Research On The Improved Algorithm Of Image Classification Based On Convolutional Neural Network

As people continue to explore artificial intelligence,convolutional neural networks,as one of the representative algorithms,have also developed rapidly.Convolutional neural networks extract more high-dimensional and abstract features from the data,summarize the distributed feature representation of the data,and discover complex non-linear relationships.Due to the rapid increase in the amount of computing in the era of big data and the more complex convolutional neural network structure,the difficulty of computing tasks continues to increase.In view of these difficulties,this paper optimizes the convolutional neural network Alex Net model.This article first introduces the basic principles of artificial neural networks and related technologies of convolutional neural networks,and analyzes the development prospects and research directions of convolutional neural network algorithms.Then introduce the convolutional neural network Alex Net model,analyzed and summarized its shortcomings,and proposed an improved network model.In the preprocessing stage,a bicubic interpolation algorithm is used to unify the image to a standard size,and then the image is normalized to maintain a similar distribution of the input data.At the same time,this article optimizes the structure,uses more refined convolution kernels to extract features,uses the receptive field to deform the convolution layer to reduce the number of parameters,abandons the local response normalization operation and grouping strategy,and uses batch normalized data processing method.By using the improved activation function to nonlinearly transform the data,the performance of the network model can be effectively improved.Secondly,this paper improves the Linear Discriminant Analysis(LDA)and proposes a new evaluation index to optimize the sample space.By combining with the network model,the convergence speed can be further improved.Finally,based on the deep learning framework Tensor Flow,the network model is trained and tested.By analyzing the dynamic changes of network model performance,error rate,training time,recognition accuracy and other factors,the experimental results prove that the method in this paper can not only effectively reduce the amount of calculation and the parameter scale can further improve the recognition accuracy of the network model.