基于MDL准则和EM算法的多光谱遥感影像变化检测  

Multispectral remotely sensed image change detection based on EM algorithm and MDL rule

作　　者：石爱业[1] 吴国宝[1] 谭德宝[2] 申邵洪[2] 黄凤辰[1] 马贞立[1] 王超[1] 

机构地区：[1]河海大学计算机与信息学院,江苏南京210098 [2]长江科学院空间信息技术应用研究所,湖北武汉430010

出　　处：《河海大学学报（自然科学版）》2013年第2期171-176,共6页Journal of Hohai University(Natural Sciences)

基　　金：国家自然科学基金(61271386;60901003)

摘　　要：针对由实际遥感地物类型难以确定导致的多光谱遥感影像变化检测精度较低的问题,提出一种结合最小描述长度(MDL)准则的EM变化检测方法。首先,采用主成分变换与相关系数融合法相结合的方式构造差异影像;其次,利用分支数为k的高斯分布混合模型对差异影像进行建模,并利用MDL-EM算法自适应估计模型各参数;最后,利用基于统计最小错误率的Bayes判别准则确定变化检测的阈值来实现多光谱遥感影像变化检测。实际遥感数据验证结果表明,所提检测方法应用于多光谱遥感影像变化检测中是可行、有效的。In order to overcome the disadvantages of multispectral remotely sensed image change detection due to the difficulty in determining the actual ground object type,a change detection algorithm is proposed through combination of the minimum description length（MDL） rule and the EM algorithm.First,the difference images were constructed by the principal component analysis（PCA） and the correlation coefficient fusion method.Then,a distribution model for the difference images was built with the Gaussian mixture model（GMM）,and its related parameters were estimated adaptively by the minimum description length-expectation maximization（MDL-EM） algorithm.Finally,the change detection was implemented by the threshold determined by the Bayes rule.Experimental results on the multispectral images show that the proposed algorithm is feasible and effective.

关 键 词：遥感影像处理 变化检测 EM算法 MDL准则 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]