Design And Implementation Of Golden Monkey Face Recognition Software Based On Deep Learning

Golden monkey is a national first-class animal that has important research value in environmental ecology and animal behavior research.In the process of protecting and researching of golden monkey,the primary task is to identify the individual of golden monkeys accurately.It is also necessary to identify the advanced attributes of golden monkey: gender and age group.Aiming at the large number of golden monkey images and video data already existing in related departments,the traditional method of human recognition is difficult to meet the heavy task of recognition and analysis.And the existing face recognition algorithms are difficult to apply to the problem of monkey face recognition directly.In this thesis,the research on the golden monkey recognition algorithm based on deep learning is carried out.The rapid and accurate automatic recognition of golden monkey will greatly advance the research and protection of golden monkey.This thesis first elaborates the research significance and related development status of face recognition of golden monkey,then summarizes the related theoretical basis of Deep Learning(DL)and Convolutional Neural Network(CNN),and analyzes the difficulty of recognition due to similarity,varying angles,partial occlusion,etc.Combining with the basic methods and theories of convolutional neural networks in deep learning,neural network structures for individual identification and gender-age group recognition of golden monkey are designed,and the corresponding recognition algorithms are given.Finally,the golden monkey face recognition software basing on Qt open source platform is implemented.(1)Aiming at the difficulties in the individual recognition task of golden monkey and the analysis about the face features of golden monkey,the GKP-Net(Global and Key Part-Net)network model is designed,and a golden monkey face individual recognition algorithm based on GKP-Net is given.This algorithm pays attention to two different ranges of face image including the entire face image and the skin area in it.Two different network substructures are designed to extract the features of different areas of the face image,and we combine with the information contained in the whole face image and the key skin area to identify the individual identity of golden monkey.Experiments show that this algorithm can improve the accuracy of individual recognition of golden monkey.(2)Aiming at the different characteristics of different genders of golden monkey in different age groups,an improved convolutional neural network model FF-Net(Feature Fusion-Net)is designed,and a gender and age group recognition algorithm of golden monkey based on FF-Net is given.Through the analysis of the differences between golden monkey face images of different genders and different age groups,combining with the requirements of practical application for the recognition of gender and age group of golden monkeys,features extracted from different levels of the network are fused to enhance the network's ability to express features.On the basis of the existing loss function,a loss function is designed to solve the problems that samples in the dataset of golden monkey have uneven distribution and different learning difficulties.These improvements enhance the performance of this algorithm.(3)According to the research results of the individual recognition of golden monkey and the gender-age group recognition of golden monkey,a golden monkey face recognition software based on Qt open source platform is designed and implemented.This software combines the Open CV computer vision library and uses the C++ language to develop the function modules,such as image loading,saving,individual recognition and gender-age group recognition of golden monkey.