Method and device for obtaining projection transformation matrix, sample classification method and device

Abstract

The invention provides a method and device for obtaining a projection transformation matrix and a sample classification method and device. The method includes the steps of building a classification matrix through classification tags of training samples, and adding the classification tags into the computed projection transformation matrix according to the classification matrix. According to the method, due to the fact that classifications of the training samples can be accurately expressed through the classification tags, the classification matrix can accurately reflect the classifications of the training samples; on the basis that an Euclidean distance sum value is minimum, classification information is added, supervised learning is achieved, the correct projection transformation matrix can be accordingly obtained, the aims of shortening the distances between the training samples in the same classifications and prolonging the distances between the training samples in the different classifications are well achieved, and then the classification performance gets better.

Description

technical field [0001] The present invention relates to the field of artificial intelligence, in particular to a method and device for obtaining a projection transformation matrix, and a sample classification method and device. Background technique [0002] Today, with the gradual development of artificial intelligence, it is often necessary for the computer to classify the samples to be classified, that is, to determine the category of the samples to be classified in the training samples and categories, for example: based on multiple known faces and names corresponding to known faces one by one, Determine the name of the face to be classified. [0003] The samples to be classified and the training samples are the data of real-world objects, which are generally expressed in the form of a matrix. Due to the high dimension of real-world objects, the disaster of dimensionality often occurs when classifying high-dimensional data, so the data is generally mapped to In order to r...

AI Technical Summary

Problems solved by technology

[0007] However, using Euclidean distance to judge homogeneous and heterogeneous methods is prone to classification errors, that is, the Euclidean distance between sample C and sample A may be smaller than the Euclidean distance between sample B and sample A, resulting in The classification of intra-class collections and inter-class collections is wrong, which further leads to the failure of the projection transformation matrix, shortening the distance between different classes and expanding the distance between similar classes.

Therefore, the projection transformation matrix obtained by this method is incorrect, and it cannot achieve the purpose of shortening the distance between the same class and expanding the distance between different classes, which will lead to poor classification performance.

Images