Method for analyzing and training linear classifier based on torque

Abstract

The invention discloses a method for training linear classifier based on torque analysis, which uses an initialization method to generate a hyperplane in data space, processes iterate training on the initialized linear classifier based on torque analysis, stops the training according to iterate ending conditions to obtain one linear classifier with local optimization. The invention uses torque analysis to build linear classifier with wide application in pattern recognition, which leads physical force and torque concepts into machine learning, to train one linear classifier with local optimization. The classifier trained by the inventive method can reach local minimum correct reject rate. The invention has wide application of computer vision fields as face detection and vehicle detection, with significant theory and practical values.

Description

technical field [0001] The invention belongs to the field of pattern recognition and relates to a classification method for linear classifier learning. Background technique [0002] In the past decade, driven by the fields of artificial intelligence, robotics, and human-computer interaction, machine learning methods have attracted the attention of many research institutions. The machine learning method is to obtain the distribution law of the samples through learning on the given training sample set, and the learned rules can be used to classify unknown samples or to extract the characteristics of samples. Since machine learning algorithms do not depend on prior knowledge of relevant application domains, they have become the main research direction of pattern classification. [0003] Machine learning methods can be divided into unsupervised learning methods and supervised learning methods. Among them, the former refers to the learning algorithm without giving the category ...

AI Technical Summary

Problems solved by technology

However, since the optimization objective of the maximum rejection classifier only considers the second-order moments, the algorithm still has a false alarm rate in practical applications, and its performance is not satisfactory.

Images