ENVI Tutorial:
Archaeology Hyperspectral
Analysis
Table of Contents
OVERVIEW OF THIS TUTORIAL.....................................................................................................................................2
Objectives ........................................................................................................................................................2
Files Used in This Tutorial ..................................................................................................................................2
Background ......................................................................................................................................................2
PROCESSING FLOW...................................................................................................................................................4
HYPERSPECTRAL TECHNIQUES FOR ARCHAEOLOGICAL ANALYSIS ............................................................................................5
 Tutorial: Archaeology Hyperspectral Analysis
Overview of This Tutorial
This tutorial presents a case history for use of hyperspectral techniques in archaeological analysis using Multispectral
Infrared and Visible Imaging Spectrometer (MIVIS) data from Selinunte, Sicily, Italy. It quickly guides you through ENVI s
end-to-end hyperspectral tools (EFFORT MNF PPI n-D Visualization Spectral Mapping) to produce image-
derived endmember spectra and image maps. For more detail and step-by-step procedures on performing a complete
hyperspectral analysis, refer to the series of ENVI hyperspectral tutorials (introductory through advanced) before
attempting this tutorial, and refer to ENVI Help when necessary. The MIVIS data are courtesy of ITT Visual Information
Solutions Italia, S.R.L. and the Italian National Research Council (CNR). Used with permission.
The purpose of this tutorial is not to teach you how to run the ENVI tools, but how to apply the methodology and tools to
a general hyperspectral remote sensing problem.
Objectives
Apply ENVI end-to-end hyperspectral processing methodology to an archaeology case study
Gain hands-on experience running the procedures rather than reviewing preprocessed results (althrough
preprocessed results are provided for comparison)
Perform data exploration in a loosely structured framework
Compare analysis results with known ground information
Files Used in This Tutorial
ENVI Resource DVD: envidata/selmivis
Required files
File Description
selinunte_rad.bil (.hdr)
MIVIS 92-band radiance subset
selinunte.ann
ENVI annotation for selinunte_rad.bil
mivis_waves.txt
Wavelength file for MIVIS
selinunte_ff.roi
ROI used for Flat Field correction
Background
Selinunte is an abandoned ancient Greek city with ruins of an Acropolis and numerous temples, located on the southwest
coast of Sicily, Italy, in the province of Trapani. The city was founded between 650 and 630 BC and destroyed by the
Carthaginians in 409 BC.
The city of Selinunte, the Acropolis, is situated on high ground overlooking the Mediterranean Sea. Selinunte is
surrounded by a park (Parco Archeologico di Selinunte) extending approximately 1,300 meters east to west. There are
five temples within the Acropolis, some of which have been excavated. Others are essentially jumbled piles of stones. The
entire Acropolis has never been completely excavated, but the surrounding walls were mostly reconstructed by
archaeologists in 1927. See http://www.bestofsicily.com/selinunte.htm for additional information.
The Italian National Research Council (CNR), who owns and operates the MIVIS Daedalus AA5000 MIVIS sensor, flew it
over the Selinunte Park during 1996. MIVIS is a 102-channel scanner with 92 bands covering the approximately 0.4 to 2.5
micrometer (VNIR-SWIR) range and 10 bands over the 8-14 micrometer (LWIR) range. Spatial resolution is 3 m. For this
tutorial, you will use 90 of the 92 VNIR-SWIR bands (excluding bands 63 and 87 as bad bands).
The following figure shows selected archaeological features annotated on a MIVIS image (top) and a reference map
(bottom):
2
ENVI Tutorial: Archaeology Hyperspectral Analysis
 Tutorial: Archaeology Hyperspectral Analysis
3
ENVI Tutorial: Archaeology Hyperspectral Analysis
 Tutorial: Archaeology Hyperspectral Analysis
Processing Flow
The following figure shows the hyperspectral processing flow implemented in ENVI.
4
ENVI Tutorial: Archaeology Hyperspectral Analysis
 Tutorial: Archaeology Hyperspectral Analysis
Hyperspectral Techniques for Archaeological Analysis
1. Examine MIVIS radiance data: Load selected MIVIS bands as gray scale images. Perform animation. Extract
radiance signatures for areas of high variability. Examine radiance spectra for evidence of absorption features.
Determine bad spectral bands. Load color composite images designed to enhance spectral contrast. Determine
spectral subset(s) to use for mineral mapping. Use the following files:
File Description
selinunte_rad.bil (.hdr)
MIVIS 92-band radiance subset
mivis_waves.txt
Wavelength file for MIVIS
2. Correct MIVIS data to apparent reflectance using the Flat Field method: Use the file
selinunte_ff.roi as the ROI for the Flat Field correction. Create your own apparent reflectance output, or
review the results in the file selinunte_ff.bil. Display a gray scale image of the apparent reflectance data.
Extract reflectance signatures and examine them for spectral features. Animate the data and extract spectra for
areas of high variability. Determine bad spectral bands. Load color-composite images designed to enhance
spectral contrast. Determine spectral subset(s) to use for mineral mapping. Extract reflectance signatures for
vegetation and geologic materials. Compare to spectral libraries.
3. Apply MNF transform and determine data dimensionality: Review MNF eigenvalue plot to determine the
break in slope and relate to spatial coherency in MNF eigenvalue images. Determine MNF cut-off between signal
and noise for further analysis. Create your own MNF-transformed dataset.
4. Apply PPI analysis to the MNF output: Rank the pixels based on relative purity and spectral extremity. Use
the FAST PPI option to perform calculations quickly in system memory, creating the PPI image. Display the PPI
image, examine the histogram and threshold, and create a list of the purest pixels, spatially compressing the
data. Generate your own PPI results.
5. Perform n-D Visualization of the high PPI value pixels: Use the high-signal MNF data bands to cluster the
purest pixels into image-derived endmembers. Rotate the MNF data interactively in three dimensions, or spin in
several dimensions and paint pixels that occur on the points (extremities) of the scatter plot. Use Z Profiles
connected to the apparent reflectance data and the n-D Visualizer to evaluate spectral classes. Use class
collapsing to iteratively find all of the endmembers. Evaluate mixing and endmembers. Save your n-D results to a
saved state file (.ndv). Export classes to ROIs and extract mean spectra. Compare mean spectra to spectral
libraries. Use spectral/spatial browsing to compare image spectra to ROI means. Extract endmembers and create
your own ROIs.
6. Use ENVI s mapping methods: Map the spatial occurrence and abundance of minerals at Selinunte. Hint: the
temples and portions of the Acropolis are built of Calcarenite and/or Limestone, which have distinct spectral
signatures:
At a minimum, try the Spectral Angle Mapper (SAM) and Mixture Tuned Matched Filter (MTMF) methods. Use
SAM to determine spectral similarity to image endmember spectra. If time permits, try a SAM classification using
spectral libraries. Be sure to evaluate the rule images. Use the MTMF mapping method to determine material
abundance. Be sure to use both the MF and infeasibility images in a 2D scatter plot to select the best matches
(high MF and low infeasibility score). Compare abundance image results to the endmember spectra and spectral
libraries using spatial and spectral browsing.
5
ENVI Tutorial: Archaeology Hyperspectral Analysis