基于量子-经典混合残差网络的时频信号识别  

Recognition of time-frequency signals based on quantum-classical hybrid residual networks

作　　者：王港曦 李永梅 潘成康[1] 赵文定 谢宁一 朱赟[3] 崔春风[1] WANG Gangxi;LI Yongmei;PAN Chengkang;ZHAO Wending;XIE Ningyi;ZHU Yun;CUI Chunfeng(China Mobile Research Institute,Beijing 100053,China;Huayi Boao(Beijing)Quantum Technology Co.,Ltd.,Beijing 100176,China;School of Physics and Electronic Information,Gannan Normal University,Ganzhou 341000,China)

机构地区：[1]中国移动通信研究院,北京100053 [2]华翊博奥(北京)量子科技有限公司,北京100176 [3]赣南师范大学物理与电子信息学院,赣州341000

出　　处：《中国科学:物理学、力学、天文学》2025年第4期193-207,共15页Scientia Sinica Physica,Mechanica & Astronomica

摘　　要：量子机器学习是机器学习与量子信息交叉的前沿领域,量子计算为机器学习提供了全新的范式,在扩展应用场景和提升学习能力方面展现出巨大潜力.本文提出了一种量子-经典混合残差神经网络(QCH-ResNet),用于雷达识别场景下的时频信号多任务识别.该模型融合了参数化量子线路与经典残差神经网络,在高噪声环境下实现了精准的分类与检测.实验结果表明,QCH-ResNet在分类精度和噪声鲁棒性方面显著优于传统残差神经网络,充分体现了量子计算在信号处理中的应用潜力.本研究不仅拓展了量子计算在实际应用中的边界,还为复杂时频信号分析任务提供了创新性解决方案.Quantum machine learning,positioned at the confluence of machine learning and quantum information science,offers transformative paradigms that enhance computational capabilities and broaden application domains.In this study,we propose a quantum-classical hybrid residual neural network(QCH-ResNet)tailored for multi-task recognition of timefrequency signals in radar applications.The QCH-ResNet integrates parameterized quantum circuits with classical residual neural networks,enabling accurate classification and detection even under high-noise conditions.Experimental results reveal that QCH-ResNet outperforms traditional residual neural networks in classification accuracy and noise resilience,highlighting the potential of quantum-enhanced signal processing.This work not only expands the scope of quantum computing applications but also provides a novel framework for analyzing complex time-frequency signals.

关 键 词：量子机器学习 量子-经典混合残差神经网络 雷达识别 时频信号分析 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]