| 冉宁,施高朗,张少康,郝晋渊.基于YOLOv8的遥感小目标检测算法[J].电子测量与仪器学报,2025,39(5):197-207 |
| 基于YOLOv8的遥感小目标检测算法 |
| Remote sensing small target detection algorithm based on YOLOv8 |
| |
| DOI： |
| 中文关键词: 遥感小目标 目标检测 YOLOv8 多尺度特征融合 |
| 英文关键词:remote sensing small targets object detection YOLOv8 multi-scale feature fusion |
| 基金项目:国家自然科学基金（62373132）、中央引导地方科技发展资金项目（236Z1602G）、石家庄市驻冀高校基础研究项目（241791367A）、保定市科技计划项目（2472P006）、河北大学优秀青年科研创新团队建设项目（QNTD202411）、河北大学多学科交叉研究计划（DXK202409）项目资助 |
|
|
| Author | Institution |
| Ran Ning | 1.School of Electronic Informational Engineering, Hebei University, Baoding 071002, China; 2.Laboratory of EnergySaving
Technology, Hebei University,Baoding 071002, China |
| Shi Gaolang | School of Electronic Informational Engineering, Hebei University, Baoding 071002, China |
| Zhang Shaokang | School of Cyber Security and Computer, Hebei University, Baoding
071002, China |
| HaoJinyuan | 4.HBU-UCLAN School of Media, Communication and Creative Industries, Hebei University, Baoding 071002,
China; 5.Laboratory of IoT Technology, Hebei University, Baoding 071002, China |
|
| 摘要点击次数: 1429 |
| 全文下载次数: 923 |
| 中文摘要: |
| 遥感小目标图像存在检测目标过于密集、尺寸过小、特征信息难以提取等缺陷,导致现有的目标检测算法在遥感小目标图像上的检测精度不高。为了解决上述问题,提出了一种基于YOLOv8的遥感小目标检测算法SBC-YOLOv8并采用了切片辅助超推理(SAHI)方法。首先,采用了SAHI切片方法对遥感小目标图像进行了切片操作,有效地改善了检测目标过于密集、尺寸过小的缺陷;其次,在YOLOv8的Backbone部分加入Space-to-Depth模块,增强了网络结构的特征提取能力,有效地改善了小目标特征信息难以提取的缺陷;然后,采用了加权双向特征金字塔（BiFPN）的特征融合方式并且将原本的P5层替换为P2层,增强了网络的多尺度特征融合能力,有效地提升了检测精度;最后,采用CSP-OmniFusion模块,进一步改善了遥感小目标特征信息难以提取的缺陷。实验结果表明,相比于原YOLOv8算法,采用SAHI加上SBC-YOLOv8算法的改进在VisDrone2019数据集的验证集和测试集平均精度均值（mAP）mAP@0.5分别提升了23.4%和18.3%;mAP@0.5∶0.95分别提升了17.4%和12.4%,同时在CARPK数据集和HRSID数据集上mAP@0.5分别提升了1.6%和1%,mAP@0.5∶0.95分别提升了6.1%和2.7%。因此,所提算法有效地提升了遥感小目标图像的检测效果。 |
| 英文摘要: |
| Remote sensing small target images often suffer from issues such as overly dense targets, small target sizes, and difficulty in feature extraction, leading to low detection accuracy for existing object detection algorithms. To address these problems, this paper proposes an SBC-YOLOv8 algorithm for remote sensing small target detection based on YOLOv8 and incorporates the SAHI slicing method. First, the SAHI slicing method is applied to slice the remote sensing small target images, effectively mitigating the problems of excessive target density and small sizes. Second, a Space-to-Depth module is added to the Backbone of YOLOv8 to enhance the network’s feature extraction capability, effectively addressing the challenge of extracting small target features. Then, a BiFPN feature fusion method is employed, and the original P5 layer is replaced with the P2 layer, strengthening the network’s multi-scale feature fusion ability and improving detection accuracy. Finally, the CSP-OmniFusion module is adopted to further address the difficulty of extracting remote sensing small target features. Experimental results show that, compared to the original YOLOv8 algorithm, the SBC-YOLOv8 algorithm with SAHI improvements yields a 23.4% and 18.3% increase in mAP@0.5 on the validation and test sets of the VisDrone2019 dataset, respectively; mAP@0.5∶0.95 increases by 17.4% and 12.4%, respectively. Additionally, on the CARPK and HRSID datasets, mAP@0.5 increases by 1.6% and 1%, and mAP@0.5∶0.95 increases by 6.1% and 2.7%, respectively. Therefore, the proposed algorithm effectively improves the detection performance of remote sensing small target images. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
