Research On Classification Method With Uncertain Shadowed Prototypes

With the development of artificial intelligence,machine learning methods have been widely applied in various fields.But when representing and making decisions on some data with complex distribution and structure,there is a lot of uncertainty in the data.Existing machine Learning methods have some limitations in representation and uncertainty analysis on complex data.Data-driven neural network methods are highly dependent on data volume and data quality,and the models have low interpretability.There are still many challenges in the practical application of machine learning methods.As an effective method on data representation,prototype learning can represent data distribution by directly establishing typical patterns from data,which has good adaptability to complex data representation.However,existing prototype learning methods generally lack uncertainty analysis mechanisms.The uncertainty theory shadowed set can model the uncertainty,and the decision-making method based on uncertainty theory can identify uncertain samples and reduce the decision-making risk,but it has limited ability to represent and process complex data.Based on the complementary advantages of the two methods,this paper proposes to expand the prototypes through the uncertainty theory shadowed set,construct the uncertain shadowed prototypes,and restudy the representation and uncertainty analysis on complex data from the perspective of the uncertain shadowed prototypes.And the classification methods based on uncertain shadowed prototypes are applied to medical data for verification.The main contents and innovations of the paper are as follows:(1)Proposed an uncertainty classification method based on uncertain shadowed prototypes.In order to effectively deal with the uncertainty representation and decision-making of complex data,this paper proposed an uncertain decisionmaking method from the perspective of uncertain prototypes.By reconstructing the shadowed sets to expand the representation of uncertainty on prototypes,interpretable uncertain shadowed prototypes are constructed and an adaptive uncertainty representation for complex data is formed.The uncertainty classification method based on the uncertain shadowed prototypes is implemented and extended to classify partial labeled data,which can effectively deal with the uncertain samples in the classification.Experimental results show that the uncertainty classification method based on uncertain shadowed prototypes can effectively improve classification efficiency and reduce decision-making risk.(2)Proposed the classification method based on feature encoding with uncertain shadowed prototypes.In order to further improve the feature representation and distinguishability of data,uncertain shadowed prototypes are extended on different prototype generation method,and the feature encoding method based on uncertain shadowed prototypes is proposed,which is helpful to describe the relative local relationship between samples and prototypes.The data features are expanded and represented by fusing the initial features and encoding features based on uncertain shadowed prototype,which improves the distinguishability of the data,reduces the dependence of the model on the data,and improves the classification result of the model.Experiments show the effectiveness of encoding features on uncertain shadowed prototypes,and verify the promotion in classification of feature encoding methods based on uncertain shadow prototypes.(3)Verified the effectiveness of the proposed classification methods with uncertain shadowed prototypes in the medical image classification tasks.Due to the problems of insufficient amount,limited labeling information,high noise and complex distribution of medical image data,the existing medical image diagnosis models have insufficient representation on medical data and high decision-making risk.In this paper,the uncertainty classification method based on the uncertain shadowed prototypes is used to classify bladder tumors,which can achieve efficient classification under the training of a small amount of labeled data.By identifying confusing bladder tumor samples as uncertain samples and making delayed decision,an efficient and lowrisk decision support system is realized.The classification method based on uncertain shadowed prototypes feature encoding is applied to pneumonia image.Feature encoding with shadowed prototypes help to build good feature representation between images,and the trained model can still perform effectively when the samples are insufficient.Classification diagnosis can reduce the dependence on data more than the neural network method.