Research And Application Of Image Classification Method Based On Deep Learning

Image classification is an important part of image analysis and processing and a hot topic in the field of computer vision.Traditional image classification algorithms need to design features manually,lack of good generalization,and have certain limitations.In recent years,deep learning has been widely used in the field of image classification due to its powerful feature extraction ability.Therefore,based on the classical deep learning model,this dissertation applies the improved method to the classification of different image data sets,aiming to improve the accuracy of image classification.The main research contents and relevant achievements are as follows.(1)Based on the traditional single channel convolutional neural network,an improved multi-channel convolutional neural network model is presented.The traditional single-channel convolutional neural network will lose some important image features after image convolution processing,resulting in the degradation of the image classification performance of the model.The improved model uses three types of convolution kernels of different sizes to extract features from input images,so that the model can extract richer feature information and reduce the loss of image feature information.Small convolution kernel stacking instead of large convolution kernel,batch normalization,Dropout,L2 regularization,data enhancement and other methods are used to reduce the over-fitting problem of the model.In order to verify the classification effect of the improved model,the improved model,the single-channel model,the multi-channel model and the traditional image classification model were compared based on the CIFAR-10 dataset.The experimental results show that the improved model can extract the global and local feature information of the image well,and solve the overfitting problem effectively.(2)Capsule network model is introduced to classify hyperspectral images.The traditional convolutional neural network model cannot fully extract the feature information of hyperspectral images,but the multi-scale convolution kernel adopted by the improved capsule network can effectively improve the detail extraction of the feature information of hyperspectral images,and can fully extract the spatial location relationship of the features.The improved model was used to classify Indian Pines and Pavia University hyperspectral datasets,and the results were compared with those of other classification methods.The classification results show that the improved capsule network model has good generalization ability and can fully extract image feature information,thus improving the accuracy of classification.(3)An improved convolutional neural network model for Chinese painting image classification is presented.The process of traditional painting image classification method is complicated and requires professional knowledge,which makes the feature information of the image cannot be fully extracted.Inception module is added to the model in this dissertation to extract the multi-scale features of the image,while residual connection is introduced to make full use of the underlying feature information of the image.In order to verify the feasibility and effectiveness of the improved model in the classification of Chinese painting images,the improved model,traditional convolutional neural network model,Lenet model and HOG+SVM algorithm are used to classify Chinese painting images.The experimental results show that the improved model can effectively extract the feature information of Chinese painting images,reduce overfitting,and improve the classification accuracy of the model.This dissertation analyzes some deficiencies of the deep learning model,improves it,and applies it to each image dataset,which has certain reference significance for solving image classification problems by using the deep learning model in the future.Figure[28]Table[9]Reference[101]...