Research On Face Recognition Technology Based On Deep Learning And Its Application In Oilfield Operation Area

Deep learning is a branch of machine learning,and it is a method of characterizing learning based on data.For the face recognition problem,the face features obtained by deep learning tend to have better selectivity than the manual features for face identity and attributes.Face recognition,as a biometric technology based on facial features,has been widely used in smart cities,mobile payment,criminal investigation,intelligent security and other fields.This thesis mainly studies face recognition algorithm based on deep learning,and constructs a face recognition system that integrates face detection and recognition.Finally,the performance of the proposed method is verified on different data sets.The main contents of this thesis are:(1)Firstly,the related theories of deep learning are introduced.The mainstream network structure and deep learning framework are expounded.The structure of the convolutional neural network(CNN)used in this thesis is introduced in detail.Finally,the process of face recognition and related Dataset and model performance metrics are presented.(2)Preprocessing the images used in the face recognition model training constructed in this thesis,including face detection,alignment and image augmentation.Among them,many experiments are performed on the face detection algorithm,and the traditional face detection algorithm is based on The face detection algorithm for deep learning is compared to determine the preprocessing part of face recognition.(3)There are three main development directions of face recognition research: data set,loss function and model structure.In this thesis,a lot of experiments were carried out on the loss function which needed to construct the face recognition network.The original ResNet network model was optimized and improved,trained in different data sets,and compared experiments on different test sets.Finally,the network model is improved aiming at the change of face occlusion and illumination,and the final face recognition model is obtained.The results show that the face recognition network model constructed in this thesis achieves the accuracy of 98.71%,95.7%,92.98% on the LFW,CFP,and AgeDB-30 data sets,especially,accuracy of 98.33% was achieved in MegaFace dataset.The rate,which is 7.46% higher than the original model,has achieved the desired target.