USE OF MULTIPLE SATELLITE IMAGES IN MULTIPLE SCALES FOR FEATURE EXTRACTION AND IMAGE CLASSIFICATION: A CASE STUDY OF RAMSAR WETLAND IN NORTH EAST INDIA

摘要

Several techniques are being used for extracting information from satellite images from decades. Extracting features based on segmented objects of the image is relatively a newer and efficient approach. However, for heterogeneous areas with high variations of Land Use-Land Cover (LULC) it is difficult to obtain accurate classified map from moderate resolution image like Landsat 4-7. Deepor Beel, a Ramsar wetland and its catchment in North East India is an example which consist of various topographical and LULC variations. In this study, a rule based classification algorithm was developed using spectral and spatial information of Landsat using ASTER DEM and band ratios such as NDVI (Normalized Difference Vegetation Index), NDWI (Normalized Difference Water Index), band2/band5, band5/band7 etc. as the ancillary information. The classification was done in two different scales: extended catchment scale and wetland scale. For catchment scale, the hilly forests could be easily separated from plain forests due to the use of the DEM. The water indices helped to accurately differentiate between water logged areas and urban areas. The validation using field data proved an accuracy of 94% and kappa coefficient of 0.91in case of the extended catchment scale classification. Using the classified datasets, the cover types dynamic degree change has been evaluated from 1989 to 2011.