Paper
10 October 2023 UAV image object detection network based on Yolo
Jubin Huang, Zexuan Guo, Xuebin Hong, Haohai Wu, Zhe Lin
Author Affiliations +
Proceedings Volume 12799, Third International Conference on Advanced Algorithms and Signal Image Processing (AASIP 2023); 127995M (2023) https://doi.org/10.1117/12.3005996
Event: 3rd International Conference on Advanced Algorithms and Signal Image Processing (AASIP 2023), 2023, Kuala Lumpur, Malaysia
Abstract
This paper proposes an object detection network based on YOLOv7 for UAV(unmanned aerial vehicle) aerial images which have a large field of view, complex backgrounds, small target sizes and uneven position distribution. The proposed network integrates Dyhead (a dynamic target detection head) which is based on attention mechanism into the network, causing the model to focus more on target-rich regions and alleviating noise interference caused by complex backgrounds, thereby improving detection accuracy. Additionally, the network proposes a new loss function which combines CIOU(Complete IOU) and WISE-IOU. The improved loss function balances the IOU and geometric metrics (such as center point distance and aspect ratio) in the position regression loss, increasing the model's tolerance for low-quality data samples and improving its overall generalization capability and detection accuracy. Finally, experiment on the Visdrone UAV image dataset show that the proposed network has a 2.1% and 1.3% increase in mAP@0.5(mean average precision(IOU=0.5)) and mAP@0.5:0.95(mean average precision(IOU=[.5:.95])) evaluation index compared to the base network. Increasement of accuracy are also observed in all ten subdivision categories of the dataset, validating the effectiveness of the proposed network for UAV image object detection tasks.
(2023) Published by SPIE. Downloading of the abstract is permitted for personal use only.
Jubin Huang, Zexuan Guo, Xuebin Hong, Haohai Wu, and Zhe Lin "UAV image object detection network based on Yolo", Proc. SPIE 12799, Third International Conference on Advanced Algorithms and Signal Image Processing (AASIP 2023), 127995M (10 October 2023); https://doi.org/10.1117/12.3005996
Advertisement
Advertisement
RIGHTS & PERMISSIONS
Get copyright permission  Get copyright permission on Copyright Marketplace
KEYWORDS
Object detection

Unmanned aerial vehicles

Target detection

Head

Detection and tracking algorithms

Data modeling

Small targets

Back to Top