Method and system for target recognition of online ELM based on consistent regularization

Abstract

The present invention discloses a method and system for target recognition of online extreme learning machine based on consistent regularization. The present invention includes the following steps of learning and training a classifier based on a single hidden layer feedforward neural network: Feature extraction, obtain the corresponding image feature set, and randomly divide the image feature set into multiple feature subsets; generate corresponding neighbor feature samples for the divided feature subsets respectively; introduce consistent regularization constraints into online over-limit learning The objective optimization function is to randomly generate the hidden layer node parameters of the single hidden layer feedforward neural network, select any feature subset and its adjacent samples to initialize the network weight generation, and iteratively update the network weights based on the remaining feature subsets to complete the network weight based on the remaining feature subsets. Learning and training of classifiers for single-hidden layer feedforward neural networks. The invention has the advantages of strong noise tolerance, high classification and recognition accuracy, fast learning speed and good task scalability.

Description

technical field [0001] The invention relates to the technical fields of image classification and target recognition, in particular to a method and system for target recognition of an on-line extreme learning machine based on consistent regularization. Background technique [0002] The extreme learning machine (ELM) is an efficient, generalized neural network learning algorithm originally based on a single hidden layer feedforward neural network. The theory of ELM is inspired by biological learning and attempts to explain and answer the basic question of whether biological neurons need to be adjusted during the learning process. Shortly after its proposal, well-known research institutions such as Harvard University, Stanford University, Massachusetts Institute of Technology, IBM Watson and other well-known research institutions have successively obtained direct or indirect verification in the olfactory system of mice, the visual system of monkeys and the perception system of ...

AI Technical Summary

Problems solved by technology

However, although the classifier trained by the extreme learning machine based on deep network features has significantly improved classification performance and efficiency, it will be affected by the noise existing in the input and its corresponding label, and lead to inevitable performance degradation

The reason for this problem often lies in the inherent shortcomings of traditional extreme learning machines, that is, traditional extreme learning machines are still based on empirical risk minimization, and overfitting is likely to occur under noise interference.

Therefore, the effect performance of target recognition technology based on traditional extreme learning machines is far from satisfactory, and often needs to be improved by introducing certain regularization constraints.

Images