A Matrix Classification Model Based on Local Sensitive Discrimination

Abstract

The present invention provides a matrix classification model based on locally sensitive discrimination. Firstly, the data set is collected, and the collected samples are converted into a matrix pattern; secondly, the training set is used to construct local intra-class and local inter-class subgraph sums, and regularization is used to construct , and then introduce the regularization term into the matrix pattern classifier MatMHKS to generate a new matrix pattern classification model LSDMatMHKS, and use the training set to train LSDMatMHKS, and use the gradient descent method to find the optimal solution for the model LADMatMHKS; then use the test set to test the optimal solution, and obtain the optimal decision function; finally use the obtained optimal decision function to calculate the matrix pattern of the input unknown category, and classify the unknown matrix pattern according to the output result. Compared with the traditional classification technology, the present invention introduces local sensitive discriminant information so that local patterns of the same category are as close as possible and patterns of different categories are as far away as possible, thereby improving the stability of classification and the learning ability of the model.

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 local sensitivity discrimination. Background technique [0002] Different from the traditional vectorization-oriented classifier design method, the matrix pattern-oriented classifier design method is a method that can directly classify matrix samples. In actual experiments, the design method of matrix pattern classifier can effectively improve the performance of the design method of vectorization classifier. The reasons are mainly manifested in three aspects: first, the matrix pattern classifier design method can capture more structural information in a single sample, and the matrix method requires a relatively small storage space in the storage space of a pattern ; The second point, the method of designing a matrix pattern classifier avoids the operation of converting a single sample into a sample in the form of a vector to a...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the problem that the existing matrix pattern-oriented classifier design method does not take into account the local sensitive discriminant information between matrix patterns, the solution of the present invention is to design a new regularization term to consider local sensitive discriminant information, this regularization term is based on the advantages of local sensitive discriminant analysis and transforms it into a regularization term suitable for matrix mode, thus generating a local sensitive discriminant matrix learning model

Images