| In recent years,the prevalence of brain diseases has increased year by year,and many researchers have begun to apply artificial intelligence to the detection of brain diseases.Existing EEG-based epilepsy recognition algorithms(such as SVM,KNN,etc.)need to perform data pre-processing and feature screening first.The extracted features are often lack of medical interpretation and are prone to dimension explosion.In addition,in the recognition of depression speech,the common MFCC and LPCC speech feature extraction methods have the advantages of single dimension and lack of interpretability.As a multi-layer neural network,convolutional neural network has strong self-learning ability,and it has breakthroughs in speech recognition,face recognition,general object recognition and even brain wave analysis.Based on the above background,based on the characteristics of epileptic EEG signals and depression speech signals,this paper designs an intelligent diagnosis system that can identify epilepsy and depression,and introduces a convolutional neural network algorithm in the system.And the spectrogram construction method.The main contents of this paper are as follows:(1)In view of the limitations of traditional classification algorithms in the screening of EEG signal features,this paper uses a convolutional neural network algorithm that can perform feature self-learning.Based on the EEG data samples of the Third People’s Hospital of Chengdu,the EEG signal is imaged,and the 3D matrix is constructed as a processing unit of 512*16*3 as the input of the convolutional neural network,using tensoflow A 10-layer convolutional neural network was constructed.In order to verify the validity and necessity of convolutional neural networks in the dataset classification of this paper,SVM,KNN and ANN algorithms are used to compare with this initial network model.The results show that the recognition rate of CNN algorithm is 91.32%,which is higher than other networks..In order to further improve the recognition rate,the parameters of the network model were adjusted and optimized.Finally,a 12-layer CNN model was constructed to achieve a seizure recognition rate of 99.8%.(2)For the single dimension of MFCC and LPCC speech feature extraction methods,lack of interpretability,and poor generalization,this paper uses a spectrogram construction method that can take into account the characteristics of time domain frequency domain.Taking the published AViD-Corpus depression sample as the dataset of the classification study of depression disease,this paper draws on the advantages of convolutional neural network in image processing,and constructs the spectral dataset as the input of the network.The influences of parameters such as learning rate,input form and all-connected layer nodeson the network model were analyzed.Three sets of experiments were designed respectively.The network model and specific parameters were determined,and the recognition rate of depression finally reached 86.3%.(3)Finally,two kinds of algorithm models for the classification of epilepsy and depression were established.According to the needs analysis of the intelligent diagnosis system for brain diseases,the design and implementation of the system was completed. |
