论文部分内容阅读
The thesis topic is land cover mapping using remote sensing (RS) and geography information system (GIS) techniques.ASTER data with 14 spectral bands including from Visible and Near infrared sensor systems (VNIR), Short Wave Infrared sensor systems (SWIR) and Thermal Infrared sensor systems (TIR) and LANDSAT 7 ETM+ with 6 spectral bands from Visible and Near infrared, Short Wave Infrared sensor systems are used to perform digital image classification in this research. Ground truth data was acquired for performing remote sensing images pre-processing and performing image classification for training samples selection and accuracy of the classification result.Image pre-preparation was conducted before performing image classification. Remote sensing images are geo-referenced to UTM zone 48, with ellipsoid datum WGS 84.Unsupervised classification was performed. Accuracy assessments results show that unsupervised classification needs further improvement to allow a more accurate land cover mapping. Then supervised classification was performed. In supervised classification, images with different sensor systems combination from ASTER data are classified and it is proved that different sensor system combination contribute to the land cover information extraction. GIS technique was used to improve supervised classification results by designing and running a model to combine the land cover classes with high accuracy from different classified images into a final land cover image. It is proved that GIS manipulation results in the improvement of RS image classification results. Comparing the unsupervised and supervised classification results, it is concluded that supervised classification method obtains higher overall accuracy and higher user’s and producer’s accuracy for each of the classes than those of unsupervised classification method. Overall accuracy obtained is less than 60% by unsupervised image analysis approach while 70% far by supervised classification method and 80% far after GIS manipulation of the supervised classification results, using independent ground truth for accuracy assessment.