Applications of Texture Analysis for Rock Types Discrimination
Abstract
Aimed at developing image processing methods for rock types analysis with LANDSAT data, numerous experiments were conducted using supervised and unsupervised classification techniques under the general concept of texture analysis with LANDSAT digital covering two geological quads of Nevada. The results indicate that the supervised classification method is very effective in the extraction of granite regions when: (1) data were in ratio format; (2) feature variables included both tone and texture information; and (3) the classifier is capable of handling non-normally distributed data. Classification errors occurred when there exists pixels of non-granite category whose spectral and textural properties are statistically similar to that of granite pixels. Two cases of errors can be noted: Type 1 pixels located at the periphery of the granite regions, and Type 2 pixels located far away from the core of the granite areas. The final decision regarding the delineation of the granite regions is based on the intersection of two classification maps using a simple map overlay analysis. The result yields a correct classification rate of about 95 percent based on a visual comparision between the composite classification map and the ground truth information given in the U.S.G.S. geological map of the study area.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 1982
- Accession Number
- ADA117076
Entities
People
- Shin-yi Hsu