Pattern Recognition Of The Saline-alkali Land Using Landsat/OLI-TIRS Data

Remote sensing based pattern recognition of saline-alkali land is a important approach to obtain the accurate surface saline-alkali land information timely by using remote sensing techniques and it also is a very complicated process. Improvement of extraction accuracy, and decrement of overestimation and underestimation, which have important implications for monitoring of saline-alkali land trough remotely sensed image based approach. Duo to the Landsat/OLI-TIRS data have large imaging area that is good for getting synchronous land surface information and it is also free of charge this data has been used for extracton of saline-alkali land distribution information in this thesis.This study chose the serious salinization area from Horqin left middle banner, Tongliao City, Inner Mongolia Autonomous Region as study area which is located in from latitude of 43° 58'N to 44° 08'N and longitude of 122 ° 26'E to 122°48'E, then on the basis of establishment of land cover classification system and collection of ground truth data, carried out extraction of distributed information for saline-alkali land cover type by using support vector machine algorithm and both of the traditional classification algorithms such as maximum likelihood classification method and ISODATA method based on Landsat OLI/TIRS images with path number 120 and row number 29 acquired in 23 May,2013. Above main data source of Landsat OLI/TIRS data include 23 variables, which are consist of nine original bands from band 1 to band 7 and band 9 to band 11, and six environmental index's bands derived from original Landsat OLI/TIRS bands which include NDVI (Normalized Difference Vegetation Index), NDWI (Normalized Difference Water Index), NDSI (Normalized Difference Soil Index), NDBI (Normalized Difference Building Index) and NDMI (Normalized Difference Moisture Index), and eight texture bands which include mean, variance, homogeneity, contrast, dissimilarity, entropy, second moment and correlation. On the basis of the accuracy assessments which consist of rondom samples based approach and visual interpretation based approach, this thesis finally got accuracies and comparison for different classification results of saline-alkali land information.After comparative analysis between three Saline-alkali land pattern recognition methods the results show that the support vector machine Saline-alkali land pattern recognition result better than the Saline-alkali land pattern recognition result from maximum likelihood method and ISODATA method. When using support vector machine algorithm, Saline-alkali land has the misclassification with residential area and grassland, and when using the maximum likelihood method and ISODATA method, Saline-alkali land not only has the misclassification with the grassland and residential, and also misclassified with the cropland at some extend. The support vector machine classification method with the best result for Saline-alkali land information extraction followed by the maximum likelihood classification method, ISODATA pattern recognition method with the significant misclassification result.