Research On Image Classification Algorithm Based On Deep Learning

As a kind of image classification method,deep learning has been developed in recent years.Its network model structure is complex and the input data samples are limited,which can easily lead to model over fitting,long training time and so on.This paper aims at improving the classification accuracy of the convolution neural network model,shortening the training time of model and improving the overfitting of the model.The researches in this paper include:First of all,the training process of convolution neural network model is studied,and the influence of initialization of weight parameters on the accuracy of classification is analyzed.Then,aiming at the long time of deep learning training,the convolution neural network structure model of multi model fusion is studied.Experiments are carried out on the Caltech-101 data set and the 2017 Baidu image competition data set.The experimental results show that the classification accuracy of convolution neural network with multi model fusion is higher than all of single model.In the end,in view of the low classification accuracy of the training sample data,a data augmentation method is proposed to optimize the classification of the data,to augment the data of the single category with poor classification effect,and to improve theoverall classification accuracy.The paper work shows that convolutional neural network based on multi model fusion and data augmentation method based on optimized classification can improve the accuracy of image classification.