关键词: 图像分类, 波利亚罐模型, 最大似然算法, 分类精度

Abstract:

Maximum likelihood (ML) algorithm is a widely used supervised method for remote sensing images classification.In ML algorithm,samples selections need to be extremely accurate,which also reduce the algorithmic efficiency.Therefore,a new method combined ML algorithm and Polya urn model to achieve panchromatic remote sensing images classification is presented.First,ML algorithm is used to calculate subordination probabilities of pixels.The numbers of balls of different colors are calculated by these probabilities,and the urn model of the image can be initialized.Urns' compositions are iteratively updated by the random sampling process of Polya urn model.The balls of the neighborhood are also combined to determine the next state of the urn.Finally,by steadying the quantity proportions of balls the final classification is achieved.The proposed method could classify images more precisely,and there is no request of samples selections;random selections of samples are practicable,which make the classification process simple.The results obtained on both synthesized and real remote sensing images show that the proposed method works well and is very promising.

Key words: image classification, Polya urn model, maximum likelihood algorithm, classification accuracy