Dynamic selection and circulating integration method for categorizer

Abstract

The invention provides a classifier dynamic selection and cycle integration method which is a novel classifier selection and integration method belonging to the computer science and technology field. Though the prior classifier selection and integration method improves the recognizing ability of a multi-classifier system to a certain degree, the problems such as low efficiency of optimal subset selection, lack of sufficient flexibility for the integration method still exist. In order to further improve the performance of the multi-classifier system, aiming at eliminating the defects of the prior method, the invention provides a novel classifier selection and integration method, that is, the classifier dynamic selection and cycle integration method. According to the new method, firstly classifier candidates are ordered according complementary index, secondly classifier subsets are dynamically selected according to target recognition status, and finally a recognition result is acquired through cycle integration method. The new invention enables selecting different numbers of classifiers for integration and recognition according to different targets to be recognized. Concerning a target which is easier to be recognized, only one or a small number of classifiers are enough, while concerning a target which is difficult to be recognized, a large number of classifiers have to be selected for carrying out a plurality of cycle integrations and obtaining a correct recognition result as much as possible. The invention which can greatly improve the efficiency, recognition rate and generalization of the multi-classifier system with flexibility, high efficiency and easy realization can be applicable to various aspects of pattern recognition such as character recognition, image recognition, biological feature recognition, medical diagnosis, military automatic target recognition and earthquake prediction.

Description

technical field [0001] The present invention proposes a new classifier selection and integration method—classifier dynamic selection and loop integration method, which can better improve the performance and generalization ability of multi-classifier systems, and can be applied to text recognition and image recognition , Biometric identification, medical diagnosis, military automatic target recognition and earthquake prediction involve all aspects of pattern recognition. Background technique [0002] The multi-classifier system is an important direction in pattern recognition. Great progress has been made in recent years. A large number of theoretical and experimental results have shown that the multi-classifier system can not only improve the accuracy of classification, but also improve the generalization of the pattern recognition system. capability and robustness. [0003] The key to the design of a multi-classifier system lies in the selection and integration of classifi...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the low efficiency of optimal subset selection in the current multi-classifier system design and the lack of sufficient flexibility of the integration method, the present invention proposes a new classifier selection and integration method—classifier dynamic selection and loop integration method, which consists of two main parts: one is candidate classifier ranking, the other is dynamic classifier subset selection and cyclic integration

Images