基于深度混合卷积的多传感器信息快速融合  

Fast Multi-Sensor Information Fusion Based on Deep Hybrid Convolution

作　　者：韩海豹 王孝国 HAN Hai-bao;WANG Xiao-guo(School of Intelligent Engineering,Jinzhong College of Information,Taigu Shanxi 030800,China;College of Agricultural Engineering,Shanxi Agricultural University,Taigu Shanxi 030800,China)

机构地区：[1]晋中信息学院智能工程学院,山西太谷030800 [2]山西农业大学农业工程学院,山西太谷030800

出　　处：《计算机仿真》2022年第5期288-291,298,共5页Computer Simulation

基　　金：山西农业大学博士科研启动项目(2020BQ80);山西省教改项目(J2019259)。

摘　　要：为了解决传统方法对多传感器信息进行融合时存在的实时性与精准性不佳的问题,提出基于深度混合卷积的多传感器信息快速融合方法。采用误差反向传播算法去掉全连接层,设计一种深度混合卷积神经网络优化模型。利用离散小波变换处理多传感器信息的连续信号,重建离散小波逆变换信号。通过误差反向传播算法训练神经网络,调整网络层隶属函数参数与网络权重,经参数学习律离线训练深度混合卷积神经网络模型,实现多传感器信息快速融合。仿真阶段,由信息滤波效果、位移与速度估算值结果验证得出,所提方法降噪质量与融合精度较高,信息融合速度较快,能够满足实际的应用需求,具有良好的可行性。The traditional optimization method has poor real-time performance and low accuracy in the process of multi-sensor information fusion.This paper explores a fast multi-sensor information fusion method based on deep mixed convolution.The error back propagation algorithm was introduced to delete the full connection layer,so an optimization model based on a deep hybrid convolution neural network was designed.Discrete wavelet transform was applied to process the continuous signals of multi-sensor information in order to reconstruct the inverse discrete wavelet transform signal.According to the error back propagation algorithm,the neural network was trained to adjust the network layer membership function parameters and network weight.The deep hybrid convolution neural network model was trained off-line to quickly fuse multi-sensor information via the parameter learning law.In the simulation experiment,the information filtering effect,displacement and velocity estimates were used to verify the effectiveness of the method.The results show that this method has high noise reduction quality,fusion accuracy and rapid information fusion speed.

关 键 词：深度混合卷积 多传感器 信息融合 参数学习律 隶属度函数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]