Pet Dogs RecognitionBased On Deep Convolutional Neural Network

On June 26,2008,the World Chinese Weekly reported that there were about 200 million stray dogs in the word and about 40million in China.These stray dogs are both pitiful and horrible,the stray dogs in many parts of the country injured people in an endless number of incidents,lightly wounded and bleeding,and seriously caused death.Therefore,the management of stray dogs is particularly important.If people can first identify the breed of stray dogs through mobile devices,will be conducive to the management of stray dogs on the characteristics of the life,the aggressiveness of people have an overall cognition,conducive to the management of stray dogs for classification processing.In this paper,a based on deep learning of pet dog recognition model is proposed which can accurately recognize 178 breeds of pet dogs.The main contents are as follow:(1)Studied the current situation of image recognition at domestic and foreign,including traditional image processing methods which combined with machine learning and based on deep learning methods,and their advantages and disadvantages are analyzed and discussed.In addition,the theory of convolutional neural networks,network structural features,the flow of convolutional neural networks for image recognition,and the optimization method of the network are presented.(2)Downloaded 178 breeds of pet dog images from the Internet through crawler technology.The number of pictures of each breed of pet dog is about 60 as a training set for deep learning.In addition,data augmentation is also performed through data augmentation methods,and it is verified through experiments that data augmentation can greatly improve the accuracy of the model and the generalization ability of the model.(3)Because pet dog images have the characteristics of "small differences between classes and large differences within classes",several mainstream deep learning models do not perform well in recognizing dogs.Therefore,an attention recognition model based on residual network is proposed,and experimental verification shows that this model has a great accuracy improvement over other model.