Research On Compliance Detection Of College Entrance Examination ID Photos Based On Deep Learning

The college entrance examination is a national selection examination,which is related to the further education and future development of every candidate.In order to ensure the fairness,impartiality and standardization of the college entrance examination,the examination department needs to conduct strict identity comparison and review of all candidates,including the review and verification of student ID photos.With the rapid development of computer technology,the review of ID photos has also become intelligent,informational and scientific.In order to reduce the workload of examination personnel on candidates’ identity review,the object detection technology is applied to the college entrance examination registration work,and the key points of the face(eyes,nose,mouth and eyebrows)are located,which can not only detect the key points of the face in real time during the process of ID photo collection,but also speed up the identification of the candidate’s identity and the speed of entering the examination room before the examination.However,during the detection of face key points,the detection difficulty will increase due to lighting conditions,occlusion,and low resolution.In view of these problems,this paper takes the YOLOv5 s model as the benchmark to conduct in-depth research on its Backbone network and Neck layer to improve the effect of face key point detection,and the main work is as follows:(1)In view of the problems of low resolution and difficult localization of key points caused by lighting and noise,the CBAM attention mechanism is introduced in the Backbone of YOLOv5 s.The introduction of attention mechanism can strengthen network feature extraction and reduce the influence of noise and redundant information on feature information extraction.CBAM can better fit complex target features by combining spatial attention and channel attention to assign weights to different parts of the feature map,which can improve the detection performance of the model to a certain extent.(2)Aiming at the problem of insufficient feature information caused by partial occlusion of key points of the face,the Bi-FPN structure is introduced in the Neck part of YOLOv5 s,which can better transmit the underlying features to the upper layers,enhance the expression ability of features and the perception ability of targets of different scales.The Bi-FPN structure is introduced into the model to make the key feature map of the face to be detected contain more information and effectively improve the recall rate.(3)In view of the problem of insufficient data,the data is first enhanced by adding Gaussian noise,adjusting brightness and random rotation,and then improving the Mosaic data enhancement method in YOLOv5 s,so as to expand the dataset and improve the network learning ability.Experimental results show that the introduction of attention mechanism,Bi-FPN structure and improved Mosaic data enhancement method in YOLOv5 s not only makes the convergence speed of network training fast,but also improves the performance of the model in face key point detection.