Research On Multi-label Image Classification Algorithm Combined With Dictionary Learning

In the first two decades of the 21 st world,due to a wide range of practical applications and scene needs,multi-label learning has attracted the attention of many researchers in the field of data mining and machine learning.As the number of features increases,the problem of multi-label classification becomes challenging,especially when there are many interdependent features and labels.Existing methods learn from multi-label data by manipulating the same feature set,that is,using the instance representation of each example in the discrimination process of all category labels.In order to maximize the use of this information and effectively extract feature information,it is necessary to propose a more innovative classification algorithm and hierarchical embedding model to solve multi-label image classification tasks and improve the accuracy of image classification tasks.How to design an effective multi-label classifier and cope with and deal with the lack of data labels is a major difficulty in image multi-label classification.In this article,we directly try to solve the above core problems in a unified learning framework,thus proposing a new multi-label image classification algorithm to achieve the repair work of incomplete label matrix and classification prediction of multi-label data,The purpose is to improve the classification effect.The innovation of the method proposed in this paper is that on the one hand,the joint learning of independent classifiers is realized,on the other hand,the joint learning of multi-label classification and label correlation,and the joint learning of dictionary learning and label correlation are achieved in order to achieve In order to deal with the problem of missing some labels and the problem of accurate classification of multi-label samples,in addition to this,the model mentioned in this paper also uses a hierarchical structure of hierarchical embedded classifiers,and each branch implements an independent classifier,This structure is conducive to achieving fine classification of samples.The algorithm proposed in this paper also uses the low-rank structure of the matrix.The specific method is to use the label correlation to implement the operation of repairing the incomplete label matrix of the original data.Help to further improve the accuracy of multi-label classification.The multi-label dictionary classification algorithm combined with hierarchical embedding proposed in this paper builds a unified learning framework that can jointly perform classification model and label learning,complete label repair and data classification,because they can affect each other in the synchronous learning process And promotion,so a more accurate classification model can be obtained.In addition,the additional information contained in the label correlation is used to effectively solve the problem of incomplete label information classification.Experiments conducted on subsequent multi-label data sets verified the label The robustness of the algorithm when the information is incomplete.After completing the design of the multi-label classification model algorithm and setting up the relevant experimental configuration,in the experimental link arrangement,4 common data sets l were used to participate in the experiment.In order to compare the experimental effects,this article selects 4 kinds of multi-label Multi-label classification methods that are representative of the field to compare.In addition,this paper uses a variety of different evaluation criteria for multi-dimensional results evaluation in the experiment,and combines optimization solutions to achieve the optimization problems involved in the method proposed in this paper.In the follow-up experimental comparison,the experimental results strongly indicate that,among the results of multiple evaluation criteria,the performance of the algorithm proposed in this paper is obviously superior to the other four multi-label classification algorithms.The detailed results strongly suggest that the advantage of this method is that it can work in a semi-supervised environment,that is,it can learn from data sets that contain some unlabeled data.Compared with other methods,even in the missing part The robustness of the algorithm has a remarkable performance under the bad situation of tag information.