Please use this identifier to cite or link to this item: http://localhost:8081/xmlui/handle/123456789/1601
Full metadata record
DC FieldValueLanguage
dc.contributor.authorKumar, Anil-
dc.date.accessioned2014-09-24T06:40:53Z-
dc.date.available2014-09-24T06:40:53Z-
dc.date.issued2007-
dc.identifierPh.Den_US
dc.identifier.urihttp://hdl.handle.net/123456789/1601-
dc.guideDadhwal, V. K.-
dc.guideGhosh, S. K.-
dc.description.abstractRemotely sensed data is an ideal source of data for mapping land cover and land uses at a variety of spatial and temporal scales. The easiest and usual assumption is that each pixel represents a homogeneous area on the ground, however in real world, it is found to be heterogeneous in nature. For this reason it has been proposed that fuzziness should be accommodated in the classification procedure so that pixels may have multiple or partial class membership. To preserve the information present inside a pixel sub-pixel classification was introduced. Among the most popular techniques for sub-pixel classification are artificial neural networks, mixture modeling, supervised Fuzzy c-means and Possibilistic c-Means (PCM) classification. The objective of this research work is to investigate statistical learning based subpixel classifiers. A number of statistical and fuzzy based sub-pixel classifiers have been studied in this research work, like; Statistical based classifier Maximum Likelihood Classifier (MLC), Possibilistic Maximum Likelihood Classifier (PMLC), two fuzzy set theory based classifiers - Fuzzy c-Means (FCM), Possibilistic c-Means (PCM), one statistical learning based classifier - Density estimation based on Support Vector machine (SVM) approach. Fuzzy c-Means (FCM) algorithm has been taken as a datum algorithm to compare other algorithms performance. In this research work density estimation based on the support vector machines (SVM) approach, uses Mean Field (MF) theory for developing an easy and efficient learning procedure for the SVM. The MF methods provide efficient approximations, which are able to cope with the complexity of probabilistic data models. In this research work, a full fuzzy concept has also been tried, at sub-pixel level, using density estimation based on support vector machine (SVM) as well as fuzzy c-means (FCM) Investigation in Sub-pixel Classification Approaches for Land Use and Land Cover Mapping approaches. These approaches (SVM and FCM) were evaluated with respect to fuzzy weighted matrix. The sub-pixel classification algorithms in this research work have been tested for the effect of change of spatial resolution of multi-spectral data for same area. Also the sensitivity of these classifiers had been assessed using data of varied environment, like Plain areas and Undulating/hilly terrain. The tools required to achieve the objectives mentioned in this work were not present in commercial digital image processing software's. Thus for this research work, software package in JAVA environment known as SMIC: Sub-Pixel Multi-Spectral Image Classifier had been developed. This software package has file, signature, classifier, accuracy assessment and ALCM modules required in this research work. For this research work four different study areas have been identified to achieve all objectives. Two study areas (Shaspur and Asan area) lies in Dehradun district, Uttaranchal State, India, having center coordinates 30° 24' 20.75" N - 77° 48' 35.67" E and 30° 26' 08.00" N - 77° 40 05.36 E, respectively. For topographic effect studies two study sites, Shaspur area, moderately hilly, in Dehradun district and Haridwar (center coordinates 29° 50 30.89 N / 78° 06 47.66 E), plain area, in Haridwar district had been taken. For ALCM: Automatic Land Cover Mapping study, two different sites of India, like; Asan reservoir, Dehradun District (Center coordinates 30° 26" 08.00" N/ 77° 40* 05.36" E) and Rana Pratap Sagar Dam, Kota District, Rajasthan state (Center coordinates 24° 47' 30.18 N/ 75° 34' 14.77" E) had been taken. In this study, different experiments for sub-pixel classification approach had been carried out using data sets from AWIFS, LISS-III and LISS-IV sensors of IRS-P6 (Resourcesat-1) as well as ASTER satellite sensor data. Fraction images generated from different classifiers had been evaluated using fuzzy error matrix (FERM). n Abstract When evaluating statistical as well as fuzzy set theory based classification algorithms the effect of weighted matrix was observed on sub-pixel classification output. While considering Euclidean Norm, PCM approach gives maximum sub-pixel classification accuracy of 99.29%. In case of diagonal norm, Possibilistic MLC gives maximum sub-pixel classification accuracy of 94.38%. While considering Mahalonobis Norm, FCM approach gives maximum sub-pixel classification accuracy of 90.87%. In SVM approach for density estimation using mean field theory the effect of learning parameters (C, Eta and Epsilon) have been studied using four kernels in SVM and it has been observed that learning parameters ofSVM effects sub-pixel classification accuracy. While studying the effect of different single kernels using SVM based classifier, it was found that local Inverse Multiquadric kernel yields the best accuracy. In different mixture kernel functions in SVM the maximum mixtures of kernels had given maximum sub-pixel classification accuracy when \x in mixture of kernel used is of range 0.3. Some mixture of kernels performed well when (j, value used in mixture of kernel is of range 0.5. While investigating three sub-pixel classification algorithms, i.e., FCM, PCM with Mahalonobis Norm, as well as density estimation based on SVM approach, it was observed that SVM produced with highest accuracy of 93.44% with Inverse Multiquadric Kernel, on independent set of pixels in comparison ofFCM and PCM. When fuzzy concept was applied at training, allocation and testing stage using FCM and density estimation based on SVM approaches, it was observed that overall accuracy for both Fuzzy c-Means and density estimation based on SVM approach, with Euclidean norm gives the highest value i.e. 92.33% and 95.90% respectively compare to other weighted norms. in Investigation in Sub-pixel Classification Approaches for Land Use and Land Cover Mapping To study the effect of varying across resolution, area of two-land cover such as water pond and forest have extracted from AWIFS, LISS-III and LISS-IV data sets and relatively compared with mobile GIS/GPS data. It was found that as pixel size becomes fine, information extracted from remote sensing data is more close to its ground information. Topographic effect was studied on density estimation based on SVM approach using Inverse Multiquadric kernel at sub-pixel level classification and it had been observed that classification accuracy was not much affected due to topographical effect. While analysis the effect of dimensionality of the data on sub-pixel classification accuracy, it was observed that classification accuracy increases as data dimensionality increases in the case of SVM. But with fuzzy based classifiers classification accuracy decreases after certain stage. In this work PCM algorithm has been found to extract single land cover class water from multi-spectral remote sensing image. Thus, the findings of this research illustrate that statistical learning algorithm - density estimation based on SVM approach outperformed relatively when compared with fuzzy set theory as well as statistical based classifiers. IVen_US
dc.language.isoenen_US
dc.subjectCIVIL ENGINEERINGen_US
dc.subjectSUB-PIXEL CLASSIFICATIONen_US
dc.subjectREMOTE SENSING DATAen_US
dc.subjectLAND COVER MAPPINGen_US
dc.titleINVESTIGATION IN SUB-PIXEL CLASSIFICATION APPROACHES FOR LAND USE AND LAND COVER MAPPINGen_US
dc.typeDoctoral Thesisen_US
dc.accession.numberG13274en_US
Appears in Collections:DOCTORAL THESES (Civil Engg)



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