基于深度学习的超视距雷达直升机目标检测方法
A Helicopter Target Detection Method of Over-the-Horizon Radar Based on Deep Learning
梁复台 1周焰 2董家隆 3陈新 2唐晓2
作者信息
- 1. 空军预警学院,武汉 430012;解放军31121部队,南京 210000
- 2. 空军预警学院,武汉 430012
- 3. 解放军95980部队,湖北 襄阳 441021
- 折叠
摘要
针对超视距雷达低速目标检测与识别难的问题,提出一种基于深度学习的超视距雷达直升机目标检测方法.根据超视距雷达直升机目标频谱特点,设计一种专用的超视距雷达直升机目标检测神经网络HDNet,仿真直升机目标距离多普勒谱图并构建数据集,以该数据集为源域向实测数据迁移,对网络模型进行训练并测试.经试验验证,HDNet网络对仿真直升机目标检测的AP@0.5可达92.1％,采用深度迁移学习的方法对模型微调后,网络模型对实测数据的检测效果也有了明显提升.
Abstract
Aiming at the difficulty of low-speed target detection and recognition of Over-the-Horizon Radar(OTHR),a helicopter target detection method based on deep learning is proposed.According to the helicopter target spectrum characteristics of OTHR,a special neural network HDNet for OTHR helicopter target detection is designed.The range-Doppler spectrum of the helicopter target is simulated and the dataset is constructed,the dataset is used as the source domain to be transfered to the real measured data.The network model is trained and tested.After the test and validation,the Ap@0.5 of the HDNet network for the simulated helicopter targets detection can reach 92.1％.After fine-tuning the model based on the deep transfer learning method,the detection effect of the network model for the measured data detection is improved obviously.
关键词
深度学习/超视距雷达/直升机目标/距离多普勒图/目标检测Key words
deep learning/OTHR/helicopter target/range-doppler diagram/target detection引用本文复制引用
基金项目
军队重大科研基金资助项目(JY2020A020)
出版年
2024
国家科技期刊平台
NSTL
万方数据