Transcript Click here to see the Powerpoint presentation.

Phoebe L. Hauff
Cari Deyell-Wurst
William Kerby
PRIZE
INTRODUCTION
“Spectral geology” applications (including
satellite, airborne, core scanning and field
measurements) have become very common
due to significant technological
advancements and improved
instrumentation.
This has also led to many programs that attempt to automate the sample
handling and mineral identification processes.
There are numerous libraries and algorithms available.
They are briefly described here.
We have also done a Round Robin to demonstrate effectiveness of some
of the algorithms.
Expanded summaries (www.spectral-international.com)
OVERVIEW
SPECTRAL LIBRARIES
• Spectral libraries contain reference spectra, which are
compared against an unknown spectrum using
computer automated ID techniques.
• Without them, it would be difficult to do interpretation
Minerals are highly variable –
composition, wavelength,
profile, crystallinity
Difficult to automate
CRYSTALLINITY
Data Bases
CHLORITES
contain spectra, ancillary information, physical
properties, references, associated species,
location.
KNOWN VIS-SWIR LIBRARIES
The libraries listed here are the better known ones. There are innumerable little
ones targeted at one mineral, one mineral group, vegetation. The addresses for
the common ones are included and will be on the SII website.
SpecMIN:
SPECMIN is a mineral identification system for spectroscopy that includes an extensive and dynamic
library of reference spectra for minerals, wavelength search/match tables, physical properties of each
species in the database, and literature references for the infrared active mineral phases. The spectral
library includes a minimum of two different samples per mineral that show compositional differences
within mineral species. In addition to mining applications, SPECMIN can also be used in remote
sensing applications for ground truthing.
www.spectral-international.com
CSIRO:
Inbuilt reference library of spectra of common minerals. As well as the mineral spectra, the library also
includes some artifact materials such as vegetation, plastic and marker pen which could also
potentially contribute to your project spectra.
www.csiro
USGS:
The concept and identification work basic to this library was started by Dr Graham Hunt in the 1970’s.
The library is used as a reference for materials identification in remote sensing images. It can also be
used for identification of laboratory and field spectrometer data. The software used to manage this
library is specPR. The library is available without cost as a download from the internet.
http://speclab.cr.usgs.gov/spectral.lib06
KNOWN VIS-SWIR LIBRARIES
Brown University RELAB
The public domain spectral library is supported by NASA at Brown University.
WWW.PLANETARY.BROWN.EDU/RELAB/. It contains thousands of spectra from NASA researchers and
others. Data cannot be used for commercial applications. It contains many project specific data sets
including planetary projects.
Arizona State University
MINESpectra is a data management program that interfaces with the USGS, JHU and special purpose
libraries. It is free. www.geologynet.com/minspectra.htm
Cal Tech VIS
This extensive library concentrates on the VIS range . There does not appear to be an identification
program. It does not say if this is digital or not.
Minerals.gps.caltech.edu\index.html
Johns Hopkins
FTIR , SWIR Incorporated by JPL into ASTER library. Available from JPL
KNOWN VIS-SWIR LIBRARIES
JPL
The spectral library available from the Jet Propulsion laboratory
contains 160 mineral spectra in 3 different grain sizes. The
references are well characterized. This library was incorporated
into the “ASTER” Library along with the Johns Hopkins midinfrared library growing to 2400 spectra of different infrared
materials. It is available from JPL.
Speclib.JPL.nasa.gov/documents/jpl_desc
MINEO (www2.brgm.fr/mineo/spectral.htm)
All spectra collected during field campaigns, lab analysis and
image analysis are gathered into a single spectral library. This
MINEO specific spectral library will constitute a European scale
spectral data base for mining related contaminated areas. It is
able to manage large amounts of spectra collected from
laboratory analysis, field spectrometry, as well as spectra
extracted from hyperspectral imagery.
THE MINEO PROJECT∗
Chevrel S., BRGM, Orléans – France,
Kuosmannen V., GTK, Espoo – Finland;
Belocky R., GBA, Wien – Austria; Marsh S.,
BGS, Nottingham, United Kingdom; Tukiainen
T., GEUS, Copenhagen – Denmark; Mollat H.,
BGR, Hanover – Germany; Quental L., IGM,
Lisbon – Portugal; Vosen P., DSK, Bottrop –
Germany, Schumacher V., JRC/SAI, Ispra –
Italy,Kuronen E., Mondo Minerals, Kajaani –
Finland, and Aastrup P., NERI, Copenhagen –
Denmark
ABSTRACT
MINEO is a European Research and Technological
Development project which aims at developing
tools and methods for assessing and monitoring
the environmental impact of mining activities by
means of combined Earth Observation and other
relevant environmental data set. MINEO is
designed to improve the already proven
hyperspectral imagery capabilities in mineral
mapping for use in the mapping of mining-related
contaminated areas in European vegetated
environments.
MINE WASTE LIBRARY
Generation of an European scale
spectral library of contaminated areas
APPLICATIONS
AMD
SPECIAL-PURPOSE LIBRARIES
vs. ONE MAIN LIBRARY
BY ALTERATION TYPE
BY SPECIFIC DEPOSIT TYPE
SITE- SPECIFIC
USGS
EXISITING DEPOSIT-SPECIFIC
LIBRARIES
GOLD EPITHERMAL
HSS, LSS
GOLD OROGENIC, VEIN
PORPHYRY
IRON MINERALS
IOCG
URANIUM - UNCONFORMITY
SKARNS
REE
SPECIAL-PURPOSE LIBRARIES:
ALTERATION TYPES
PROPYLLITIC - ZEOLITIC
SERICITIC-CHLORITE
ARGILLIC
INTERMEDIATE ARGILLIC
ADVANCED ARGILLIC
SILICIC
CARBONATE
QSP (QUARTZ-SERICITEPYRITE)
TOURMALINE
SKARNS
ADVANCED ARGILLIC
POTASSIC
PHYLLIC
IRON OXIDES
ADVANTAGES OF SPECIAL PURPOSE LIBRARIES
FEWER WRONG CHOICES = MORE ACCURATE MATCHES
Example:
Porphyry Deposit – Chile – Lithocap Zone
50 Random samples selected through mineralized zone
Samples run against:
SPECMIN Database – SII library 1536 reference samples
Deposit Library – 557 samples from porphyry
Environments + selected deposit references
TSG (TSA) Database
Mineral suite seen through zone, alunite, dickite, kaolinite, smectite, illite, muscovite,
chlorite, epidote, gypsum, silicification, goethite, hemitite, jarosite
Procedure:
1 - Samples run against large SII library
2 – Samples run against a location-specific porphyry library
Results:
SII vs. Porphyry library
Results between SII vs. porphyry database identified 14% of
first order minerals were misidentified with SII Main library
Minerals causing largest issue were smectite-illite-muscovite
presence.
Results:
Porphyry Library vs. TSG
22% of first order minerals misidentified with TSG
5% of first order minerals should have been identified as
second order presence
34% of second order minerals were identified as NULL in TSG
(mineral presence did exist in 90%)
37.5% of remaining second order minerals were misidentified
Only 23% of second order minerals correctly identified
Example: Porphyry Cu-Au Deposit
Mineral ID Using General Library
SpecMIN - FeatureSearch
Top matches not relevant
Example: Porphyry Cu-Au Deposit
Mineral ID Using Deposit-Specific Library
SpecMIN - FeatureSearch
Top matches correct!
Example: Porphyry Cu-Au Deposit
Mineral ID Using General Library
SpecMIN - FeatureSearch
Top matches not relevant
Example: Porphyry Cu-Au Deposit
Mineral ID Using Deposit-Specific Library
SpecMIN - FeatureSearch
Top matches correct!
SPECIAL PURPOSE LIBRARIES WILL
PROVIDE BETTER MATCHING STATISTICS
THEY ELIMINATE WRONG CHOICES
“AUTOMATED” MINERAL ID ALGORITHMS
The Spectral Geologist (CSIRO, Australia)
The Corescan Interpretation Software (CoreScan)
FeatureSearch (Steve Mackin, specMIN)
SpecPR (USGS) - NOT ID PROGRAM
The original hyperspectral ID program - NOT AN ID PROGRAM
(Neil Pendock, Phil Harris, Paul Linton, Anglo American-DeBeers)
Newmont – Dave Coulter - NO LONGER USED
Rio Tinto - Alistair LAMB - NO LONGER USED
MINEO – French Consortium - PROJECT ENDED
BHP - PROJECT STOPPED WHEN PIMA APPEARED
“AUTOMATED” MINERAL ID ALGORITHMS:
TYPES of PROGRAMS
Wavelength-based
Least squares
Profile based
Linear regression
Neural nets
Non-linear regression
“AUTOMATED” MINERAL ID ALGORITHMS:
TYPES of PROGRAMS
SHAPE BASED – Feature Position
simple lookup table
example: Feature Search; Tetracorder (USGS)
SHAPE MATCHING
Pearson correlation Matrix
Matched filtering
VECTOR SPACE ALGORITHMS
remote sensing classification method
TSG/TSA appears to use this type of algorithm
The Spectral Geologist (CSIRO, Australia)
Specialist processing
and analysis software
package designed for
analysis of field or
laboratory
spectrometer data.
It is automated
It uses a spectral
library developed by
CSIRO
The Corescan
Interpretation Software
(CoreScan)
Corescan is a global services
company specialising in the
hyperspectral scanning,
processing and analysis of drill
core, rock chips and other
geological samples for the
mining, oil and gas, and
geotechnical industries.
FeatureSearch (Steve Mackin, specMIN)
FeatureSearch is a semi-automatic mineral identification package for
determining mineralogy based on features observed in an "unknown"
spectrum collected by a field spectrometer.
Ideal for novice users with little experience in spectral identification or for
advanced users trying to determine low proportion end-members in mixtures.
The software is spectrometer-independent and operates with data from
specTERRA™, ASD, GER, SEI or PIMA spectrometers.
The users can select a spectral library created with any spectrometer.
Drag and drop a file or select a file from Plot Preview, click on the "Search
Library" button and the chosen mineral library is searched in less than a
second. The results are displayed clearly to allow the user to extract and
save the information of interest in History Libraries.
Use the extracted end-members to build a deposit or environment specific
library to use in an automatic mineral identification algorithm such as SIMIS
Field 2.9, or save the results to directly import into Microsoft Excel.
SPECMIN
SPECMIN incorporates options from the
FeatureSearch program, which allows it to
unmix spectral components, do mineral
percentages, and access user-created
custom libraries.
SPECMIN is a data management
system that puts spectral data into an
easy access format. It provides
numerous spectral libraries including
ASD, PIMA, USGS, and JPL mineral
libraries. It contains spectra from
hyperspectral imagery such as
AVIRIS and SFSI. SPECMIN also
contains libraries for soils, and
libraries specific to precious metals
deposit types.
Example: Mineral Spectral Analysis Software
Potential to
change mineral
list?
Spectra
Kaolinite
Illite
Dolomite Dickite
Phlogopite
Epidote
Calcite
Alunite
Pyrophyllite
Chlorite
sample 1
61.3
11.4
0.1
13
0.1
0.1
0.1
13.1
0.5
0.1
sample 2
39.7
59.3
0.1
0.1
0.1
0.1
0.4
0.1
0.1
0.1
sample 3
41.4
21.9
0.1
3.4
0.1
16.5
0.1
0.1
0.1
16.5
sample 4
3.9
1
17.7
0.1
30.3
3.4
0.1
0.1
0.1
43.2
sample 5
25.1
22.6
6
0.1
3.2
7.3
7.8
0.1
0.1
27.6
sample 6
0.2
0.2
98.2
0.2
0.2
0.2
0.2
0.2
0.2
0.2
sample 7
38.4
16.5
0.1
5.6
0.1
11.4
0.1
0.1
0.1
27.7
Potential to be effective in localized environments!
Total Sums
to 100%
WHY SO FEW???
IT IS HARD
CHALLENGES FOR AUTOMATED
MINERAL ID PROGRAMS
CHEMICAL VARIABILITY OF MINERALS
ABSORPTION CO-EFFICIENTS ARE UNKNOWN
NON-LINEAR ASSOCIATIONS
MIXTURES
USER INEXPERIENCE
Ex. Absorption Coefficients
In A high (illite) and low (chlorite) reflectance mixture,
the low reflector is difficult to see. In this example,
there has to be nearly 40% Fe-chlorite present before
it can be detected
Fe-Chlorite & illte
OBJECTIVES: ROUND ROBIN TEST STUDY
SURVEY OF AUTOMATED MINERAL ID PROGRAMS
•S:N.
•Quality of spectrum accepted
•What the algorithms are prejudiced towards
•Where they do not do well
•Poor libraries?
•Special purpose data libraries
•Artifacts of the spectrum
•Mixtures
•Experienced user required? Or “out of the box”?
ROUND ROBIN: SUMMARY OF PROCEDURE
• 48 spectra total (natural samples, computergenerated mineral mixtures)
• specTERRA, TerraSpec, FieldSpec Pro
• Anonymous participants
• Samples were run through automated mineral ID
programs of participants’ choice
• The entries were scored
• A winner determined based on:
• Greatest number of correct minerals identified
• Penalized for wrong answers
MINERALS IN THE STUDY
actinolite
alunite-Na
alunite-K
apophyllite
beryl
biotite
buddingtonite
calcite?]
cerite REE
chlorite
Chondrodite
clinohumite
diaspore
dickite
diopside
dolomite
dravite
enstatite
dumortierite
Fe-chlorite
elbaite
goethite
hornblende
illite
jarosite
kaolinite
lepidolite
M.L. I\S
sheridanite
monazite REE
montmorilllonite
muscovite
natrolite
nepheline
opal
phlogopite
prehnite
pyrophyllite
saponite
Scapolite
shorl
synchysite REE
szmolokite
topaz
tremolite
KEY
MINERALS IN THE ROUND ROBIN
Kaolinite
Dickite
RR
kaol dik
alun PYROdias TOPAZ
DUMOwm
ill
RR02
RR03
RR05
RR06
RR07
RR08
RR10
RR11
RR12
RR13
RR14
RR17
RR18
RR21
RR22
x
x
x
x
x
RR23
RR26
RR27
RR28
RR29
RR30
RR31
RR35
RR37
RR38
RR40
RR41
RR43
RR45
RR46
RR47
RR48
RR49
RR54
RR55
RR56
RR57
RR60
RR61
RR66
RR67
RR68
RR69
RR70
RR72
RR75
RR76
RR84
RR85
RR86
RR87
RR88
RR89
RR90
RR91
RR92
RR96
RR98
x
x
mont ML
sap
SCA sil
x
dolo CAL chl
AMP PYX tour BIO diop apop preh pump REE hem goe ?
x
x
x
x?
x
x
x
x
x
x
x
x
x
op
x
Diaspore
x
?
x
pyrophyllite
x
x
x
Alunite
x
x
x
??
x
Topaz
x
x
x
x
x
x
x
x
Dumortierite
x
x
x
x
x
x
Xxxxxxxxxxxxx wm
x
x
x
Illite
x
x
X
X
X
Smectite
X
X
X
X
X
X
X
X
X
Mixed Layer I/S
X
Saponite
X
X
X
X
X
X
X
X
Scapolite
X
X
X
Silica
X
X
X
X
X
X
X
X
X
X
Calcite
X
X
Chlorite
X
X
X
X
X
X
X
Dolomite
X
X
X
X
X
X
X
Amphibole
X
X
X
SZM
OP?
X
X
Tourmaline
X
X
X
X
X
X
?
X
PHLG
X
X
X
Biotite
X
X
X
X
X
X
X
X
X
X
ZEO
X
X
Apophylite
X
Prehnite
X
X
X
X
X
X
Pumpellyite
X
X
NH4
BERYL
X
X
Diopside
X
X
X
X
Pyroxene
X
X
X
X
JARO
REE
Hematite
Goethite
ROUND ROBIN SPECTRA: EXAMPLES
Szmolnokite
Elbiate + Lepidolite
Monazite
Dickite + alunite + pyrophyllite
ROUND ROBIN:
MINERAL ID PROGRAMS ENTERED
•GRAMS
•TSG (several versions)
•FEATURE SEARCH
•TNT-MIPS
•IN-HOUSE C+
•MSA (MINERAL SPECTRAL ANALYSIS)
Summary of Programs Submitted to Round Robin:
Strengths and Weaknesses
Program
Strengths
The Spectral Geologist Up to 5 minerals
(TSG) – TSA (The
potentially identified
Spectral Assistant)
Weaknesses
Minor components of
mixtures poorly
identified
TNT MIPS
• Good attempt at
Second and third order
mixtures – up to 5
phases of mixtures not
phases
well identified
• Fe-phases identified
MSA – Mineral
Spectral Analyst
Mineral ‘matrix’ could
work very well for
major components
No measure of
reliability – will always
fit spectra to given
minerals
“In-house” C
Attempt to determine
VNIR phases
Matches only 1 spectra
ROUND ROBIN: Participants and winner
Participant #
Program
Version
Data Base
Country
8011
TNT MIPS
PRO-3.1
SPECMIN
Turkey
8014
TSG
7.1
TSG
Australia
8015
GRAMS
Specmin
USA
8016-1
FS
SPECMIN
Argentina
8017
TSG
unspecified
USA
8018
in-house C
Specmin
ENVi format
USA
8021
TSG
7.1 3.0
Australia
8022
MSA Mineral Spectral Analysis
V3.6
China
8024
TSG
5.03
IN TSG
Netherlands
8025
TSG
7.1
NG TSG?
Australia
8030
TSG
2
answers
in TSG
Chile
8035
proprietary
1
specmin
USA
OBSERVATIONS & COMMENTS
BIGGEST OBSERVATION
not a lot of progress made over the last 10 years
TSG CAN GENERATE HIGHLY VARIABLE RESULTS
Specialized libraries definitely improve matching statistics
Library must be comprehensive to provide accurate answers
The less complicated the procedure, the more accurate will be the
results
Less choices, better results – i.e. special purpose libraries provide
the best answers
Computer program:
The Spectral Geologist v. 5.03
Database: The Spectral Assistant (TSA)
Dr. F.J.A. (Frank) van Ruitenbeek
University of Twente, Netherlands
PRIZE
IF USER DOES NOT KNOW
WHAT ANSWERS ARE WRONG,
HOW CAN ANSWERS FROM
AUTOMATED PROGRAMS BE
EVALUATED??