摘要
针对船用通信系统光信号通常缺乏大量标记样本,且无标记样本未得到充分利用的情况,提出基于RBM-DSN的船用通信失真信号半监督识别方法。采集光通信信号,并对其进行转化处理;将转化后的光通信信号作为输入,利用限玻尔兹曼机学习挖掘光通信信号潜在特征,利用对比散度算法求解最优参数,并以此初始化深层堆叠网络输入权值,采用多模块堆叠结构,通过各模块独立的有监督训练,实现对光通信信号特征的分层提取,并采用梯度下降法微调网络参数,通过最小化误差函数优化识别效果。实验结果表明:对于不同失真光通信信号,本文方法的识别决定系数高,在样本标记率在20%时,该方法能够兼顾识别精度与计算成本,达到理想的失真信号识别效果。
Aiming at the situation that the optical signal of marine communication system usually lacks a large number of labeled samples and the unlabeled samples are not fully utilized,a semi-supervised identification method of marine communication distortion signal based on RBM-DSN is proposed.Collecting optical communication signals and converting them;Taking the converted optical communication signal as input,the potential features of optical communication signal are mined by Boltzmann machine learning,the optimal parameters are solved by contrast divergence algorithm,and the input weights of deep stack network are initialized.The multi-module stack structure is adopted,and the features of optical communication signal are extracted hierarchically through independent supervised training of each module.The network parameters are fine-tuned by gradient descent method,and the recognition effect is optimized by minimizing the error function.The experimental results show that the identification determination coefficient of this method is high for different distorted optical communication signals.When the sample labeling rate is 20%,this method can give consideration to the identification accuracy and calculation cost,and achieve the ideal distortion signal identification effect.
作者
胡晓光
刘涛
HU Xiaoguang;LIU Tao(School of Software and Communications,Tianjin Sino-German University of Applied Sciences,Tianjin 300350,China)
出处
《舰船科学技术》
北大核心
2025年第23期147-151,共5页
Ship Science and Technology
关键词
深度学习
船用通信系统
失真信号
梯度下降法
特征提取
deep learning
marine communication system
distorted signal
gradient descent method
feature extraction
