Research And Application Of Convolution Neural Network

Convolution neural networks have shown good results for handwritten digital recognition early.Due to the limitations of memory and lacking of a lot of training data about hardware at that time,making the network can't expand the larger image,study reduced.Due to the increased availability and computational power of the GPU,2012 Alex Krizhevsky introduced a large data set in the convolution neural network model,reducing the classification error record from 26% to 15%,to a large extent inspired deep network computer vision usage of.Over the past few years,convolution neural network in computer vision has gained a lot of breakthroughs in previous research achievements and results.Its powerful features of learning and classification skills has been a great deal of attention in the whole field.(1)Firstly,the development process and related history and background of convolution neural network are introduced.The theory of the basic structure and operation of convolution nerve is introduced.(2)As the beginning of the deeping learning,introduced the basic structure of AlexNet network,elaborated the advantages of the model,then,large-scale network and depth model one after another was put forward,and then derived convolution neural network learning ability,fine-tuning.(3)With the research and improvement of the convolution neural network more and more,from the structural improvement,regularization,activation function selection of these three aspects of the convolution neural network in the recent exploration made a summary and discussion analysis.Finally,we summarize the related research of the convolution neural network and the latest research results in the application field.Finally,we also clarify the defects and future research of the convolution neural network and the direction of the application.