A classification method and system for missing multi-view data

Abstract

The invention discloses a classification method and system for missing multi-view data. The method includes: reconstructing the first latent space from the missing multi-view training sample data, reconstructing the second latent space from the missing multi-view sample data to be tested; reconstructing the loss function from the first latent space, missing multi-view training sample data and , train the multi-view multi-channel neural network model, input the first latent space and real class labels into the trained model, and adjust the first latent space with the total loss function as the objective function until the reconstruction loss function and the total loss function converge , get the trained model and the first complete latent space; input the second latent space into the trained model, adjust the second latent space with the reconstruction loss function as the objective function, and obtain the second complete latent space; from the first complete The latent space and the second complete latent space enable the classification of missing multi-view samples to be tested. The invention can improve the classification accuracy for missing multi-view data.

Description

technical field [0001] The invention relates to the technical field of image classification, in particular to a classification method and system for missing multi-view data. Background technique [0002] Multi-view data is very common in life, for example, magnetic resonance imaging and computed tomography in the medical field. These different types of data contain complementary information that is effective for disease diagnosis. However, it is difficult to integrate multiple types of data and make full use of them. In addition, the lack of partial view data makes modeling more difficult. [0003] Although the field of multi-view learning has developed rapidly in recent years, it is still limited by the effective modeling of complex relationships, and it is difficult for existing technologies to effectively solve the situation of missing views. In dealing with the problem of missing views, some technologies discard missing data and only keep complete data, which will lose ...

AI Technical Summary

Problems solved by technology

These different types of data contain complementary information that is effective for disease diagnosis. However, it is difficult to integrate multiple types of data and make full use of them. In addition, the lack of partial view data makes modeling more difficult.

[0003] Although the field of multi-view learning has developed rapidly in recent years, it is still limited by the effective modeling of complex relationships, and it is difficult for existing technologies to effectively solve the situation of missing views.

In dealing with the problem of missing views, some technologies discard missing data and only keep complete data, which will lose a lot of data information, especially when the sample size is scarce; some technologies are grouped according to the missing data, and each group is trained independently. It will not be able to fully mine the relationship between the data, and it will also lead to complex grouping when the missing situation is diverse

This leads to the fact that the existing classification methods for missing multi-view data cannot balance the consistency relationship and information complementarity between multi-view data, resulting in low classification accuracy.

Images