Application Research Of Convolution Neural Network In Image Classification

Image classification based on Convolutional Neural Network(CNN) is one of the applications of deep learning on image processing.The advantage of CNN is that it can make convolution computation with pixels to extract features, which simulates the vision learning method of human brains.And the weights sharing and pooling layer in CNN reduce the number of network parameters largely,which simplify the network structure and improve the learning efficiency.This thesis mainly focus on the network structure design and the parameters optimization.The convolutional neural network’s classification performance depends on network’s layers and parameters.How to design convolutional layers,the number of hidden layers and optimize the convolutional neural network parameters properly are very important parts in convolutional neural network applied research field.The most work in this thesis will be implemented on caffe platform.Firstly,I will design a new simple CNN with 5 layers,and use mnist and cifar-10 to train,test and optimize the proposed CNN.Testing results show that the simple CNN can achieve high accurate rate on mnist,but fails to cifar-10.Then,I will design a complex CNN with 9 layers and use cifar-10 and cifar-100 to train,test and optimize the proposed CNN.Testing results show that the complex network can handle complex images classification tasks like cifar-10,cifar-100.A simple five-layer and a complex nine-layer network’s experimental results show that the network layer of depth influences network’s performance.This thesis study the structure and parameters optimization methods of network and tested on different databases,and conclude some practical rules about how to apply deep learning into image classification.Due to parameters play great roles in alike learning networks,it will be a better guide for solving some engineering problems.The conclusions can also be applied into some other deep networks.Finally,this thesis implements a easy-used image classification demo system based on the practical rules.The system not only shows research achievements,but explains the image classification theory to non-professionals.