Ransac-algorithm-based robust AdaBoost classifier construction method

Abstract

The invention proposes a Ransac-algorithm-based robust AdaBoost classifier construction method. The method comprises: according to a Ransac algorithm, constructing a sample subset; extracting a sample; on the basis of an AdaBoost algorithm, training a strong classifier; calculating a classification precision value corresponding to a classification model; constructing a classifier model Ci; determining all training samples meeting the newly constructed classifier model; repeating the above-mentioned steps until obtaining a model Ci and a corresponding consistency set under Ransac algorithm estimation; determining whether the number of different established classifier models exceeds Nmax; and determining a finally selected classifier model and calculating a corresponding classification precision value. According to the method provided by the invention, the optimal classifier model is established by samples containing exterior points and the influence on the classifier model constructionb the exterior points by using the iteration process of the Ransac algorithm.

Description

technical field [0001] The invention relates to the technical field of robust classifiers, in particular to a method for constructing a robust AdaBoost classifier based on the Ransac algorithm. Background technique [0002] The term AdaBoost is derived from the abbreviation of Adaptive Boosting (Adaptive Boosting). It is a machine learning meta-algorithm proposed by Yoav Freund and Robert Schapire. Its design guiding principle is to ensure that the current training samples have the highest classification accuracy. By reasonably combining different weak classifiers (the so-called weak classifier here means that the classification accuracy is slightly better than random guessing), a strong classifier is formed. Although the classification accuracy of each weak classifier is not high, the final strong classifier The classifier will get a huge boost in classification performance. The AdaBoost algorithm is self-adaptive in a sense, by adjusting the weights of the misclassified s...

AI Technical Summary

Problems solved by technology

However, like everything has two sides, although the AdaBoost algorithm has many advantages, it is sensitive to outliers, and in some cases it is more likely to be affected by this, resulting in the degradation and failure of the overall performance of the classifier.

This is because the samples that cannot be classified correctly are constantly weighted, especially the continuous weighting of the outliers, which makes the weight of the outliers grow too fast

Excessively large outlier weights will cause the classifier algorithm to continuously deviate from the outliers, and then deviate from most normal samples, which will inevitably lead to the degradation of the designed classifier model

Images