Classification Of Remote Sensing Image Based On Fuzzy Sets

Fuzzy classification of remote sensing image was researched in this paper. Because pixel brightness are decided by mathematics and physics methods in images classified by computer, classifying an image clearly is difficult for those mixture pixels which have integrated spectral information because of spatial resolution limitation. The inherent fuzziness of remote sensing image causes traditional classification method can't get ideal classification results. While fuzzy sets theory can describe fuzziness of matters. In this paper, fuzzy sets theory was applied in classification of remote sensing image, classification of remote sensing image was researched.9 bands multispectral image was classified by C-mean algorithm and fuzzy C-mean algorithm firstly, and the classification results were compared, two advantages of FCM was find: (1) The clustering result of FCM is better than that of C-means algorithm; (2) FCM isn't sensitive to initial clustering center, while C-mean is sensitive to it. With different initial clustering center, classification results are greatly different too.Fuzzy maximum likelihood classification was proposed based on fully research of supervise and non-supervise classification methods. Selecting train sample on the basis of fuzzy C-mean clustering can improve accuracy of train sample, singleness of train samples can be satisfied. On the other hand, train samples distributed widely by selecting it after clustering and meet the integrity criterion at the same. The experiment results showed that classification results by using the proposed method is better than purely clustering. Selecting train sample on the basis of fuzzy C-mean clustering decreased subjective factor affecting selecting train sample, so higher classification accuracy can be achieved.