Research And Application Of Multi-task Face Attribute Recognition Based On Deep Learning

With the continuous development of the Internet and the popularity of image acquisition devices,the related issues in the field of computer vision have received more and more attention.In many computer vision application scenarios,such as face recognition,autopilot,and medical image detection,the end-to-end training methods of convolutional neural networks has proved to be more efficient than traditional manual extraction features such as Gabor filtering,local binary pattern(LBP)features,biomimicry(BIF)features,etc.Especially in the rapid development of computer hardware equipment in recent years,GPU as computing power has greatly increased,and convolutional neural networks have demonstrated unparalleled advantages in the field of image recognition.Based on the existing research work,this paper uses the convolutional neural network in deep learning to study the relevant issues of face attribute recognition,which mainly involves the attributes of face age estimation,gender recognition,and face recognition.This article from the perspective of multi-task learning,design and improve the structure of convolutional neural networks to improve the accuracy of face recognition,mainly related to the following three aspects:a)For different properties of attributes,different loss functions are used to fit the distribution of attributes on the data set.Comparing the performance improvement brought by multiple loss functions,the final experiment shows that the depth expectation method is more suitable for the data distribution on the current data set.b)Compare the performance differences in age estimation for different network structures(mainly related variant structures of Resnetl8 and Resnet80).c)Taking into account the correlation between attributes in attribute recognition,based on the multi-task learning convolutional neural network,embedding loss functions of different attributes in the same network,using end-to-end training,and testing the model analysis.The experimental results are better than the single-attribute model(face age estimation model).