Research And Implementation Of Facial Recognition Of Golden Monkey Based On Attention Mechanism

Golden monkeys are national first class protected animals in China and are endangered species that require protection urgently.The protection of golden monkeys and related research are based on the ability to identify the individual accurately.Traditional methods based on artificial identification are expensive,inefficient and susceptible to individual subjective factors,which restricts the progress of related research.In view of the rapid development of computer vision technology,it is a feasible method to automatically process the image data of golden monkeys by using corresponding computer vision methods.Convolutional Neural Network(CNN)has excellent feature extraction ability,and has had significant study in the field of image processing.The attention mechanism plays an important role in image comprehension.Based on the basic methods and theories of CNN,this thesis introduces the attention mechanism and designs a facial recognition algorithm for golden monkeys to realize the automatic identification of individual rapidly and accurately,so as to provide technical support for the research and protection of golden monkeys in the era of big data.Firstly,this thesis expounds the research significance of facial recognition of golden monkeys,and analyzes the development status of related fields.Secondly,it analyzes the relevant basic theories of convolutional neural networks,and expounds the basic research direction of this algorithm.Thirdly,the characteristics of the facial image data of golden monkeys are analyzed.The network structure of the facial recognition algorithm of golden monkeys is designed according to the characteristics of the facial image data,such as complex background,varied light,high degree of individual similarity and so on.The validity of the algorithm was demonstrated through experimental comparison.Finally,based on the Qt platform,golden monkeys facial recognition software is designed and implemented.(1)To solve the problem of monkeys' facial recognition,this thesis designs an improved convolutional neural network model AKP-CNN(Attention Key Part Convolutional Neural Network)based on the analysis of the characteristics of the facial data of golden monkeys,and implements an algorithm based on AKP-CNN.The algorithm uses Attention-ProposalNetwork(APN)to focus on key part areas,and extracts global and local features respectively through two networks with the same structure,then fuses the two features through feature fusion layer for recognition tasks.Based on APN's characteristics of locating and focusing on key part regions,this thesis designs a corresponding feature fusion mechanism by heuristic method in order to utilize different features of multiple scales.Experiments show that the algorithm can improve the accuracy of face recognition of golden monkeys.(2)According to the research on the algorithm of face recognition of golden monkeys,this thesis designs a software for face recognition of golden monkeys based on Qt platform and C++.The software utilizes the computer vision library Open CV to accomplish the loading and processing procedure of images,and uses open source deep learning framework Caffe to load network models and predict image categories.Considering the need of field research,the software is able to sample and save unfamiliar data.The software has a complete graphical operation interface and exception handling mechanism.It realizes many functions such as image reading,model loading,result display and so on.The software is simple to operate and easy to learn.It is an excellent assistant in the work of individual identification of golden monkeys.