Please use this identifier to cite or link to this item: http://localhost:8081/xmlui/handle/123456789/6075
Title: A, STUDY OF VARIOUS FEATURE SELECTION TECHNIQUES IN DIGITAL REMOTE SENSING ANALYSIS
Authors: K., Deepak
Keywords: CIVIL ENGINEERING;SATELLITE DATA;FEATURE SELECTION TECHNIQUES;REMOTE SENSING ANALYSIS
Issue Date: 1994
Abstract: The purpose of applying feature seleotIon techniques in mulbIspectral data is to provide a trade-off between the cost and the accuracy of classification In order to reduce the dimensionality of satellite data, as well as computational time for the analysis. Feature selection undertakes the task of selecting a subset of hands from available number of bands of a " sensor. These bands are selected on the basis of either separability measure or degree of overlap between the classes present In the area. In the present study, Mina and its surrounding area lying In the Bihar State of India has been selected as the study area. Various feature selection techniques such as Divergence, Transformed Divergence, Dhatlacharya Distance, Jeffreys Matusita Distance and Brightness Value Overlapping Index FTVOI) have been used employing digital satellite data of LANDSAT-6 TM, 1992. The study has been carried out In tkm stages. In the first stage, feature selection techniques have been applied In order to. find out the best comblnatIon of two and three bands. In second stage, the entire Image has been classified into six classes Vi2. water bodies, vegetation, drY sand. wet sand, urban area. Imd boulders using the best combination of bands. In addition, the classification has been performed using the worst combination of bands. ACCUPaCy assessment of the classification has also been ULU carried using error matrix. ts show that classification acqiirmy: improves significantly when the best combinatiaiM Or speqhral bands Is used. The 131,01 technique is fentalitdhairs the best slies;only combination of best bands le VoibOROV.iOYWllid11 for the chasSification. If Inter class separability is tchthutindertaIned, then Transformed Divergence and Jeffreys NatutaikidiniOW.Nde are the' most 'useful techniques of feature selectionnill,, rri 1 lit IS suggested thatadhnhadsfasifInielsig Oen, [VAS, Of classes must be further examinedMISYCOIDI.1.301tablihilibthlinahribination of specific band on a Dationalnaihhal DiCsilard'hiridadij h 1,17. r19.1.1111,1 /1 410 LI," ' ;1 la 310h 11.1124,R 311113, ,1,10111111,1 11111,3.1 rid .,1111,, , ( in)
URI: http://hdl.handle.net/123456789/6075
Other Identifiers: M.Tech
Research Supervisor/ Guide: Ghosh, S. K.
Garg, P. K.
metadata.dc.type: M.Tech Dessertation
Appears in Collections:MASTERS' DISSERTATIONS (Civil Engg)

Files in This Item:
File Description SizeFormat 
246489CE.pdf1.56 MBAdobe PDFView/Open


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