Development of hybrid spectral matching approaches for improved mapping of selected vegetation and minerals

Abstract

In the arena of spectral sensing and research, matching algorithms newlinehave revolutionized the process of analysis and extraction of information from newlineremote sensing imagery. Several spectral matching approaches, ranging from newlinedeterministic to information theoretic-based processes, have been utilized for newlinetarget identification and mapping. However, each of these methods is limited newlineby an inherent factor which reduced their ability to utilize the band-level newlineinformation during matching. Combining two or more spectral matching newlineapproaches proved to be successful as they outperformed their individual newlinecomponents. This thesis presents a novel spectral matching technique which newlineintegrates the Jeffries-Matusita measure (JM) and the Spectral Correlation newlineMapper (SCM) algorithms for enhancing information extraction from newlinehyperspectral and multispectral imagery. newlineThe developed JM-SCM algorithms (tangent and sine version) are newlineimplemented in characterizing the selected vegetation (complex mangrove newlineecosystem) of Pichavaram and Muthupet regions, southern India using Earth newlineObservation (EO) -1 Hyperion imagery. The image-derived reference spectra newlinefor matching are that of: Avicennia, Rhizopora, paddy, groundnut, mudflat, newlinesand, clear and turbid water (in Pichavaram); and Avicennia, Prosopis, newlinemudflat, marsh, sandy/saline soil, clear and turbid water (in Muthupet). For newlinethe matching-based classification of Pichavaram imagery, the combined JMSCM newline(TAN) and JM-SAM (SIN) algorithm resulted in an increased accuracy newlineof 93.75 % and 91.25 % compared to 86.25%, 85 %, 76.25%, 75% and 71.25 newline% of JM-SAM(TAN), JM-SAM(SIN), JM, SCM, SAM respectively. For newlineMuthupet imagery, it yielded 91.25% and 88.75 %, compared to 85%, newline