Music Genre Recognition Research Based On Improved AlexNet

To solve the problem that machine learning model has weak ability to identify music genre features,a Music Genre Recognition model based on Deep Convolutional Neural Network(DCNN-MGR)is proposed.First,the model extracts audio information through Fast Fourier Transformation(FFT),generates spectrums that can be input to the DCNN and slicing the generated spectrums.Next,some improvements have been made to AlexNet.Given that neurons do not learn when the independent variables of Rectified Linear Unit(Re LU)activation function enter the negative range,the Re LU function in the convolutional layer of AlexNet is replaced with the Leaky Rectified Linear Unit(Leaky Re LU)function to solve the neuronal necrosis induced by Re LU function.The Re LU function in the fully-connected layers of AlexNet is replaced with the Hyperbolic Tangent(Tanh)function.The zero mean of Tanh function is used to enhance the effectiveness of network in extracting music features during the iterative process.The Local Response Normalization(LRN)layer with weak generalization ability of music feature recognition is removed from AlexNet,so that the network can be more easily parallelized.The number of output nodes in the first convolutional layer,the second convolutional layer and the third fully-connected layer of AlexNet is reduced to 64,192 and 10,respectively.By cutting down the superfluous parameters,the converged network is accelerated to reduce the training time for network.After AlexNet extracts the fully-connected layer of features,the classifier Softplus is added to improve the effectiveness of feature classification and recognition.The spectral slices above are input into the improved AlexNet for multiple rounds of training and verification to extract and learn music features.A network model for distinguishing music features is finally obtained and applied to test its effectiveness in recognizing music genres.The experimental results show that the improved AlexNet is superior to AlexNet,VGGNet and other commonly used DCNN in terms of accuracy of music feature recognition and network convergence effect.The DCNN-MGR model is 4%-20% higher in music genre recognition accuracy than Support Vector Machine(SVM),K-Nearest Neighbor(KNN)and other machine learning models.There are 30 figures,16 tables and 64 references in this paper.