate. Finally, the Inner-IoU is introduced into the original CIoU loss function and optimized into the Inner-CIoU loss function, which enhances the assessment of prediction bounding boxes and improves the model's localization precision. The experimental results on the VisDrone2019 dataset show: compared with the YOLOv8 model, the BDI-YOLO model in accuracy mAP@50 and mAP@50:95 has increased by 3.8% and 2.7% respectively, with a 4% increase in recall, a 9.4% decrease in computational complexity, and a 28.8% decrease in parameter count. The BDI-YOLO model can adapt well to the target detection task of unmanned aerial vehicle aerial images in complex scenes. |
|---|---|
| ISSN: | 1673-5048 |