http://www.iisc.ernet.in/
Advanced Machine Learning Algorithms based Free and Open Source Packages for Landsat ETM+ Data Classification
 http://wgbis.ces.iisc.ernet.in/energy/
Uttam Kumar1,2,3          Anindita Dasgupta1          Chiranjit Mukhopadhyay2           T.V. Ramachandra1,3,4,*
1Energy and Wetlands Research Group, Centre for Ecological Sciences [CES], 2Department of Management Studies, 3Centre for Sustainable Technologies (astra),
4Centre for infrastructure, Sustainable Transportation and Urban Planning [CiSTUP], Indian Institute of Science, Bangalore – 560012, India.
*Corresponding author: cestvr@ces.iisc.ernet.in

INTRODUCTION

With the development of remote sensing (RS) technology, space borne data have been widely used to classify land use (LU), permitting to update maps more frequently and nearly on a real-time basis [1]. Radiance/reflectance measurements obtained in various wavelength bands for each pixel provide spectral patterns that can be classified and correlated to different LU classes on the ground. Our ability to analyze RS data is important because it allows changes in the Earth's surface to be monitored as they occur [2]. By monitoring changes in the Earth's surface, a better understanding of the environmental control issues is possible. However, the unprecedented wealth of RS sensors and image-based geospatial information produce large volumes of data and result in large imagery-based data repositories [3]. Therefore, deriving LU information from these data is an important phase for the determination of land use/land cover information using machine learning algorithms.

The overall objective of classification is to assign all pixels in the image to particular classes or themes (e.g. water, forest, etc.). The resulting classified image represents a particular theme, and is essentially a thematic map of the original image [1]. In this context, machine learning algorithms tend to find hidden patterns, trends and relationships in data and classify them into user defined categories. Due to its broad applicability to many fields, machine learning has attracted tremendous attention from both researchers and practitioners.

In recent times, non-parametric methods, such as K-Nearest Neighbour, Neural Network, Random Forest, etc. have been recently practiced that have the advantage of not needing class density function estimation thereby obviating the training set size problem and the need to resolve multimodality [4-5]. In this paper, we perform a comparative analysis of six advanced machine learning algorithms viz. Decision Tree, K-Nearest Neighbour, Neural Network, Random Forest, Contextual Classification using sequential maximum a posteriori estimation (SMAP), and Support Vector Machine. These algorithms are evaluated and their performances are assessed on Landsat ETM+ data using Free and Open Source (FOS) Packages.

The paper is organised as follows. Section 2 describes the six machine learning algorithms along with a brief description of their FOS Packages, followed by data and study area in section 3. Section 4 presents the result and discussion and concluding remarks are given in section 5.

BACK   «   TOP   »   NEXT
Citation :Uttam Kumar, Anindita Dasgupta, Chiranjit Mukhopadhyay and Ramachandra. T.V., 2012, Advanced Machine Learning Algorithms based Free and Open Source Packages for Landsat ETM+ Data Classification., Proceedings of the OSGEO-India: FOSS4G 2012- First National Conference "OPEN SOURCE GEOSPATIAL RESOURCES TO SPEARHEAD DEVELOPMENT AND GROWTH” 25-27th October 2012, @ IIIT Hyderabad , pp. 1-7.
* Corresponding Author :
Dr. T.V. Ramachandra
Energy & Wetlands Research Group, Centre for Ecological Sciences, Indian Institute of Science, Bangalore – 560 012, India.
Tel : +91-80-2293 3099/2293 3503-extn 107,      Fax : 91-80-23601428 / 23600085 / 23600683 [CES-TVR]
E-mail : cestvr@ces.iisc.ernet.in, energy@ces.iisc.ernet.in,     Web : http://wgbis.ces.iisc.ernet.in/energy, http://ces.iisc.ernet.in/grass
E-mail    |    Sahyadri    |    ENVIS    |    GRASS    |    Energy    |      CES      |      CST      |    CiSTUP    |      IISc      |    E-mail