Image Recognition Based On Improved Convolutional Neural Network

The rapid development of Deep Learning technology has greatly improved the performance of Machine Learning algorithms.Deep Learning that are composed of multiple processing layers can learn the feature of abstract data with multiple levels.Deep Learning uses the back propagation algorithm to indicate how a network model should change its internal parameters that used to compute the feature in each layer from the feature in the previous layer,discovering complex structure in large data sets.The Convolutional Neural Network(CNN)in Deep Learning has been widely used in the field of computer vision and has become an important weapon for image recognition.The focus of this paper is to optimize and improve the Convolutional Neural Network model.Firstly,Knowledge Distillation is introduced to compress the complex model into a simple model,so that the recognition accuracy of the simple model after distillation learning is higher than that of the simple model trained by the traditional method.Then,the Convolutional Neural Networks have good feature extraction capabilities for image data,but are not optimal for recognition classification.However,Ensemble Learning has good classification performance and high recognition accuracy,but it is difficult to achieve good classification effects for the complex feature of image data.Therefore,this paper proposes a classification method combining with Deep Learning and traditional Machine Learning techniques.In this paper,the training dataset is cifar-10.Firstly,the simple model AlexNet is trained by the traditional method.Then the VGG16 model is pre-trained to obtain the soft target,which as the prior probability guides the AlexNet model training.The recognition accuracy is obviously improved.After an image recognition method combining multi-layer features of convolutional neural network and XGBoost algorithm is used.AlexNet model after distilling learning extracts multilayer feature information,using the serial fusion features and principal component analysis(PCA)to reduce the features dimension.The feature vector with more important information uses the XGBoost classifier instead of the softmax classifier of the AlexNet model.The Convolutional Neural Network improves the recognition performance and generalization ability.