Matrix classification model based on inter-class discrimination

Abstract

The invention provides a matrix classification model based on inter-class discrimination, comprising the following steps: collecting a data set, and converting collected samples into matrix samples; constructing a regularization term R<BC>; introducing the regularization term R<BC> to MatMHKS, generating a new matrix-pattern-oriented classification model CBCMatMHKS, training the model with a training set, and using a gradient descent method to solve the model CBCMatMHKS in order to get the optimal solution to the model; testing the optimal solution with a test set to get an optimal decision function; and finally, using the optimal decision function to calculate an input matrix sample of which the class needs to be judged, and classifying the matrix sample according to an output result. Compared with the traditional matrix classification model, the distance between local samples of different classes is maximized by introducing inter-class discrimination information and using the cluster center to represent the samples in a region, and the accuracy of classification is improved.

Description

technical field [0001] The invention relates to the field of pattern recognition, in particular to a method of matrix learning machine model based on class discrimination. Background technique [0002] At present, most classifiers can only process vector-type samples, and matrix-type samples need to be converted into vector-type samples before they can be processed. For example, for a face picture, a vector-type classifier needs to convert it into a vector-type sample before processing it, but this loses the structural discrimination information inside a single sample to a certain extent. The design method of matrix pattern classifier can directly classify matrix samples. At the same time, experiments show that the design method of matrix pattern classifier can effectively improve the performance of vectorization classifier design method to a certain extent. [0003] The original matrix pattern classifier design method ignores the discrimination information between categor...

AI Technical Summary

Problems solved by technology

[0005] For the problem that the existing matrix pattern-oriented classifier design method does not take into account the discriminative information between the matrix patterns, the solution of the present invention is to design a new regularization term to consider discriminative information between classes, resulting in a locally sensitive discriminant matrix learning model

Images