Landsat-8 Image Classification using Support Vector Machine Classifier

Abstract

The extent of Built-up Area (BUA) is continuously increasing with rapid globalization. Identification of BUA provides vital information required for territorial planning as well as the impact of land cover changes on the environment. Therefore, detection of changes in land cover should be carried out periodically. However, it is difficult to extract built-up areas using satellite images because of the confusion between spectral values with other land cover types. Presently, satellite sensors provide continuous data in multiple spectral channels, which are becoming very useful for monitoring earth surface over large areas. The primary challenge is to accurately retrieve class information from the enormous set of data. The selected study area comprises of a scene taken from Haridwar District, India. The bounding coordinates of the chosen area are, long. 77 48’ 32.4’’ E and Lat. 29 54’ 50.4’’ N at upper left and long. 77 57’ 28.8’’ E and Lat. 29 45’ 46.8’’ N at lower right. In the last few decades, rapid urbanization has been taken place in this area, which results in increased infrastructural growth and urban expansion. The area mainly consists of land cover types such as built-up regions, agricultural land, water bodies, river sand and fallow land. The satellite data used in this study consists of multispectral bands acquired by Landsat-8 Operation Land Imager (OLI) sensor on 10 December 2014 with path- row number 146-39. The image represents a diverse land class scenario with 572 463 pixels in seven bands ranging from the wavelength of 0.43–2.29lm in the spectrum and having a spatial resolution of 30 m. In this study, medium resolution Landsat-8 data is used because it is suitable for mapping of land cover classes such as built-up area. However, the conventional methods are failed to provide the accurate classification performance. So, this work considered the machine learning based Support Vector Machine (SVM) classifier for obtaining the labelled samples from Landsat-8 Image.