基于地理加权的k-|NN高分辨率遥感影像分类算法改进

与中低分辨率相比,高分辨率遥感影像的信息比较丰富,在使用常规k-NN分类方法基于像元进行高分辨率遥感影像分类时会产生大量的“椒盐噪声”和地物类别错分。根据地理学第一定律,引入地统计模型,将地理权重加入到常规k-NN分类方法中,形成新的地理权重k-NN分类器(Geographically Weighted k-NN,GWk-NN)。该方法首先通过条件概率函数计算出训练样本数据的空间分布特征,然后通过地统计模型对空间分布特征进行拟合,为每种地物选择合适的权重模型,这样既保留了遥感影像中地物的光谱特征,又融入了地物的空间特征,在一定程度上减少甚至消除了“椒盐噪声”,提高了分类精度。GWk\|NN和常规k\|NN分类器分析对比表明：GWk-NN分类方法提高了高分辨率影像的分类精度。

The Modified k-NN Classifier for High Spatial Resolution Remote Sensing Images based on Geographical Weighted Models

k-NN classifier has been widely used in remote sensing image classification due to its simple concept and easy implementation.However,it may produce a large amount of “salt and pepper” noise and wrong classification results in high resolution remote sensing classification because of its rich texture information.According to the first law of geography,this paper attempts to present a new geographically weighted k-NN classification method (GWk-NN) to solve these problems by incorporating geographical statistical models into the traditional k-NN classifier.First of all,the spatial distribution characteristics of the training samples of each land cover class have been calculated through conditional probability function;Secondly,two kinds of geographical statistical models (exponential and Gaussian model) are fitted and the suitable weighting model for each land cover is selected by the least residual error.Finally,a subregion of Nanjing city (SPOT5,2.5 m spatial resolution) is taken as an example to illustrate the validation of GWk-NN method.By comparing the classification results of GW k-NN and k-NN,the study finds that the new GWk-NN classifier can reduce or even eliminate the “salt and pepper noise” and eventually improve the classification accuracy by making use of spatial and spectral signature of the remote sensing images.