Research On Image Classification Model Of Improved CNN

Image classification task has always been the hot topic in the field of computer vision.In the face of massive image data,the traditional feature processing method has been stretched out.With the improvement of computing performance for computer,the existing convolutional neural networks solved most of the image classification problems,but the problems of network construction,low classification accuracy,and large calculation consumption need to be solved.In view of the above problems,from the perspective of feature fusion and weight sharing,two improved convolution neural network models are proposed,as follows:First,the existing convolution networks only relies on the learning features to predict the category of the image.This mothod ignores the influence of the artificial design features.An image classification algorithm based on salient features and deep learning is proposed.Firstly,the salient feature and the BGR feature of the original image are fused.Then,the convolution network performs automatic feature learning on the fused multi feature image to extract the multiple features of the image.Finally,the image is classified by softmax classifier.The experiment shows that the algorithm is a new image classification algorithm with strong robustness and good classification effect.Secondly,Most of the existing convolution networks make use of the huge amount of parameters caused by the network depth and lack robustness to scale features of input image.In view of the problems,a weight-shared parallel convolutional neural network(WPCNN)is proposed.Firstly,the image on the training set and the image after zooming are input into the weight-shared parallel network to extract multi-scale input features.Then,in the hierarchical network,multi-layer convolutions are used to extract the same input image features.Next,the idea of feature fusion are used to fuse two features to enhance the robustness of the network for multi-scale features.Finally,the network parameters are optimized to achieve efficient feature extraction.The experimental results on cifar,fer2013 and other datasets show that the proposed model achieves high classification accuracy.And the results prove that the convolution neural network with feature fusion and weight sharing is one of the effective methods to solve the problem of multi-scale image classification accuracy.The paper has 32 pictures,19 tables,and 85 references.