Research And Application Of Scene Classification Based On Convolution Neural Network

Scene classification is the important research directions in the field of image processing.With the development of technology of computer and the Internet,a large number of image data swarm into our life and work.Facing such a huge image information,the traditional scene classification methods and techniques show a lot of deficiencies.In recent years,the Convolutional Neural Network(CNN)has made many breakthroughs in the field of image processing.It is a process which is simulating learning of human brain,extracting image features directly from image pixels.Then the feature extraction and classifier are combined into a learning framework to classify and recognize the related objects.In addition,the local connection,weight sharing and down sampling of convolutional neural network can greatly reduce the training parameters of the network,simplify the network model,and further improve the training efficiency of the network.In this dissertation,we combined the convolution neural network method to classify the scene,focusing on the complex variability of the scene image and the problem that the generalization ability of the traditional scene classification methods are not strong.Convolution neural network classification performance is mainly determined by the hierarchical structure of the network,so this paper researched the factors which influence the performance of classification,and designed a new model based on the convolution neural network model to apply in scene classification.The specific work is as follows:1.Aimed at selecting the hierarchy problem a shallow convolution neural network model has been designed to finish the scene image classification in the Scene-15 and the SUN-397 data set.It also can be used to study the effects of different size and different number of convolution cores,different activation functions,different sampling methods.The study shows that the neural network which uses a smaller convolution kernel,a larger number of cores,the maximum sampling and the ReLU activation function,can increase the convolution neural network classification performance.2.In order to better adapt to the requirements of the actual scene image,this dissertation has improved the neural network model based on the above research,and designed an 8-layer convolution neural network.The convolution layer of the network uses a smaller convolution kernel and increases the number of convolution cores,which can extract more image features and improve the classification performance.At the same time,the sampling layer uses the maximum sampling method and the Re LUactivation function in this paper.Then the modified convolution neural network model is compared with the Alex Net model and the VGGNet model on the Scene-15 data set and the SUN-397 data set as the contrast.The experimental results show that our model has a good classification effect in scene classification application.This dissertation uses the MatConvNet toolbox to construct the structural design and finish parameter optimization of convolution neural network in MATLAB software,and analyzes the factors that affect the classification performance of convolution neural network.Based on above,a convolutional neural network model was designed and applied to scene classification.A large number of experiments show that our network model has good classification performance in scene classification application and a certain generalization ability.