The use of contextual techniques and textural analysis of satellite imagery in geological studies of arid regions

Abstract

This Thesis examines the problem of extracting spatial information (context and texture) of use to the geologist, from satellite imagery. Part of the Arabian Shield was chosen to be the study area. Two new contextual techniques; (a) Ripping Membrane and (b) Rolling Ball were developed and examined in this study. Both new contextual based techniques proved to be excellent tools for visual detection and analysis of lineaments, and were clearly better than the 'traditional' spatial filtration technique. This study revealed structural lineaments, mostly mapped for the first time, which are clearly related to regional tectonic history of the area. Contextual techniques were used to perform image segmentation. Two different image segmentation methods were developed and examined in this study. These methods were the automatic watershed segmentation and ripping membrane/Laserscan system method (as this method was being used for the first time). The second method produced high accuracy results for four selected test sites. A new automatic lineament extraction method using the above contextual techniques was developed. The aim of the method was to produce an automatic lineament map and the azimuth direction of these lineaments in each rock type, as defined by the segmented regions. 75-85% of the visually traced lineaments were extracted by the automatic method. The automatic method appears to give a dominant trend slightly different (10° — 15°) from the visually determined trend. It was demonstrated that not all the different types of rock could be discriminated using the spectral image enhancement techniques (band ratio, principal components and decorrelation stretch). Therefore, the spatial grey level dependency matrix (SGLDM) was used to produce a texture feature image, which would enable distinctions to be made and overcome the limitations of spectral enhancement techniques. The SGLDM did not produce any useful texture features which can discriminate between every rock type in the selected test sites. It did, however, show some acceptable texture discrimination between some rock types. The remote sensing data examined in this thesis were the Landsat (multispectral scanner, Thematic Mapper), SPOT, and Shuttle Imaging Radar (SIR-B).