基于高斯混合模型的遥感信息提取方法研究

摘要

高斯混合模型(Gaussian mixture model,GMM)可以描述遥感数据的概率密度函数,通过估计各高斯分布的参数,计算后验概率,实现信息提取.为了提高利用GMM进行遥感信息提取的准确度,首先在GMM中使用马尔科夫随机场(Markov random field,MRF)计算各像元邻域内各类地物的先验概率,代替各类地物的混合概率,使其反映出各类地物的空间相关性；然后在参数估计过程中利用模拟退火(simulated annealing,SA)思想获得全局最优的参数估计值；最后利用该参数估计值求出每个像元对于各类地物的后验概率,获得各类地物的空间分布.通过对遥感实验场的图像数据进行信息提取,发现所述新方法取得了更好的效果,证明了上述改进的有效性.%Gaussian mixture model ( GMM) is used to describe the probability density function of remote sensing data. According to the process of parameter estimation and the calculation of posterior probability, remote sensing information extraction can be realized. For the purpose of improving the accuracy of the extraction by GMM, Markov Random Field ( MRF) is applied to calculate the prior probability of each feature in the pixel' s neighborhood to replace the mixing probability of the feature, and spatial correlation is reflected by this way. Then simulated annealing ( SA) is utilized for the acquisition of overall optimum estimation of parameters. With the parameters, posteriori probability for every feature of each pixel is computed and the distribution of features is obtained. Extracting information from the images obtained from the remote sensing test site reveals that the new method has a better performance, thus proving the effectiveness of the above - mentioned improvements.