Improvement And Application Of Convolutional Neural Network

Since Yann LeCun et al.proposed the first mature convolutional neural network(LeNet-5)in 1998,convolutional neural network has attracted great attention from industry and research areas,but it can not be widely used due to the limitations of computer memory and hardware at that time.In recent years,with the extensive use of GPU and the further development of deep learning,convolutional neural network has been applied in image,character,speech and so on.A lot of breakthroughs have been achieved.This paper studies the theory,structure,training and classical model of convolution neural network.Basis on the loss function of classical convolution neural network,the regular constraints of Triplet Network are added.An improved convolution neural network algorithm based on the constraints of Triplet Network model is obtained.In addition,the features extracted by convolution neural network are combined with the features extracted by wavelet scattering.It is applied to image retrieval.The specific work is as follows:1.The background,structure,principle and classical network model(LeNet-5)of convolutional neural network theory are introduced in detail.2.The improvement of traditional loss function based on convolution neural network is introduced,and the validity of this method is verified on handwritten characters.3.The feature obtained by convolution neural network and the feature obtained by wavelet scattering are fused by vector stitching and applied to the field of image retrieval.