Face Detection Based On Depth Residual Network

Compared with the traditional face detection method,the face detection algorithm based on the deep learning concept has a great advantage in detection accuracy.It has a deep backbone network,can extract more abstract feature values from the image,the model structure is more complex,and the parameter settings such as the candidate box are more reasonable,so that the model is not affected by the manual parameters,and the recognition results are more accurate.With the development of hardware and software and the different needs of all walks of life,in recent years,the scene of face detection has evolved from a single indoor background to a complex background such as streets,shopping malls and public transport platforms.At the same time,the technical requirements of face detection are also increasing.In the detection image,there are many problems,such as face angle is different,face is covered by masks and other objects,and the number of faces is densely overlapped.The emergence of these problems has brought many challenges to face detection,many traditional detection methods are not competent.Throughout the development of face detection technology,there are many kinds of methods.In recent years,the rapid development of face detection methods based on deep learning is the most representative.Among many convolution neural network models,residual neural network has outstanding performance and excellent effect in feature extraction.Deep convolution network has excellent training effect because of the residual structure,the loss function will not appear gradient explosion and other problems,so it is suitable to be the backbone network of face detection framework.In terms of detection framework,Faster RCNN is a widely used general target detection algorithm,which has outstanding effect in the field of face detection,and the detection speed and accuracy are guaranteed.In order to better solve the current face detection problems,this paper proposes to improve the residual network and the detection framework Faster RCNN.In this paper,the application of convolutional neural network in face detection is studied based on the theoretical basis of convolutional neural network,and then the improvement of RPN network in Faster RCNN is proposed according to the existing problems,so that the detection framework can deal with face images in multi-scale,using Soft-NMS instead of the original screening mechanism,and optimizing the problem of face overlapping occlusion of detection framework.After that,this paper conceives the improvement direction of the residual neural network,introduces the attention mechanism module into the residual learning branch of the residual network,and adds the se part in the residual branch,so that the feature channels can be correlated with each other,and the feature response of the channel can be automatically adjusted,the residual network and the attention model can be integrated,so that the attention mechanism can effectively obtain the details of the face image,and The improved residual network is used as the backbone network of the detection framework.Finally,the specific parameters of the improved algorithm model are analyzed,and the loss function style is designed.The model is trained,tested and optimized on the open data set,and the detection effect of the improved model is given according to the evaluation index.