Mode recognition method of mixed linear SVM (support vector machine) classifier with hierarchical structure

Abstract

The invention discloses a mode recognition method of a mixed linear SVM (support vector machine) classifier with a hierarchical structure. The iteration process of the mode recognition method mainly comprises the following steps of: firstly updating the training weight of each sample by applying the neighbor geometrical structural relation of training data and the classification error of the current H-MLSVMs classifier; and secondly training a linear SVM classifier with a weight by utilizing the samples with the weights, embedding the classifier into the current H-MLSVMs, and updating the classification error of the classifier. The mode recognition method disclosed by the invention is an effective mode recognition and classification method and is a universal method. Experimental results show that compared with other classical SVM methods, the classification method has the popularization capability which is always the same with a kernel function-based SVM method, has no shortcomings of difficulty in selection of a kernel function and parameters thereof, very large occupation of a memory and the like and particularly has the classification time complexity in the same order with the linear SVM; and furthermore, the classification method is more effective when the classification method is used for classifying new samples, can be further applied to a real-time mode recognition system, and has great application prospects.

Description

technical field [0001] The invention relates to the field of computer pattern recognition, in particular to a pattern recognition method of a mixed linear SVM classifier with hierarchical structure. Background technique [0002] Classifier design is the focus of research in the field of computer pattern recognition, because classifiers are the basic tools for pattern recognition research. In general, the current classifier algorithm mainly has two key issues: (1) the generalization ability of the classifier; (2) the time it takes to classify new samples. [0003] Generally speaking, the generalization ability of a classifier is the ability of the classifier to predict the category of unknown samples, that is, the accuracy of classification. The SVM classifier based on statistical learning theory and structural risk minimization theory has become the most successful classifier in recent years because of its global optimal solution and good promotion ability, and is widely us...

AI Technical Summary

Problems solved by technology

In general, the current classifier algorithm mainly has two key issues: (1) the generalization ability of the classifier; (2) the time it takes to classify new samples

This makes SVM unable to deal with classification problems with large amounts of data.

[0006] 3) The classification time complexity is high: when the SVM classifier based on the kernel function classifies a new sample, it needs to calculate the kernel function between the new sample and all support vectors

However, since this classifier does not introduce the concept of soft intervals, it cannot handle those two types of classification data that overlap each other, which greatly limits the scope of application.

Images