UAV-to-Ground Target Detection Based on YOLO-DSBE

To address the issues of complex background, small target scale, mutual occlusion and high missed detection rate in UAV captured images, this paper proposes a ground target detection algorithm based on YOLO-DSBE.The DC-ELAN and DC-MP modules based on deformable convolution are proposed to adapt to i...

Full description

Saved in:
Bibliographic Details
Main Author: Meng Pengshuai, Wang Feng, Zhai Weiguang, Ma Xingyu, Zhao Wei
Format: Article
Language:zho
Published: Editorial Office of Aero Weaponry 2025-04-01
Series:Hangkong bingqi
Subjects:
Online Access:https://www.aeroweaponry.avic.com/fileup/1673-5048/PDF/2024-0064.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850149315791552512
author Meng Pengshuai, Wang Feng, Zhai Weiguang, Ma Xingyu, Zhao Wei
author_facet Meng Pengshuai, Wang Feng, Zhai Weiguang, Ma Xingyu, Zhao Wei
author_sort Meng Pengshuai, Wang Feng, Zhai Weiguang, Ma Xingyu, Zhao Wei
collection DOAJ
description To address the issues of complex background, small target scale, mutual occlusion and high missed detection rate in UAV captured images, this paper proposes a ground target detection algorithm based on YOLO-DSBE.The DC-ELAN and DC-MP modules based on deformable convolution are proposed to adapt to input features of different shapes and sizes, and to improve the network’s ability to parse features in complex backgrounds; A high-resolution multi-scale detection layer is designed to boost the algorithm’s capability in extracting small target features, thereby improving the detection accuracy of minute targets. The algorithm deeply integrates the BiFormer dynamic sparse attention mechanism into the improved feature fusion network, eliminating irrelevant feature information, enhancing the focus on pertinent details, and reducing the missed detection rate. Moreover, the EIoU boundary loss function is incorporated to address the ineffectiveness of the CIoU shape penalty term, enhancing model convergence speed and detection accuracy. The experimental results show that the improved algorithm achieves an average accuracy of 56.1% on the UA-DETRAC dataset, and compared to the original algorithms, and improve by 3.5% and 2.8% respectively on the VisDrone2019 dataset, effectively improving the accuracy of UAV image recognition.
format Article
id doaj-art-51dd21fbbc0b48dc8d3f0362af5eb91b
institution OA Journals
issn 1673-5048
language zho
publishDate 2025-04-01
publisher Editorial Office of Aero Weaponry
record_format Article
series Hangkong bingqi
spelling doaj-art-51dd21fbbc0b48dc8d3f0362af5eb91b2025-08-20T02:26:59ZzhoEditorial Office of Aero WeaponryHangkong bingqi1673-50482025-04-013229410310.12132/ISSN.1673-5048.2024.0064UAV-to-Ground Target Detection Based on YOLO-DSBEMeng Pengshuai, Wang Feng, Zhai Weiguang, Ma Xingyu, Zhao Wei01. College of Electronic Information Engineering, Taiyuan University of Technology, Jinzhong 030600, China;2. College of Electrical and Power Engineering, Taiyuan University of Technology, Taiyuan 030000, ChinaTo address the issues of complex background, small target scale, mutual occlusion and high missed detection rate in UAV captured images, this paper proposes a ground target detection algorithm based on YOLO-DSBE.The DC-ELAN and DC-MP modules based on deformable convolution are proposed to adapt to input features of different shapes and sizes, and to improve the network’s ability to parse features in complex backgrounds; A high-resolution multi-scale detection layer is designed to boost the algorithm’s capability in extracting small target features, thereby improving the detection accuracy of minute targets. The algorithm deeply integrates the BiFormer dynamic sparse attention mechanism into the improved feature fusion network, eliminating irrelevant feature information, enhancing the focus on pertinent details, and reducing the missed detection rate. Moreover, the EIoU boundary loss function is incorporated to address the ineffectiveness of the CIoU shape penalty term, enhancing model convergence speed and detection accuracy. The experimental results show that the improved algorithm achieves an average accuracy of 56.1% on the UA-DETRAC dataset, and compared to the original algorithms, and improve by 3.5% and 2.8% respectively on the VisDrone2019 dataset, effectively improving the accuracy of UAV image recognition.https://www.aeroweaponry.avic.com/fileup/1673-5048/PDF/2024-0064.pdf|object detection|uav image|yolo-dsbe|deformable convnets|biformer
spellingShingle Meng Pengshuai, Wang Feng, Zhai Weiguang, Ma Xingyu, Zhao Wei
UAV-to-Ground Target Detection Based on YOLO-DSBE
Hangkong bingqi
|object detection|uav image|yolo-dsbe|deformable convnets|biformer
title UAV-to-Ground Target Detection Based on YOLO-DSBE
title_full UAV-to-Ground Target Detection Based on YOLO-DSBE
title_fullStr UAV-to-Ground Target Detection Based on YOLO-DSBE
title_full_unstemmed UAV-to-Ground Target Detection Based on YOLO-DSBE
title_short UAV-to-Ground Target Detection Based on YOLO-DSBE
title_sort uav to ground target detection based on yolo dsbe
topic |object detection|uav image|yolo-dsbe|deformable convnets|biformer
url https://www.aeroweaponry.avic.com/fileup/1673-5048/PDF/2024-0064.pdf
work_keys_str_mv AT mengpengshuaiwangfengzhaiweiguangmaxingyuzhaowei uavtogroundtargetdetectionbasedonyolodsbe