摘 要：采用深度学习方法抑制毫米波雷达时频域干扰面临着实测样本不足的问题，数据集的大小和质量会影响模型的干扰抑制性能和泛化性能。文章提出一种改进 DCGAN 算法来生成毫米波雷达时频域干扰图像，以扩充训练深度学习干扰抑制模型的实测训练样本。改进 DCGAN 算法对网络结构做出调整，使用带梯度惩罚的 Wasserstein 距离替代 DCGAN 的损失函数。实验结果表明，在原始仿真数据集中加入改进 DCGAN 算法生成的样本，能够有效提高 CNN 模型的干扰抑制性能。
关键词：毫米波雷达；干扰抑制；深度学习；DCGAN
DOI:10.19850/j.cnki.2096-4706.2022.013.014
中图分类号：TP391 文献标识码：A 文章编号：2096-4706（2022）13-0055-07
Research on Generation of MMW Radar Mutual Interference Time-Frequency Image Based on Improved DCGAN —A Case of Performance Effect of the Generated Samples on the CNN Interference Suppression Model
CHEN Zewei, YAN Yuanpeng
(Guangdong University of Technology, Guangzhou 510006, China)
Abstract: Using deep learning method to suppress the MMW radar time-frequency field interference is faced with the problem of lacking actual measurement samples. The size and quality of datasets will affect the interference suppression and generalization performance of the model. This paper proposes an improved DCGAN algorithm to generate the time-frequency field interference images of the MMW radar to expand the actual measurement training samples for training deep learning interference suppression model. The improved DCGAN algorithm adjusts the network structure and the Wasserstein distance with gradient penalty is used to replace the loss function of DCGAN. Experimental results show that adding samples generated by the improved DCGAN algorithm into the original simulation dataset can effectively improve the interference suppression performance of CNN model.
Keywords: MMW radar; interference suppression; deep learning; DCGAN
参考文献：
[1] CARAZO B J，RUFAS D C，BIEMPICA E，et al. Resourceconstrained machine learning for ADAS：a systematic review [J].IEEE Access，2020，8：40573-40598.
[2] NIDAMANURI J，NIBHANUPUDI C，ASSFALG R，et al. A Progressive Review-Emerging Technologies for ADAS Driven Solutions [J]. IEEE Transactions on Intelligent Vehicles，2022，7（2）：326-341.
[3] RISTEA N C，ANGHEL A，IONESCU R T. Fully Convolutional Neural Networks for Automotive Radar interference Mitigation [J/OL].arXiv：2007.11102 [eess.SP].[2022-05-18].https:// arxiv.org/abs/2007.11102v1.
[4] RISTEA N C，ANGHEL A，IONESCU R T. Estimating the Magnitude and Phase of Automotive Radar Signals Under Multiple Interference Sources With Fully Convolutional Networks [J/OL].arXiv：2008.05948 [eess.SP].[2022-05-17].https://arxiv.org/abs/2008.05948.
[5] GOODFELLOW I J，POUGET-ABADIE J，MIRZA M，et al. Generative adversarial nets [C]//The 27th International Conference on Neural Information Processing Systems Advances in neural information processing systems.Montreal：MIT Press，2014（2）：2672-2680.
[6] LEDIG C，THEIS L，HUSZÁR F，et al. Photo-realistic single image super-resolution using a generative adversarial network [C]//2017IEEE conference on computer vision and pattern recognition （CVPR）.Honolulu：IEEE，2017：105-114.
[7] DONG C，LOY C C，HE K M，et al. Image super-resolution using deep convolutional networks [J].IEEE transactions on pattern analysis and machine intelligence，2016，38（2）：295-307.
[8] CHEN Z，TONG Y. Face Super-Resolution Through Wasserstein GANs [J/OL].arXiv：1705.02438 [cs.LG].[2022-05-19]. https://arxiv.org/abs/1705.02438v1.
[9] ISOLA P，ZHU J Y，ZHOU T H，et al. Image-to-image translation with conditional adversarial networks [EB/OL].[2022- 05-18].https://www.zhangqiaokeyan.com/open-access_resources_ thesis/0100091258873.html.
[10] KIM T，CHA M KIM H，et al. Learning to Discover CrossDomain Relations with Generative Adversarial Networks [J/OL].arXiv： 1703.05192 [cs.CV].[2022-05-19].https://arxiv.org/abs/1703.05192.
[11] BROCK A，DONAHUE J，SIMONYAN K. Large Scale GAN Training for High Fidelity Natural Image Synthesis [J/OL].arXiv： 1809.11096 [cs.LG].[2022-05-19].https://arxiv.org/abs/1809.11096.
[12] WANG Q，FAN H J，SUN G，et al. Laplacian pyramid adversarial network for face completion [J].Pattern Recognition：The Journal of the Pattern Recognition Society，2019，88：493-505.
[13] RADFORD A，METZ L，CHINTALA S. Unsupervised representation learning with deep convolutional generative adversarial networks [J/OL].arXiv：1511.06434 [cs.LG].[2022-05-17].https://arxiv. org/abs/1511.06434.
[14] GULRAJANI I，AHMED F，ARJOVSKY M，et al. Improved Training of Wasserstein GANs [J/OL].arXiv：1704.00028 [cs. LG].[2022-05-19].https://arxiv.org/abs/1704.00028.
作者简介：陈泽伟（1995—），男，汉族，广东潮州人，硕士研究生在读，研究方向：车载毫米波雷达信号处理、深度学习及其应用等；严远鹏（1995—），男，汉族，江西赣州人，硕士研究生在读，研究方向：车载毫米波雷达信号处理。