A Comparative Analysis of Supervised Learning Techniques for Pixel Classification in Remote Sensing Images

Abstract

Predicting the class labels for each pixel in a remote sensing image is a very challenging task. Due to the high spatial resolution of the remote sensing data, each pixel in a remote sensing image has a meaningful information. Therefore, identifying the homogeneous regions and annotating them with significant land cover information remains an open challenge. To handle this challenge supervised machine learning methods are adopted and they play a key role in dealing with these high dimensional data and understanding the landcover information of the geographical surfaces in a remote sensing image. The main aim of this study is to analyse the performance of different supervised learning algorithms for labelling each pixel for the images obtained from International Society for Photogrammetry and Remote Sensing (ISPRS) Vaihingen. From the comparative analysis it is concluded that the fine Gaussian support vector machine outperforms the other state of the art techniques with an overall classification Accuracy of about 75.1448%.