| 引用本文: | 李家强，任梦豪，危雨萱，陈金立. 基于轻量化网络的车载雷达目标分类方法[J]. 雷达科学与技术, 2023, 21(6): 597-604.[点击复制] |
| LI Jiaqiang, REN Menghao, WEI Yuxuan, CHEN Jinli. Target Classification Method for Automotive Radar Based on Lightweight Network[J]. Radar Science and Technology, 2023, 21(6): 597-604.[点击复制] |
|
| 摘要: |
| 针对现有车载毫米波雷达目标分类方法存在静止杂波干扰和网络模型复杂度高的问题，本文将Ghost模块和MobileNet相结合设计了G?MobileNet轻量化神经网络，并提出了一套完整的车载毫米波雷达目标分类流程。首先对雷达原始AD采样信号进行向量均值相消处理，滤除静止杂波，再进行二维快速傅里叶变换（FFT）得到目标的距离?多普勒（RD）图像，最后使用G?MobileNet对RD图像特征进行提取及分类得到分类结果。实测数据处理结果表明，所提方法能够消除静止杂波对多普勒特征产生的干扰，且分类网络模型复杂度低，在具有较高的分类准确率的同时节省了网络模型储存空间和运行网络所需内存。 |
| 关键词: 毫米波雷达 目标分类 向量均值相消 距离⁃多普勒图像 轻量化网络 |
| DOI：DOI:10.3969/j.issn.1672-2337.2023.06.002 |
| 分类号:TN957.51 |
| 基金项目:国家自然科学基金（No.62071238） |
|
| Target Classification Method for Automotive Radar Based on Lightweight Network |
|
LI Jiaqiang, REN Menghao, WEI Yuxuan, CHEN Jinli
|
|
School of Electronics and Information Engineering, Nanjing University of Information Science and Technology, Nanjing 210044, China
|
| Abstract: |
| This paper presents a complete process for target classification in automotive millimeter wave radar, addressing the challenges of high network model complexity and the presence of stationary clutter interference in existing methods. The proposed approach combines the Ghost module with MobileNet to design a lightweight neural network called G?MobileNet. First, the radar’s raw AD sampling signal undergoes vector mean cancellation to filter out stationary clutter interference. Then, a two?dimensional fast Fourier transform （FFT） is applied to obtain the range Doppler （RD） image of the target. Finally, G?MobileNet is used to extract and classify the RD image features to obtain the classification result. Experimental results show that the proposed method effectively eliminates the interference of stationary clutter on Doppler features and the classification network model complexity is low. Furthermore, it saves network model storage space and reduces the memory required to run the network, while achieving high classification accuracy. |
| Key words: millimeter wave radar target classification vector mean cancellation RD image lightweight network |
