Mathematics – Logic
Scientific paper
Dec 2008
adsabs.harvard.edu/cgi-bin/nph-data_query?bibcode=2008agufm.p53c1468i&link_type=abstract
American Geophysical Union, Fall Meeting 2008, abstract #P53C-1468
Mathematics
Logic
0540 Image Processing, 3672 Planetary Mineralogy And Petrology (5410), 5194 Instruments And Techniques, 5410 Composition (1060, 3672), 8094 Instruments And Techniques
Scientific paper
An autonomous Geologist's field assistant (GFA) is being developed to aid in the classification of remote geological data. This abstract focuses on experimental results from the unsupervised classification of unaltered igneous rocks. GFA uses a high-resolution digital camera to acquire close-up images of rock samples. All samples are classified by geologists and a database is created with ground truth data. Autonomous classification of an unknown rock is performed using an image retrieval method that uses the database (of more than 400 samples to date) to determine properties such as main texture (aphanitic, porphyritic, phaneritic) and general composition (felsic, intermediate, mafic). GFA's performance is evaluated using a "leave-one-out" procedure that provides a large data set to train the system against itself. While earlier experiments with low-resolution images required less computing power at the expense of losing important textural data, the latest experiments use high- to full-resolution images with comparable results. Training experiments show that, using black-and-white Gabor texture-based image retrieval, the highest classification rates occur for phaneritic samples (averaging at 80% correct with some instances achieving up to 91%), with considerably lower rates for aphanitic and porphyritic samples. At the same time, data indicate that many aphanitic samples are mistakenly identified as porphyritic. In response, several approaches that offer complementary information to fill the performance gap have been considered. For example, mean-shift color clustering seems to improve porphyritic classification rates by an average of 14%, which may prove crucial to improving overall classification; the aforementioned imaging procedures are being combined with Raman spectral data, which is expected to greatly improve results since accurate classification often hinges on detecting the presence of certain minerals; various other methods have been proposed including color Gabor texture image retrieval methods, and segmentation and thresholding to increase the sensitivity to grains and porphyries.
Gulick Virginia Claire
Ishikawa S. T.
No associations
LandOfFree
Geologist's Field Assistant for Remote Science Exploration: Using High-Resolution Images does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.
If you have personal experience with Geologist's Field Assistant for Remote Science Exploration: Using High-Resolution Images, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Geologist's Field Assistant for Remote Science Exploration: Using High-Resolution Images will most certainly appreciate the feedback.
Profile ID: LFWR-SCP-O-1239464