Research On Characteristics Reconstruction Of Label Distribution Learning And Its Application In Emotion Recognition

Label distribution learning as an extended research on multi-label learning has become one of the hot topics of machine learning.In real life,people often necessary to collect a lot of sample data in order to better study marker distribution learning.However,when a lot of sample data are obtained,the similarity between some samples is too high,and there may be noise interference in the collected samples.So,how to reduce redundant samples and avoid noise interference is very important to improve the classification accuracy of label distribution learning.There are many problems in traditional multi-label learning,but few scholars have extended their research to label distributed learning.Based on this,this paper proposes two algorithms to deal with these two problems.The main contents are as follows:(1)Existing algorithms all use conditional probability to build parameter models,but do not consider the links between samples fully.Based on it,the spectral clustering algorithm is introduced.The clustering problem is transformed into the problem of the graph's global optimal partition based on the similarity relationship between samples,thus,a label distribution learning algorithm with spectral clustering is proposed(SC-LDL).Firstly,calculate samples similarity matrix.Then,transform the matrix to construct feature vector space.Finally,cluster the data to establish parameter model with the K-means algorithm,and then this new model is used to predict the label distribution of unknown samples.The comparison between SC-LDL and the existing algorithm on multiple data sets shows that this algorithm is superior to multiple contrast algorithms,and statistical hypothesis testing further illustrates the effectiveness and stability of the SC-LDL algorithm.(2)At present,most of the algorithms are designed with complete data,and not consider noise in the data.Therefore,combining the noise reduction characteristics of the self-encoder and the stability of the kernel extreme learning machine,the Label Distribution Learning Algorithm based on Kernel Extreme Learning Machine with Self-encoder is proposed in this paper.Firstly,we use the self-encoder in kernel extreme learning machine to map the original feature space to obtain more robust feature representation.Secondly,we construct the extreme learning machine model that adapts to the label distribution learning as a classifier to improve the classification efficiency and performance.Finally,the experimental results show the proposed algorithm has certain advantages over other label distribution learning algorithms,and the hypothesis test method further illustrates the effectiveness of the algorithm.Label distribution Learning is closer to the real world than traditional multi-label learning.In order to further study,this paper constructs a face emotion recognition model.Firstly,LBP is used to extract the features of two-dimensional face samples and construct the emotional distribution.Then the KELM classifier is used to predict the emotional distribution.The experimental results show that the constructed emotional distribution is more in line with human's emotional judgment,which indicates that marker distributed learning is a learning method closer to the real world.