Please use this identifier to cite or link to this item:
|Title:||SAR POLARIMETRY FOR CHARACTERIZATION AND RETRIEVAL OF EARTH AND LUNAR SURFACE|
|Publisher:||Dept. of Electronics and Communication Engineeing|
|Abstract:||This thesis “SAR Polarimetry for Characterization and Retrieval of Earth and Lunar Surface” explores the utilization of polarimetric Synthetic Aperture Radar (SAR) data for extracting constructive information in the form of polarimetric parameters to be used in various applications, like land cover identification, classification, and surface parameter retrieval for Earth and lunar surface. The main objective of the thesis is to pursue these tasks by using concepts of SAR polarimetry and electromagnetic wave modelling with emphasis on developing algorithms, which may require minimum or no ‘a priori’ information. Four tasks have been carried out in this thesis; (i) Study of model based decomposition methods and its analysis to visualize the effect of decomposition and deorientation for enhancement of land cover identification using polarimetric SAR data, (ii) Development of adaptive land cover classification algorithm using spatial statistics of polarimetric indices, (iii) Application of transmission line theory for development of algorithm for retrieval of soil moisture in bare soil and vegetation covered soil using minimum or no ‘a priori’ information, and (iv) Study and analysis of hybrid polarimetric MiniSAR data, and the development of algorithm for possible existence of water-ice deposits on lunar surface. Fully polarimetric ALOS PALSAR and/or Radarsat-2 data of Roorkee city in the state of Uttarakhand, India, have been used for characterization and soil moisture retrieval of Earth surface. The characterization of lunar surface in terms of identification of possible water-ice deposits has been performed by using hybrid polarimetric MiniSAR data of Peary and Rozhdestvenskiy craters. The thesis includes seven chapters. Chapter one presents the introduction, which consists of motivation, scope, and objectives of the thesis. In Chapter two, the state-of-the-art in the fields of advances made in SAR polarimetry, land cover classification methods, and soil moisture retrieval approaches has been briefly described. This chapter also elucidates the theoretical background for the presence of water-ice deposits on lunar surface. Critical reviews of presently available approaches for identifying regions having possible water-ice deposits on lunar surface have been presented, along with the vi discussion of limitations and challenges associated with them. In Chapter three, the study of incoherent model based decomposition methods with and without applying deorientation has been performed. The advantage of model based decomposition methods is their ability to extract polarimetric information from SAR data by expressing average scattering mechanism as the summation of individual scattering components, such as volume (Pv), surface (Ps), double-bounce (Pd), and/or helix component (Pc). In literature, there are several three- and four component model based decomposition methods. However, it is observed that due to similar polarimetric response, several land covers such as vegetation and oriented building blocks, decomposed into same volume scattering component by model based decomposition methods. In order to overcome this problem, it is suggested to apply deorientation i.e., rotation of target matrix (coherency or covariance) by radar line-of-sight, prior to decomposition. The deorientation effect results in getting same scattering response from differently oriented similar targets, and different scattering response from distinct targets, which might be producing same response before deorientation. Thus, this chapter analyses seven different three- and four-component model based decomposition methods, in which two methods are without deorientation, and other five are with deorientation. The methods without deorientation are, three component model based decomposition (TCM) proposed by Freeman and Durden in 1998 and four component model based decomposition (FCM) proposed by Yajima et al. in 2008. Model based decomposition methods with deorientation are, three component model based decomposition method with deorientation (TCMD) proposed by An et al. in 2010; three component model based decomposition method with double deorientation and adaptive volume scattering model (TCMDDA) proposed by Cui et al. in 2012; four component model based decomposition method with deorientation (FCMD) proposed by Yamaguchi et al. in 2011; four component model based decomposition with deorientation and additional volume scattering model (FCMDA) proposed by Sato et al. in 2012; and four component model based decomposition method with double deorientation (unitary transformation along with rotation) (FCMDD) proposed by Singh et al. in 2013. The results of these decomposition methods have been evaluated by performing visual and quantitative analyses for ALOS PALSAR data sets of Roorkee, Meerut, and Delhi cities of India. Two types of quantitative analysis have been performed; first, by analysing the variation in number of pixels for each scattering contribution; and second, by observing the scattering behaviour in terms of percentage of scattering power for different land covers. First quantitative vii analysis shows that in TCMD and TCMDDA, the pixel percentages of Ps and Pd increase as compared with TCM. In FCMD, FCMDA, and FCMDD, Pd increases drastically in terms of pixel percentage as compared with FCM. The pixel percentage of Ps is incremented by approximately 2% in FCMDA, and is invariant in FCMDD, as compared with FCMD. Pixel percentage having helix contribution (i.e., Pc) is increased by 2% in FCMD as compared with FCM, and is invariant in both FCMDA and FCMDDA. By second quantitative analysis, it has been observed that uncertainty always lies in the description of scattering mechanism of urban, tall vegetation, and short vegetation regions, because there is no distinct scattering mechanism which is dominant for these land covers in all decomposition methods. Only bare soil provides distinct pattern by having very strong contribution of surface scattering. After deorientation, double-bounce power is definitely enhanced, however, it is not the dominant scattering mechanism in urban area. This may occur due to the presence of large amount of vegetation within urban area (Roorkee city). In Chapter four, the problems associated with fixed-threshold based land cover classification algorithms and the need for the development of adaptive classification algorithm have been discussed. This chapter presents the development procedure for image statistics (median and standard deviation) based adaptive land cover classification algorithm by using best-selected polarimetric indices on the basis of separability index criterion. The algorithm provides optimum value of polarimetric indices on the basis of user-specific requirements (i.e., overall accuracy). The algorithm has been developed and validated on two different ALOS PALSAR data of same site i.e., Roorkee. For first ALOS PALSAR data, the overall accuracy is obtained as 87.59%, whereas producer accuracy is obtained as 98%, 71%, 86%, 92% and 95% for bare soil, water, tall vegetation, short vegetation, and urban, respectively, For second ALOS PALSAR data, the overall accuracy is obtained as 78.43%, whereas producer accuracy is obtained as 98%, 57%, 66%, 84%, and 97% for bare soil, water, tall vegetation, short vegetation, and urban, respectively. In Chapter five, the key issues related to the problems involved with the retrieval of soil moisture under vegetation cover by SAR data have been discussed. Considering the limitations of currently available soil moisture retrieval algorithms, this chapter presents multilayer model for retrieval of soil moisture in both bare soil and vegetation covered soil using the classical concept of transmission line theory. In this chapter, two different models have been developed for characterization of scattering from vegetation and bare soil regions. In case of vegetation, v iii three layer model having one layer of vegetation canopy and two layers of soil have been considered, whereas in case of bare soil, due to exclusion of vegetation layer the model consists of only two layers of soil. For both the models, calculated backscattering coefficients have been obtained as a function of complex dielectric constant and thickness of each layer involved in respective models. The observation depth for retrieval of soil moisture varies from one tenth of wavelength to one quarter wavelength. Therefore, the first layer of soil is considered to have thickness of 5 cm and second layer of soil is taken as infinite. In case of three layer model, the thickness of vegetation-air mixed layer is considered to have thickness in the range 5 cm to 400 cm, assuming all agricultural vegetation fall within this range. Now, the complex dielectric constant of each layer involved in respective models are retrieved though Genetic Algorithm (GA) approach by minimizing cost function. The cost function is formed by taking backscattering coefficient calculated by each model and HH polarized backscattering coefficient measured by SAR data. The developed algorithm has been applied on two data sets of ALOS PALSAR and one data set of Radarsat-2 of Roorkee city, and quite satisfactory results have been obtained by comparing the retrieved soil moisture values with observed soil moisture values. The advantages of the proposed approach are its capability to estimate soil moisture with good accuracy and requirement of minimum ‘a priori’ information. In Chapter six, a decision tree algorithm has been developed for finding the possibilities of water-ice deposits on lunar surface using MiniSAR data. In radar based missions, high value of received radar circular polarization ratio (μc >1) has been traditionally used as a key criterion for determining the evidences of possible water-ice deposits in cold dark permanently shadowed regions. However, rough and dry surfaces containing rocks, lava flows, ejecta etc., also represent μc >1 due to double-bounce effect. Differentiation on the basis of criterion μc > 1 is very challenging because of two different phenomenon associated with lunar surface, namely volume scattering caused due to presence of planetary water-ice, and surface roughness caused by ejecta, rocks, or lava flows. Therefore, in this chapter, the information of two different approaches has been fused which are polarimetric approach (m- and m-χ decomposition) and fractal approach (fractal dimension ‘D’). The polarimetric approach helps in obtaining scattering information of lunar surface, whereas fractal dimension ‘D’ helps in retrieving roughness information. After exhaustive study, various criteria have been obtained and incorporated in a decision tree. In this decision tree, the criteria for icy craters proposed by Thompson et al., have also included in order to provide confidence about regions having ix possibility of water-ice deposits. It has been observed that there are certain common regions inside anomalous craters on the floor of Peary and Rozhdestvenskiy craters which satisfy all the conditions of proposed approach. In this chapter, the study of electrical and physical properties like, dielectric constant of lunar surface (ε= ε’-j ε’’), loss tangent (tan δ), and regolith bulk density (ρ0), has also been performed. Finally, the work carried out in this thesis has been concluded in Chapter seven. This chapter presents the contributions of the thesis and the prospects of extending the tasks of thesis in future.|
|Appears in Collections:||MASTERS' DISSERTATIONS (E & C)|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.