提出了一种基于变差函数模型拟合的SAR图像分类新方法。该方法首先采用球状模型拟合不同地物的变差函数图,计算模型参数——变程值;然后,用变程值所对应的变差函数纹理特征构成分类特征矢量;最后,利用K-近邻分类器实现训练与分类。对不同RadarSat图像的实验表明该方法可有效用于SAR图像的分类,且分类性能优于经典的最大似然法和灰度共生矩阵法。

A method of urban SAR image classification based on variogram modeling is proposed in this paper. The proposed method mainly consists of three steps as follows. Firstly, variograms of different classes are fitted by the spherical model to obtain ranges. Secondly, with the obtained ranges used as lags, corresponding variogram texture features are extracted. Finally, K-nearest neighbor (KNN) classifier is trained and used for classification. The experimental results of different RadarSat images show the feasibility of the proposed method. Quantitative performance evaluation shows that our method outperforms the two classic methods, maximum likelihood classification and gray level co-occurrence matrix texture classification.

国家自然科学基金资助项目