【作者】 马捷； 马越； 张若愚； 缪晨； WU Wen；

【Author】 MA Jie;MA Yue;ZHANG Ruoyu;MIAO Chen;WU Wen;Key Laboratory of Near Range RF Sensing ICs & Microsystems,Ministry of Education,School of Electronic and Optical Engineering,Nanjing University of Science and Technology;

【机构】 南京理工大学电子工程与光电技术学院,近程射频感知芯片与微系统教育部重点实验室； Key Laboratory of Near Range RF Sensing ICs & Microsystems,Ministry of Education,School of Electronic and Optical Engineering,Nanjing University of Science and Technology；

【摘要】 本文针对低慢小目标的识别问题,基于LSS-FMCWR-1.0数据集,提出了一种改进的时间卷积网络(TCN)模型,用于实现低慢小目标的高效分类识别。该模型通过多层一维卷积结构提取时序特征,并引入残差连接与Dropout策略,以提升模型的稳定性与泛化性能。将改进TCN与传统机器学习方法(SVM、随机森林、KNN)以及深度学习方法进行了系统对比。从分类准确率、收敛速度和抗噪性能等角度全面评估各模型在低慢小目标识别任务中的性能。实验结果表明:在低信噪比条件下,TCN的性能与传统方法相当,而随着SNR提升,TCN的分类准确率逐步提升并最终优于对比方法。进一步通过ROC曲线分析可知,所提模型在不同目标类别上均表现出良好的判别能力,其中部分类别的AUC达到0.97以上,验证了模型在复杂环境下的泛化能力。更多还原

【Abstract】 This paper investigates radar-based recognition of low,slow,and small(LSS) aerial targets using the LSS-FMCWR-1.0 dataset.We propose an improved Temporal Convolutional Network(TCN) that leverages multi-layer 1D convolutions,residual connections,and Dropout to enhance feature extraction,stability,and generalization.The model is systematically compared with classical machine learning methods(SVM,Random Forest,KNN) and deep learning approaches under varying signal-to-noise ratio(SNR) conditions.Experimental evaluations demonstrate that the improved TCN achieves comparable performance to traditional methods at low SNR,while its accuracy increases steadily with higher SNR,ultimately outperforming other approaches.ROC analysis further indicates strong discriminative ability across target categories,with AUC values exceeding 0.97 for some classes.These results validate the robustness and effectiveness of the proposed TCN,providing a promising solution for practical deployment of LSS target recognition in complex environments.更多还原