Research On Image Classification Algorithm Based On Convolution Neural Network

With the accelerated development of global technology and the rapid popularization of artificial intelligence in the global scope,people from all kinds of intelligent devices,more and more image data acquisition and storage,so people thirst for image processing and classification is becoming more and more obvious,therefore,take an effective image classification processing mechanism and good classification accuracy is an inevitable trend.It is also this growing demand that pushes deep learning to the peak of the peak,and various deep learning frameworks emerge,such as the Tensor Flow and Caffe of Google.At the same time,it is an important basis and premise for deep learning and artificial intelligence application to classify images well,and it is also a reasonable demand for data processing in our daily life.At the same time,it is an important basis and premise for deep learning and artificial intelligence application to classify images well,which is also the reasonable demand for data processing in our daily life.Using convolution neural network to classify images is an important application of deep learning about image processing.it has directly to a convolution convolution layer,so you can directly with the target image pixels of convolution,and extract features of target image.Because of the unique nature of convolution neural network,it is very suitable for image analysis and processing direction.CIFAR-10 is a data set for object recognition.CIFAR-10 collected the advantage of the universality of the object,and applied to the classification of CIFAR-10 was chosen because it is complex enough to verify algorithm most of function,and can be expanded into a larger model.At the same time,due to the smaller model,the training speed is very fast,which is suitable for testing new ideas and testing new technologies.When the convolution neural network is applied to the image,because the weight sharing attribute and the pooling layer make the network need to train the parameters greatly reduced,the time complexity,the computation amount and the operation time are reduced correspondingly.Initialization method based on the improved weight and pooling operation process such as classifying CIFAR-10,the experimental results show that Compared with the traditional classification method of convolutional neural network,the method proposed in this paper has a great improvement in accuracy and speed.Because of the parameters of convolutional neural network and the limitations of the model,the more network layers in the more deep-seated network,the more the network can not improve the performance of the network.Sometimes the more the network layer is,the more performance is reduced.And in the process of the network training error and the design of network layer,gradient increased with the increase of network layer gradually degradation problems,regarding this,this paper puts forward an improved weights initialization method and the pooling and operating process of the integration of residual network undertake inhibition of gradient process,raise the depth of the convolution neural network is used to analyse the image classification accuracy,the results show that the traditional method in this paper,the improved method can improve the accuracy of image classification partly,follow-up of deep learning efficient work has obvious significance.