Study On The Band Math Classification Method Based On The Multi-temporal TM Images

With the development of the global economy, the environmental issues arebecoming increasingly prominent. Desertification is one of the most seriousenvironmental problems that arise the international community more attention.While, due to the unique nature of the desert itself, the desert change monitoringhas a big limitation. With the continuous development of remote sensingtechnology and its unique advantage, it has been widely used in the desertificationdetection. In this paper, we use the Landsat TM images of Shanshan KumtagDesert to explore the feasible classification method, find out the scope of thedesert, and the time series change detection. The main contains are as follow:(1) In the data pre-processing procedure, the Geometric Correction, ImageSubset, and Minimum Noise Fraction are applied in the raw images.(2) Using Maximum Likelihood method, Support Vector Machine, IterativeSelf-organizing Data Analysis Algorithm (ISODATA), NDVI, MSAVI to classifythe Shanshan Kumtag Desert; and then qualitatively evaluate the experimentalresults.(3) By the analysis of TM remote sensing image,we found that: on theoriginal TM image, Shanshan Kumtag desert region on blue and green band (Band1) shows the strong absorption effect, and the thermal infrared band (Band6)shows a strong reflection effect; After MNF processed images, the desert regionhave significant absorption and reflection effect on the band of1and2; given thesignificant differences in features, proposing three band math index-Desert Index(DI), the Normalized Differential Desert Index (NDDI), and Modified Soil-adjustDesert Index (MSADI); and the1995’s original image and MNF processed imagesare taken as an example to test the three indexes to find the optimal index. Andthen, the optimal band math is used to classify the four time series (1995,2000,2005,2010) images;(4) The confusion matrix and statistical area are used to verify the accuracy ofthe Maximum Likelihood method results, Support Vector Machine results andMSADI results, thus proving the proposed method’s correction and effectiveness:Compared with these methods, we can concluded that: Using the MSADIband math for dynamic monitoring can accurately, real time, and quickly detectchanges, which has a great significance for sand.