【摘要】 研究并构建了一个结合脑电信号处理与深度学习的车内噪声评价模型,该算法通过自我学习实现脑电信号特征提取,使用同步似然方法构建delta、alpha和beta频段的脑功能网络。将3个频带的脑功能网络扁平化处理后作为输入,通过无监督的堆栈自编码器(RSAE)自主提取脑功能网络的特征。通过几个高阶特征训练前后对比,证实了RSAE自主学习到与噪声评价有关的脑神经特征。最终将RSAE与普遍使用的SVM回归模型进行比较,同时将脑功能网络与传统的基于心理声学声音品质的车内噪声评价进行对比。结果表现,所提出的脑功能网络RSAE模型的平均决定系数高达98.69%,明显优于其他方法。更多还原

【Abstract】 In this paper,an in-vehicle noise evaluation model combining EEG signal processing and deep learning was studied and constructed.This algorithm can realize EEG signal feature extraction through self-learning.The synchronous likelihood method was used to construct the brain function network of delta,alpha and beta bands.For the unsteady characteristics of EEG signals,the synchronous likelihood method does better in finding the linear and nonlinear coupling relationship between different channels.The three frequency bands of functional brain networks were flattened as an input.Firstly,the features of the functional brain network were extracted through unsupervised pre-training,and then the initialization weights were used to train the four-layer network using the in-vehicle noise scores evaluated by the expert group as the annotation information.Through the comparison of several high-order features before and after training,the validity of RSAE’s self-learned features were confirmed to be useful.In the end,this paper compared the RSAE with SVM regression model,and compared the functional brain network with the traditional psychoacoustic sound quality-based interior noise evaluation.The results show that the proposed vehicle interior noise based on RSAE and functional brain network was better.The average decision coefficient of the functional brain network RSAE model proposed is as high as 98.69%,which is obviously superior to other methods.The results also confirmed the feasibility and effectiveness of the proposed model.更多还原