【作者】 崔明哲； 陈韬； 罗晨； 谢磊； 苏宏业；

【Author】 Mingzhe Cui;Tao Chen;Chen Luo;Lei Xie;Hongye Su;State Key Laboratory of Industrial Control Technology,Zhejiang University;

【机构】 浙江大学工业控制技术国家重点实验室；

【摘要】 本文提出了一种基于深度学习的轻量化自适应时频分析网络,为非平稳信号提供了高分辨时频表达。网络首先通过短时滤波模块自适应学习基函数并与信号内积以生成时频特征映射,然后将轻量化通道注意力机制嵌入模块,重新放缩时频特征以提高后续能量集中性能,之后通过能量集中模块实现信号的高分辨率时频表达。此外,通过对网络结构以及卷积方式的调整实现网络整体模型的轻量化。最后通过一系列模拟仿真与真实信号仿真,验证了该网络的优越性与泛化性。更多还原

【Abstract】 a lightweight adaptive time-frequency analysis network based on deep learning is proposed to provide high-resolution time-frequency representations for nonstationary signals.The network first adaptively learns basic functions through a shorttime filtering module and generates time-frequency feature mappings by taking inner products with the signal.Then,the lightweight channel attention mechanism is embedded into the module to rescale the time-frequency feature to improve the subsequent energy concentration performance.Subsequently,the high-resolution time-frequency representation of the signal is achieved through the energy concentration module.In addition,the overall network model is lightweight by adjusting the network structure and convolution method.Finally,a series of simulated and real signal simulations validate the superiority and generalization of this network.更多还原