Research Of Image Classification Technology Based On Deep Convolutional Network

In recent years,with the development of computer technology and network,the number of image databases is increasing day by day.It is one of the most important research problems to obtain image information by classifying image data.Image classification has also broadened application prospects in many fields.With the rise of deep learning technology,the accuracy of image classification based on deep neural network has been significantly improved,and this takes the image classification technology to a new era.Among them,the convolution neural network based image classification technology is one of the hottest topics in the multimedia domain.Convolution neural network is widely used in pattern recognition,image processing and other fields recently.It is an efficient and popular algorithm,because it has the advantages such as robust network structure,adaptability and so on.However,in practical application,the classification accuracy still needs to be further improved.In order to promote the precision of image classification algorithm based on depth learning,in this thesis,we firstly study the image classification by Principal Component Analysis of Multi-Channel deep features,and then study the fine-grained image classification method based on convolution neural network.The main work is as follows:(1)Image classification based on multi-channel features and principal component analysis.Most image classification algorithms only extract a single feature of the image.The extracted feature can only describe a few attributes of the image,and the image description is not comprehensive enough,resulting in insufficient information of the image.The classification accuracy of the image is not very high.Therefore,an image classification technology based on Multi-Channel Deep Feature and Principal Component Analysis(PCA)is proposed.Firstly,the feature was extracted from the three RGB channel and gray scale channel of the image,and the features extracted were more distinguishable.Then,the principal component analysis was used to reduce the dimension of the features.Finally,SVM was used to classify the features.The image classification accuracy of the method is higher,and the classification speed is faster.(2)Fine-grained image classification based on convolutional neural network.We studied fine-grained vehicle identification through Deep CNNs,which could identify different models of the same brand.Vehicle and the corresponding parts are localized with the help of RCNN and their features from a set of pre-trained CNNs are aggregated to train a SVM classifier.We create a fine-grained vehicle dataset and conduct a follow-up experiment.The preliminary results show the superiority of this method.Through experimental verification,the above two kinds of classification methods have achieved good classification results.