Full text
| |
| Author(s): |
Enrico Campos Pedroso
Total Authors: 1
|
| Document type: | Master's Dissertation |
| Press: | Campinas, SP. |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Instituto de Geociências |
| Defense date: | 1998-04-06 |
| Examining board members: |
Alvaro Penteado Crósta;
Gilberto Amaral;
Fernando Pellon de Miranda
|
| Advisor: | Carlos Roberto de Souza Filho; Alvaro Penteado Crósta |
| Abstract | |
This work presents the results of semi-automated approaches for the geological mapping of a significant metallogenic province of Brazil, the Tapajós gold province. The geology of the Tapajós region comprises Archaean to Phanerozoic rock assemblages. The main gold accumulations occur as placer deposits associated with Quaternary alluvium sediments. The Tapajós Province is located in the Brazilian Arnazon, in an area covered by a dense tropical rain forest. As optical remote sensing data is severely constrained by almost permanent cloud coverage, we have selected JERS-I SAR data, together with a 1:250,000 scale geological map as the basis for this work. ln tropical regions such as the Amazon, radar imagery is an important source of textural information. The main goal of this research was therefore to evaluate image processing techniques that allow to recognise textural domains that could be correlated to the underlying geology. Geostatistic-based semivariogram textural classifiers and the grey-level co-occurrence matrices methods were employed for the semi-automated textural recognition task. Both are based on the spatial distribution of pixel values within an image. A set of textural channels is produced by either method that can be displayed as a pseudocolor image. The technique based on the semivariogram analysis yielded the best results. This technique first calculates the value of the semivariogram function for given training areas of different textural features. After a detailed interpretation of the curves, the best lags are selected interactively by the user, i. e. the lags that best distinguish the textural units. Next in the process, a bayesian unsupervised classification is performed in the entire image, using a moving window of predefined dimensions, based on the function values previously established. A comparison of the results derived from the textural classifiers, additional information extracted by digital image processing of Landsat TM data and ground truth data showed a good correlation in the spatial distribution of the main geologic units. Furthermore, it allowed distinguishing the alluvium that host gold-mineralized placers, based on their distinctive geomorphic texture. The semivariogram classifier is a powerful mapping technique that can be successfully applied for regional geologic mapping and mineral exploration in tropical regions and in particular to the vast geologically unknown terranes of the Amazon. The integration of SAR and optical data such as JERSI /SAR and Landsat TM imagery is a very useful procedure to enhance textural and spectral information provided by these sensors (AU) | |