基于概率扩散的多光谱遥感图像分类模型

为了提高遥感图像分类精度,提出了一种基于概率扩散模型的多光谱遥感图像自动分类技术。该方法首先通过比较模糊C均值分类器（FCM）的有效性函数来自动确定最优分类数目,然后利用基于形态学的各向异性概率扩散模型来调整中心像元隶属类别的概率,最后根据概率扩散的隶属概率向量图,并按照最大后验概率估计（MAP）对像元进行分类。由于各向异性扩散具有保边缘平滑的特点,因此,该概率扩散模型不仅能够有效地抑制同质区域内部“斑点”的产生。而且使得图像上重要的边缘特征得到了较好地保留。实验结果表明,该分类算法不仅能够避免分类图像中“斑点”噪声的影响,而且分类后的总体精度达到了77．76％和Kappa系数达到了0．7198,均优于未经过概率扩散的最大后验概率估计分类算法,因而具有一定的实用价值。

Multispectral Remote Sensing Image Classification Model Based on Probabilistic Diffusion

In this paper,we propose an automatic multispectral remote sensing image classification technique based on improved probabilistic diffusion.Firstly,the optimal number of clusters in multispectral images is determined by comparing the validity functions of fuzzy c-means classifier(FCM).The posterior probability maps for each class are then smoothed by an improved version of multispectral anisotropic diffusion based on morphology.Finally,each pixel is classified independently using the maximum a posterior probability(MAP) estimate based on probabilistic membership maps.Because of the elegant property of anisotropic diffusion,edge-preserving smoothing,probabilistic diffusion,not only restrains effectively speckles in homogeneous regions,but also preserves preferably the significant physiognomy and edge features.Experimental results are given to show that the proposed method avoids the influence of "class noise" and its overall accuracy and Kappa coefficient have superiority capability over the traditional maximum a posterior probability estimate classification method without probabilistic diffusion.Thus it is an ideal remote sensing classification method.