Please use this identifier to cite or link to this item: http://localhost:8081/xmlui/handle/123456789/12476
Authors: Sharma, Ajay
Issue Date: 2011
Abstract: Advances in remote sensing technologies have provided practical means for land use and land cover mapping which is critically important for landscape ecological studies. Satellite images contain grids of pixels from surfaces that may be evaluated, based upon their level of reflected radiation. But due the complexity of earth's terrain, and presence of various phenomenon (e.g. scatterings, absorption of radiation, noises etc), and low resolution of imagery, the quality of images is not always very good, and so there is a great need of good classifiers which can come up with good pattern recognition of the terrains. In our work, we propose a classification method for satellite images which is based on texture features of the terrain. As the methods based on spectrum characteristics are very limited, texture based methods are becoming more and more popular because of their stable nature. In this proposed work, to compute various texture features, we have used gray level co-occurrence matrix (GLCM). GLCM gives a variety of, and multi-directional texture features. As most of the satellite imagery of low resolution, GLCM is a good option for computing texture features. We combine support vector machine with GLCM in the classification step. SVMs are becoming a very important tool in the field of classification because of their non-linear and non-probabilistic nature. As we don't have prior knowledge about pixel relationships in the terrain, SVM performs very well for it. In this method, we first compute feature vector with the help of GLCM in different directions, and then fed it along with a training vector to a SVMs decision tree for the final classification. We have applied this proposed method on synthetic aperture radar (SAR) images, which is a. popular satellite imagery technique. Results shows that this proposed method can classify earth's terrain in number of different classes with a good accuracy. This method is implemented in MATLAB, with some prerequisites implemented with Envi (v 4.7) tool.
Other Identifiers: M.Tech
Research Supervisor/ Guide: Niyogi, Rajdeep
Singh, Dharmendra
metadata.dc.type: M.Tech Dessertation
Appears in Collections:MASTERS' DISSERTATIONS (E & C)

Files in This Item:
File Description SizeFormat 
ECDG21025.pdf5.43 MBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.