The Estimation Of Class Probability Of Image Gray Value

With the development of remote sensing technology, we get more and moreinformation for the earth we live.especially in recent years, we have more andmore access to the methods of remote sensing images. However, for a growingnumber of remote sensing images, the challenges followed, how do we select anbetter remote sensing image to study become a field of remote sensing researchdirection in our study of remote sensing images. Despite mant remote sensingimage classifications secularing, but without a forecast of the quality of theclassification of numerous remote sensing image in advance, rushing to use theclassification method will waste a lot of manpower and material,the researchof remote sensing image may have a perceptibly opposite effect.Therfore, This paper tightly revolve how to estimate theimage quality ofimage-based classification, as the main object of study to simulate data, focusfrom the selecteding the beneficial image classification samples and reducinghuman involvement of these two aspects to be studied， on the basis of previouswork，proposing a classification method witch combined semi-supervised andunsupervised classification, in order to achieve the estimating of the varioustypes of membership of different gray value. The following is the detail:1）Both the labor participation and classification accuracy can not bebalanced in traditional classification algorithm,for this,we propose a methodbased on the K-means and semi-supervised classification.2）On the basis of the weighted support vector domain data description methodbased on semi-supervised learning，Use unsupervised clustering method andempirical knowledge to identify the most likely point labeled samples，toachieve without human intervention.3) Being improved on the basis of the weighted support vector domain data description method based on semi-supervised learnin, iterative and generatenew mark point to solve the unaccurately estimating problem of excessivenumber of categories.