Research On Gender And Age Of Human Face

With the increasing development of social media,the need to obtain facial gender and age attributes has also gradually increased.The application of face gender and age information in a human-computer interaction system brings many conveniences to people's lives.Based on deep learning methods,this paper obtains information from a large amount of face data,extracts features,establishes a network model,constructs a loss function,and finally realizes intelligent recognition.The main experimental content and innovation of the paper are as follows:(1)The main research work and innovation of this paper are as follows:An asynchronous cascaded convolutional neural network for face detection and keypoint positioning is designed.During the training process,the feature extraction of different cascaded networks is realized by designing the size of the convolution kernel and the maximum pooling step.Use the full convolution layer in the output of the first cascade network,and use the full link layer in the last two networks,so that the size of the input picture is not limited;at the same time,the idea of non-maximum suppression(NMS)is used to transform non-face data The coarse and fine culling is performed with overlapping face data to ensure the robustness of small face detection.The experimental results show that the real-time detection frame rate of the network designed in this paper can reach 40 ? 50 fps,and the minimum detectable face size is 12 × 12 pixels.(2)A multi-task shared convolution method based on deep learning is proposed.After sharing the backbone network,the output layer simultaneously outputs the results of gender and age.The Res Net network is used as the main body,and Group Convolution is used instead of ordinary convolution,which improves the feature extraction capability without increasing the amount of parameters;the SE-Net module is added to learn the feature relationship of different channels;and Bottleneck is selectively added for different layers of Res Net To achieve the effect of removing high-frequency noise.The Mobilenet network is used as the main body and improved.Deeply separable convolution is used instead of draw pooling.While ensuring a certain accuracy rate,the network operation speed is accelerated.In this paper,the age problem is transformed into a classification problem.Each age is divided into two categories.A CNN subnetwork and a corresponding 100-level sequence softmax are constructed.The score of each binary classification represents whether the age of the picture is higher than this age category.If it is higher,it is set to 1,and then the softmax fusion is performed on the output end to obtain the final recognition age.(3)On the server side,cascade the face detection network and the gender and age network to make the final network output gender and age information in real time;correct the face according to the detected face frame and key point coordinates to improve the accuracy of gender and age detection Robustness;design server-side multi-threaded video stream processing and multi-process model loading to optimize the overall structure,support multi-channel video real-time detection;design UI interface and background data interaction,interface super parameter adjustable,real-time observation effect,multi-channel display the gender and age information processed by the algorithm.