Research And Design Of Face Recognition Access Control System Based On Deep Learning

With the rapid development of deep learning,the research and application of face recognition have received more and more attention in this era of big data.In terms of security monitoring,human-computer interaction,etc.,there are more and more applications related to face recognition.At present,most face recognition systems based on deep learning will choose to place the task of computing face features on the server side due to the large amount of computation.But this kind of networking work is greatly affected by the network,and often the performance is not stable enough.In addition,most of the current face recognition systems do not have a face anti-spoofing algorithm deployed,or use a very simple face anti-spoofing algorithm to complete the live detection,but this method is easy to be attacked by illegal users and brings great harm to legitimate users.In order to develop a safe,efficient and stable face recognition access control system.This article focuses on the introduction and research of face recognition and face anti-spoofing technologies.First,based on the existing face anti-spoofing technology and deep classification network technology,this paper proposes a dual-flow registration face anti-spoofing method,and a series of experiments have verified that this method is effective.Then during the practice of the internship,by using depthwise separable convolution,collecting and cleaning a large number of data sets,and using transfer learning methods on the target domain data set,we obtained a face recognition model with high accuracy and a small amount of calculation.Finally,this paper combines the research content of face anti-spoofing and face recognition technology to design and implement an offline face recognition access control system with high accuracy,good stability and fast response time.The system uses a forward inference framework specifically designed for mobile devices such as ARM.This framework deviates from Caffe's native framework and reduces the forward calculation time to one-tenth of the original,greatly improving the system's recognition time.In addition,the interface of the system is clear and simple,and the user experience is good.