Research On Face Recognition Express System Based On Siamese Network

The traditional pick-up parcel method is easy to falsify or lose,and the pick-up process is cumbersome,leading to problems such as pretending to pick-up a parcel,picking up accidentally the wrong parcel,and its process is slow.By substituting face recognition for traditional pick-up proof.The siamese network is trained with a custom few sample dataset to solve the problem of traditional signing in and the problem of performance degradation caused by insufficient samples when the industry trains the traditional model of face recognition.Finally,according to the pick-up needs in daily life,use Django to build a face recognition express system,and integrate the model into the system pick-up function to realize picking up the parcel by face recognition.The specific works are as follows:(1)Aiming at the model performance degradation problem that most dimensions of prototype vector close to zero in the middle and later period of training,investigates the research status of face recognition and the status of small sample learning field,and uses the unsupervised momentum contrast training method to obtain a dictionary queue containing a large number of prototype vectors.By setting a large momentum value,the prototype vector can be stable in the whole training process.Combined with the accuracy of the LFW test set and the visualization of model parameters,the results reflect that there is no gradient disappearance in the training period,so that the parameters of the model are updated effectively,and the problem of model degradation is solved to a certain extent;(2)Aiming at accelerating the convergence speed of the face recognition model and updating the front-end parameters of the model under gradient reverse transfer,channel and spatial attention are introduced into the inversion residual block of classical Mobilefacenet to improve the stable attention ability of the model to the local key information in the feature graph,to speed up the convergence speed of the model The attention mechanism is added to analyze the influence of the reverse transfer of gradient on the updating of the front,middle and back-end parameters of the model.Combined with the accuracy of the LFW test set and the visualization of model parameters,the results show that the model with attention mechanism converges faster,and the accuracy of the model added in the middle and back-end will be higher;(3)Aiming at problems such as pretending to pick-up a parcel,picking up accidentally the wrong parcel,and its process is slow.By comparing the similarities and differences between the traditional sign-in and face recognition sign-in process,an express system based on face recognition using Django and My SQL has been constructed.By testing the information storage function and face recognition pick-up function of the express system,this paper shows The ability of the express system to deal with the different matching relationship between the receiver and the parcel proves that the face recognition express system based on Siamese network can meet the delivery requirements in daily life.