Object based image analysis of support vector machine and rule based image classification for building extraction/ Nazatul Asyikin Arham

• Main Author: Arham, Nazatul Asyikin
• Format: Thesis
• Language: English
• Published: 2020
• Subjects: Analysis; Image processing
• Online Access: https://ir.uitm.edu.my/id/eprint/34565/1/34565.pdf; https://ir.uitm.edu.my/id/eprint/34565/

Building extraction is one of the main procedures used in updating digital maps and geographic information system databases. This is a challenging task in a remote sensing community to extract buildings from high spatial remote sensing imagery because of the spectral similarity between man-made objects such as buildings, parking lots, roads, in the urban areas. This study utilizes Pleiades-1A satellite image data of Shah Alam areas to extract buildings in urban area. The main goal of this study is to demonstrate the capability of object-based image analysis (OBIA) in building extraction from high spatial remote sensing imagery. Different classification approaches, including support vector machine (SVM) and rule-based classification, were applied to the Pleiades-1 A. Results show that rule-based classification has a better overall accuracy closeness index with 0.07 while SVM had 0.14 of overall accuracy closeness index. The rule-based classification resulted in fewer buildings that under-segmentation and over-segmentation. The classification accuracy of the result obtained is approximately 95% for SVM and 83% for rule-based classification. The overall accuracy and kappa coefficient for SVM is 95.11% and 93% respectively and the classification accuracy using rule-based image classification shows 83.49%) and 76%) of overall accuracy and kappa coefficient respectively. The map of building extraction using SVM shows the distribution of building, tree, road, waterbody, land, grass and shadow area are 14%, 19%, 23%, 6%, 12%, 26%, and 0% respectively and the map of building extraction using rule-based image classification shows 26%), 24%o, 14%), 3%o, 30%), 3%) and 0% of building, grass, land, road, tree, waterbody and shadow area respectively.