信号处理2025，Vol.41Issue(5)：853-866,14.DOI:10.12466/xhcl.2025.05.007
基于改进YOLOv8的凝视雷达小目标检测算法
Small Target Detection Algorithm Based on Improved YOLOv8 for Staring Radar
摘要
Abstract
Small target detection plays a crucial role in applications such as low-altitude aircraft management,environ-mental monitoring,and border security,serving as a key technology for ensuring low-altitude airspace safety and pro-moting the development of the low-altitude economy.However,conventional radar target detection algorithms often poorly perform when dealing with"low,slow,and small"targets due to challenges such as limited target size,low signal-to-noise ratio(SNR),and interference from background clutter.This study proposes feeding the radar Range-Doppler(RD)plane into an improved YOLOv8 model and validating the approach using real radar data,resulting in en-hanced small target detection performance to address these issues.First,a holographic staring radar is used to collect real-world data,followed by data annotation and dataset preparation.In terms of model design,this paper introduces the Efficient Channel Attention(ECA)mechanism,which effectively captures dependencies between channels by replacing traditional fully connected layers with one-dimensional convolutions without adding extra parameters.This enhancement improves the ability to select relevant features of the network across different channels,thereby enabling more accurate small target detection.Additionally,a dedicated small target detection layer is incorporated into the model.This layer ad-justs the resolution of feature maps,allowing the model to extract and recognize features from even smaller targets,ad-dressing the shortcomings of conventional detection networks in handling small targets.Furthermore,multiple models are compared in the experimental section.The results show that the proposed improved model outperforms other models across multiple evaluation metrics.Compared to the baseline YOLOv8n,the improved model achieves a 2.3%increase in precision,a 1.9%increase in mAP@0.5,and a 3.5%increase in mAP@0.5-0.95.Overall,the proposed model dem-onstrates superior detection performance in terms of precision,recall,and mAP@0.5,validating its effectiveness for radar-based small target detection.关键词
全息凝视雷达/雷达小目标检测/YOLOv8/ECA注意力机制/小目标检测层Key words
holographic staring radar/radar small target detection/YOLOv8/ECA attention/small target detection layer分类
信息技术与安全科学引用本文复制引用
周昶雯,宋强,张月..基于改进YOLOv8的凝视雷达小目标检测算法[J].信号处理,2025,41(5):853-866,14.基金项目
国家自然科学基金(U2133216) （U2133216）
广东省科学技术厅先进智能感知技术重点实验室科技规划项目(2023B1212060024) The National Natural Science Foundation of China(U2133216) （2023B1212060024）
Guangdong Provincial Department of Science and Technology Key Laboratory Project on Advanced Intelligent Sensing Technology(2023B1212060024) （2023B1212060024）
