No description available.

No description available.

No description available.

Thesis (MSc (Geography and Environmental Studies))--University of Stellenbosch, 2005. / 81 Leaves printed on single pages i-xi, preliminary pages and numbered pages 1- 70. Includes bibliography, list of tables and list of figures. / Digitized at 300 dpi color PDF format (OCR), using KODAK i 1220 PLUS scanner. / ENGLISH ABSTRACT: Geological field mapping is often limited by logistical and cost constraints as well as the scope and extent of observations possible using ground-based mapping. Remote sensing offers, among others, the advantages of an increased spectral range for observations and a regional perspective of areas under observation. This study aimed to determine the accuracy of a collection of image classification techniques when applied to ASTER reflectance data. Band rationing, the Crosta Technique, Constrained Energy Minimization, Spectral Correlation Mapping and the Maximum Likelihood Classifier were evaluated for their efficiency in detecting and discriminating between greenstone and granitoid material. The study area was the Archaean Barberton Greenstone Belt in the eastern Mpumalanga Province, South Africa. ASTER reflectance imagery was acquired and pre-processed. Training and reference data was extracted from the image through visual inspection and expert knowledge. The training data was used in conjunction with USGS mineral spectra to train the five classification algorithms using the ERDAS's software package. This resulted in abundance images for the target materials specified by the training data. The Maximum Likelihood Classifier produced a classified thematic map. The reference data was used to perform a rigorous classification accuracy assessment procedure. All abundance images were thresholded to varying levels, obtaining accuracy statistics at every level. In so doing, threshold levels could be defined for every abundance image in such a way that the reliability of the classification was optimized. For each abundance image, as well as for the output map of the Maximum Likelihood Classifier, user's- and producer's accuracies as well as kappa statistics were derived and used as comparative measures of efficiency between the five techniques. This information was also used to assess the spectral separability of the target materials. The Maximum Likelihood Classifier outperformed the other techniques significantly, achieving an overall classification accuracy of 81.1% and an overall kappa value of 0.748. Greenstone rocks were accurately discriminated from granitoid rocks with accuracies between 72.9% and 98.5%, while granitoid rocks showed very poor ability to be accurately distinguished from each other. The main recommendations from this study are that thermal infrared and gamma-ray data be considered, together with better vegetation masking and an investigation into object orientated techniques. / AFRIKAANSE OPSOMMING: Geologiese veldkartering word algemeen beperk deur logistiese en koste-verwante faktore, sowel as die beperkte bestek waartoe waarnemings met veld-gebasseerde tegnieke gemaak kan word. Afstandswaarneming bied, onder andere, 'n vergrote spekrale omvang vir waarnemings en 'n regionale perspektief van die area wat bestudeer word. Hierdie studie was gemik daarop om die akkuraatheid van 'n versameling beeld-klassifikasie tegnieke, toegepas op ASTER data, te bepaal. Bandverhoudings, die Crosta Tegniek, "Constrained Energy Minimization", Spektrale Korrellasie Kartering, en Maksimum Waarskynlikheid Klassifikasie is evalueer op grond van hul vermoë om groensteen en granitoied-rotse op te spoor en tussen hulle te onderskei. Die studiegebied was die Argalese Barberton Groensteengordel in die oostelike Mpumalanga Provinsie in Suid Afrika. 'n ASTER refleksie beeld is verkry, waarop voorverwerking uitgevoer is. Opleidings- en verwysingsdata is van die beeld verkry deur visuele inspeksie en vakkundige kennis. Die opleidingsdata is saam met VSGO mineraalspektra gebruik om die vyf klassifikasie algoritmes met behulp van die ERDAS sagteware pakket op te lei. Die resultaat was volopheidsbeelde vir die teikenmateriale gespesifiseer in die opleidingsdata. Die Maksimum Waarskynlikheid algoritme het 'n geklassifiseerde tematiese beeld gelewer. Met behulp van die verwysingsdata is 'n streng akkuraatheidstoetsing prosedure uitgevoer. Vir alle volopheidsbeelde is 'n reeks drempelwaardes gestel, en by elke drempelwaarde is akkuraatheidsstatistieke afgelei. Op hierdie manier kon 'n drempelwaarde vir elke volopheidsbeeld vasgestel word sodat die drempelwaarde die betroubaarheid van die klassifikasie optimeer. Vir elke volopheidsbeeld, asook vir die tematiese kaart verkry van die Maksimum Waarskynlikheid klassifikasie, is gebruikers- en produsent-akkuraathede en kappa statistieke bereken. Hierdie waardes is gebruik as vergelykende maatstawwe van akkuraatheid tussen die vyf tegnieke, asook van die spektrale skeibaarheid van die onderskeie teikenmateriale. Die Maksimum Waarskynlikheid klassifikasie het die beste resultate gelewer, met 'n algehele klassifikasie akkuraatheid van 81.1%, en 'n gemiddelde kappa waarde van 0.748. Groensteenrotse kon met hoë akkuraathede van tussen 72.9% en 98.5% van granitoiedrotse onderskei word, terwyl granitoiedrotse 'n swak vermoë getoon het om van mekaar onderskei te word. Die belangrikste aanbevelings vanuit hierdie studie is dat termiese uitstralingdata asook gamma-straal data geimplimenteer word. Beter verwydering van plantegroei en 'n studie na die lewensvatbaarheid van objekgeorienteerde metodes word ook aanbeveel.