Research On Dog Face Breed Recognition Algorithm Based On Deep Learning

With the rapid development of pet industry at home and abroad,family pets represented by pet dogs occupy the main market share,but the management problems caused by pet dogs are becoming more and more prominent,and the identification of pet dogs has become the key to improve the efficiency of pet dog management.Common pet dog identification schemes are limited to a small range of promotion,but dog face recognition technology can not only effectively identify pet dog individual identity information,but also has the advantages of low cost and convenience,which is conducive to a wide range of promotion and application.Based on the above realistic background,this paper mainly studied the dog face breed recognition algorithm and proposed two recognition schemes,the specific work is as follows:In the first part,three kinds of public dog face image data sets are re-screened and divided to make the dog face image meet the task requirements of this paper.A multialgorithm image enhancement scheme combining bilateral filtering algorithm and RGB-HSI transform algorithm was proposed to solve the problems of excessive noise,uneven light and dark and unclear contour of dog face image.The fusion algorithm effectively solves the above problems and obtains ideal image enhancement effects,which lays a foundation for the subsequent model to obtain more effective feature points in dog face images.In the second part,a dog face breed recognition network model based on convolutional neural network is designed.Firstly,batch standardization algorithm,Dropout algorithm and Adam algorithm were used to optimize the designed network model,and 92.03% dog face breed recognition rate was achieved.Then,whether the image enhancement algorithm in the first part is beneficial to improving the recognition rate of the network model is analyzed and explained.Finally,the obfuscation matrix is used to evaluate the classification effect of the model.In the third part,a dog face breed recognition network model based on transfer learning is designed.Firstly,Celeb A face image dataset was used to pre-train the Google EfficientnetB7 model to obtain the source area network model.Then,parameters of the full connection layer and Softmax classification layer of the source area network model were adjusted,and the recognition rate of dog face breeds was 96.83%.Finally,in order to illustrate the advantages of transfer learning in dog face breed image recognition,this paper selects two representative classical face recognition algorithms to construct a traditional dog face breed recognition network model for comparative analysis.