Image classifying method and system for self-adaption weight fusion of multiple classifiers

Abstract

The invention relates to an image classifying method and system for self-adaption weight fusion of multiple classifiers. Multiple features capable of giving consideration to different kinds of objects are selected, and comprise spectral feature, textural feature, fractal dimension feature and topic feature; for different features, different classifiers are selected and then fused, compared with the single classifier, the fusion classifier has higher classifying precision and better classifying effect. Meanwhile, weights of component classifiers are self-adaptively designed according to classifying reliabilities of the component classifiers to the samples without being fused and decided by one weight which is fixed for all samples.

Description

technical field [0001] The invention belongs to the field of remote sensing image applications, in particular to a remote sensing image classification method and system for multi-classifier self-adaptive weight fusion. Background technique [0002] Remote sensing image classification is an important aspect of remote sensing interpretation research, and many remote sensing applications are directly related to image classification. Traditional remote sensing image classification techniques are mostly pixel-level methods based on probability and statistics models. Statistical classification methods are generally divided into two types: supervised classification methods and unsupervised classification methods. Supervised classification methods require certain prior knowledge, such as the probability distribution of samples and high-quality training samples. Because remote sensing images are usually complex, the data sample features have many dimensions, and data samples with mu...

AI Technical Summary

Problems solved by technology

Doing so brings at least three problems: all features are arranged into feature vectors, which will bring about the problem of "dimension disaster" and increase the complexity of calculation; for different types of features, their representation forms, physical meanings, and dimensions are different. Similarly, scale normalization is required to arrange them into feature vectors, and effective normalization of different features is very difficult; different features are simply combined together, and it is difficult to choose a suitable classifier

Its disadvantage is that when the number of training samples is small or the quality of training samples is not high, the weight of the classifier may not correctly reflect the role of the classifier in the fusion system

For some improved weighting algorithms based on the recognition performance matrix, while using the recognition performance matrix to evaluate the classification performance of the classifier, the characteristics of the training samples are also considered to make the weight of the classifier more accurate. The disadvantage is that there is still no comprehensive Considering the overall distribution of samples and the characteristic differences of each sample, it is not possible to assign an appropriate weight value to the classifier for the sample based on the recognition performance of each component classifier for a certain sample and the degree of discrimination reliability.

Images