Face Attribute Recognition Based On Multi-label Learning

As one of the core research topics in the field of artificial intelligence,deep learning has made rapid development in recent years.Multi-label learning combines the theoretical foundations of stochastic processes,support vector machines,regularization,and deep learning for multi-label learning modeling,which belongs to a deep learning method.At present,it has been widely used in the fields of speech processing,image processing,data mining and disease prediction.This paper mainly studies the face attribute recognition algorithm based on multi-label learning.The traditional single-label learning cannot meet our needs,and the current face attribute recognition algorithms based on multi-label learning have the following problems:First,the face attribute recognition algorithm based on multi-label learning is in the training process between tags and tags.Restrictions,it is impossible to get good performance for every classification.Second,the recognition accuracy of face attribute recognition algorithms based on multi-label learning on attributes such as age is generally low.Third,the amount of labeling of multi-label data is very large,and the existing data sets have artificial mislabeling and mislabeling,which makes the model unable to converge well during the training process and affects the final accuracy of the model.To this end,this paper uses the combination of LOSS-based data filtering,attention mechanism,network shunting and step-by-step training to solve the problem of low age recognition accuracy in the existing face attribute algorithms.The main work of this paper includes the following aspects:(1)The overall process of implementing the algorithm is introduced and the LOSS-based data filtering method and the strategy of increasing the attention mechanism at the input are proposed to make the network focus on the specific characteristics of the image for training.(2)A multi-label implementation method for network shunting based on face attributes is proposed.This problem transforms single-label learning into multi-label learning by means of network shunting.A step-by-step training method is proposed.The problem is solved step by step,so that more difficult learning tasks can also obtain better recognition accuracy.(3)A method for implementing network shunting and multi-labeling based on face attributes is proposed,which solves the problem of mutual restriction between tags and tags in the training process of face attribute recognition algorithms.A step-by-step training method is proposed,which is to solve the problem step by step through a learning method that is difficult first and then easy,so that more difficult learning tasks can also obtain better recognition accuracy.(4)The effectiveness of the proposed face attribute recognition algorithm based on multi-label learning is verified.On the four data sets of Adience,UTKFace,Multi-PIE Extended-ear and FERET,the proposed algorithm is compared with the current popular multi-label face attribute recognition methods,and the experimental results prove that the proposed method is more effective.In short,the multi-label learning method combining LOSS-based data filtering,attention mechanism,network shunting and step-by-step training effectively improves the accuracy of face attribute recognition.