Some issues in contextual fuzzy c -means classification of remotely sensed data for land cover mapping

  • PDF / 1,625,109 Bytes
  • 10 Pages / 612 x 792 pts (letter) Page_size
  • 119 Downloads / 217 Views

DOWNLOAD

REPORT


J. Indian Soc. Remote Sens. (March 2010) 38 : 109 –118

RESEARCH ARTICLE

Some Issues in Contextual Fuzzy c-Means Classification of Remotely Sensed Data for Land Cover Mapping A. Dutta . A. Kumar . S. Sarkar

Received: 09.07.2009 / Accepted: 26.11.2009

Keywords Contextual information . Markov Random Field . Metropolis Algorithm . Gibbs Samples.

Abstract Earlier for the hard classification techniques contextual information was used to improve classification accuracy. While modelling the spatial contextual information for hard classifiers using Markov Random Field it has been found that Metropolis algorithm is easier to program and it performs better in comparison to the Gibbs sampler. In the present study it has been found that incase of

A. Dutta1( ) . A. Kumar1 . S. Sarkar2 1 Indian Institute of Remote Sensing (NRSC) Dehradun – 248001, India 2 Department of Geography, University of Calcutta, Kolkata, 700073, India.

e-mail: [email protected]

soft contextual classification Metropolis algorithm fails to sample from a random field efficiently and from the analysis it was found that Metropolis algorithm is not suitable for soft contextual classification due to the high dimensionality of the soft outputs.

Introduction Remote sensing data provides the unique opportunity to prepare land cover maps at the national level in a much faster and economic way, compare to the traditional ground surveying methods. Land cover map plays an important role in natural resources management and conservation. Conventionally in remote sensing thematic maps, one pixel is assigned to one thematic class, which is known as ‘hard’ or ‘crisp’ classification method (Foody, 1996), though the real scenario is some thing different; very often on the ground land cover class changes gradually from one to another. For example forest class gradually changes into the grass land area. A traditional hard

110

J. Indian Soc. Remote Sens. (March 2010) 38 : 109-118

classification technique does not take into account this continuous change in land cover classes and only assigns the single class level which dominates in a the pixel, it leads to loss of information (Kumar et al., 2007). Fuzzy or soft classification methods have received growing interest where land cover classes are still vague. In the field of remote sensing Fuzzy cMeans (FCM) clustering algorithm has been widely used to classify satellite images with vague land cover classes (Zhang and Foddy, 1998), which decompose the pixel into its class proportions. Earlier in contextual FCM classification of remotely sensed data, it was found that the contextual information could be useful to map the real world phenomena more accurately (Dutta et al., 2008). Perhaps there are few technical issues in contextual FCM classification and it needs to be discussed in a more detailed way. Therefore, the aim of this research work was to explore the important issues in contextual FCM classification using Markov Random Field (MRF) model. Since any commercial software packages does not include MRF bas