遥感极大似然分类法讲义(英文)学案.pptVIP

Maximum Likelihood Classifier for Remotely Sensed Data by Chen xi the Effect of Classification development of tourism resources engineering exploration transportation planning and management natural disasters survey etc definition 1.calculating the attribution probability of each pixel for a category 2.assigning the pixel into the category of the largest attribution probability Process g1(x) Maxg(x) g2(x) gn(x) selector discrimination pixels decision Basic idea Every category of each dimension data on its own axis form a normal distribution,and its multidimensional data make up a multidimensional normal distribution of the category Algorithm the probability density function of a specific category i in a single band: where the is the average of category i, is the variance, n is the pixel number of the training sample When the number of band increases,for a m dimension subspace, the pixels(X), the mean vector of each category( ) and the covariance matrix ( ) can represented as followed: where: A probability density function in m dimension is then derived as followed: we used Bayes method to judge a point belongs to which category, if then we consider the point belongs to i category.However we dont know the prior probabilitys,generally we make them be equal. then above-mentioned function derived as followed: Data platform：ENVI Data：太行山2009/07/252013/06/02 Path124Row34（114:26E 38:03N）include 6 bands bare_land=[[92,45,58,76,97,48]; ,[94,47,59,71,95,51] ,[99,50,66,72,107,59] ,[98,49,64,79,99,52] ,[96,47,59,69,95,52] ,[95,47,59,86,102,52] ,[94,47,57,78,97,49] ,[95,46,58,76,101,53]] plant=[[85,39,37,92,64,27]; ,[82,39,36,94,71,28] ,[83,37,36,88,64,26] ,[84,37,35,95,61,21] ,[88,41,41,86,75,31] ,[86,38,37,89,71,29] ,[83,37,36,84,62,26] ,[88,40,39,94,71,28]] urban=[[108,48,53,56,81,47]; ,[111,53,60,66,89,55] ,[115,57,67,64,99,69] ,[128,64,73,68,97,63] ,[101,49,52,49,71,48] ,[116,59,68,64,96,62] ,[115,55,65,61,96,63] ,[113,55,64,66,89,58]] water=[[117,59,55,29,