The Research Of Music Sensing Based On Deep Learning

During the treatment of children Autistic Spectrum Disorder(ASD)by the way of music sensing,perceived effects and the development of the disease are mainly reflected in the fluctuation of brain waves.There are inaccuracies in the clinical judgment of EEG.Fortunately,deep learning has great advantages in signal feature extraction and classification.In this paper,the Deep Belief Network(DBN)is used as the theoretical basis.This article takes the DBN in deep learning as the theoretical foundation,and then proposes an approach that combines an optimized Restricted Boltzmann Machine(RBM)feature extraction model with a softmax classification algorithm.This method is used to perform brainwave tracking analysis on children with autism who received different music perception treatments.The purpose is to improve the accuracy of classification and to accurately determine the condition of the disease.One of the reasons for the inaccuracy of EEG manual judgment is that the collected EEG data is doped with a series of noises such as power frequency interference,eye-electron artifacts,etc.In order to provide relatively pure data support for subsequent EEG classifications,this article uses convolution neural network method for the detection of the original signal eyelid artifacts,and use the Hilbert-Huang Transform in combination with FastICA to remove noise.Compared with traditional de-noising methods,CNN's pre-detection method greatly saves the workload of subsequent noise removal.HHT shows great advantages in non-stationary signal processing.FastICA has faster calculation speed and less CPU usage than traditional ICA methods.Through the evaluation of the de-noising effect,it is found that the signal to noise ratio(SNR)after de-noising is improved and the root mean square error(RMSE)is reasonable.Aiming at the shortcomings of the traditional DBN recognition performance,this paper proposes an improved algorithm based on double characteristics of the band energy ratio and the moving average sample entropy.The method firstly extracts the characteristics of both from the sample EEG data according to the extraction process of FBER and MVSE.Then these feature composition matrices are input into the DBN network as representative information of the original wave-forms.Through constant adjustment and optimization of the weight matrix in the model,we get a stable identification model.The simulation results show that this optimization algorithm can effectively improve the recognition performance of DBN,and the accuracy reaches 94% in a certain environment.Compared with other traditional classification methods,it has a better classification effect.