Research On Generation Method Of Specific Face In Monitoring System Based On Deep Learning

In the application environment of airport,railway station,classroom and other public surveillance systems,the collected face images are often uncontrolled,with a variety of changes.For example,the lack of face image information caused by the influence of face pose,illumination and small occlusion leads to the rapid decline of recognition performance.Some traditional machine learning methods use shallow structure.This kind of neural network can not explore complex functions within the sample,such as BP neural network and support vector machine,and is vulnerable to external factors such as illumination,complex background and so on.Generalization ability is obviously insufficient.In-depth learning can learn and characterize the distributed representation of input data through deep neural network structure,and fit complex functions,which reflects the ability of in-depth learning to extract the deep features of input data.Therefore,this topic chooses the method of generating model based on deep learning to overcome the influence of partial information missing in uncontrolled face images,and to forecast and supplement the missing data.This paper mainly studies the learning principle,training parameters and the effect of face generation of depth generation model(such as Generative Adversarial Network,GANs).The training set is constructed by extracting static face images in public surveillance,normalizing the captured face images and simulating occlusion by mosaic.Using TensorFlow deep learning framework,the processed training set is input into DCGAN,context coder-based DCGAN and Pix2Pix-based generative adversarial network to train and adjust.The model structure is further optimized according to the application background,the trained model is saved and the training results are improved.Line comparative analysis.In the framework of Keras in-depth learning,VGGNet is used to extract and classify the features of students' facial expressions under classroom monitoring,and the methods used in the experimental system are evaluated by comparing the recognition accuracy of facial expressions.The experimental system uses PyQt for research and development.Mosaic processing is carried out on the occluded area of the input face image.Complete face images are generated by simulation.Finally,facialexpression recognition is carried out on the restored face image.The experimental results prove the effectiveness of the algorithm.