Transcript Remote Sensing Application Supporting Regional Database

Remote Sensing Applications Supporting
Regional Transportation Database Development
CLEM 2001
August 6, 2001
Santa Barbara, CA
Chris Chiesa, [email protected]
(520) 326-7005 ext. 106
Remote Sensing Application Supporting Regional
Database for Transportation Planning
In Partnership with:
Presentation Overview
Project Summary
Project Objective
Approach
Benefits
Technical Discussion
Land Cover Change Detection and
Mapping
Road Feature Characterization and
Extraction
Project Objective
Develop tools and methods to facilitate
regional transportation road network
database development and maintenance
Utilize commercial remote sensing sources to
identify and map changes in land use and
transportation infrastructure
Automate procedure for extracting and attributing
road vectors
Develop procedures within COTS software
environment (ERDAS IMAGINE / CAFÉ)
Promote awareness of tools and processes
through outreach activities
Training / Workshops
Web-based Interactive Tutorial
Commercial Remote Sensing Sources
High-resolution IKONOS
LANDSAT Thematic Mapper
Approach
1.
Use multi-date Landsat Thematic Mapper imagery
to identify areas within a large region where
intensive urban development (hot spots) has
occurred.
May 26, 1984
June 15, 2000
Urban development between 1984
and 2000
Approach
2.
Acquire high-resolution (IKONOS) imagery over
hot spots and enhance road network with one or
more spectral features developed for the types of
roads present and the geographic environment.
IKONOS panchromatic band
(1-meter)
IKONOS false color composite
(4-meter)
Road feature derived from
linear combination of
IKONOS multi-spectral
bands (4-meter)
Approach
3.
Extract road locations in newly developed
regions and store as vector coverage’s using
Veridian’s Lines of Communication (LOC)
extraction software.
Approach
4.
Assign attributes (e.g. surface type, width) to
vector coverages.
2-lane roads
3-lane roads
Benefits
The LANDSAT program provides an inexpensive
means of identifying landcover change over a
large area.
Landsat Coverage
IKONOS Coverage
Benefits
Automated (i.e., user-assisted) road extraction
using road spectral features and/or LOC toolkit
can be faster, less tedious and less error prone
than traditional processing of hand digitizing
from aerial photography or satellite imagery.
Panchromatic Aerial Photograph
Road feature derived from
Multispectral Imagery
Change Detection and Feature
Extraction Process
Change detection over a large area
Radiometric normalization
Categorize both dates
Categorical change
Radiometric change
Hybrid change
Feature extraction and attribution
Identify regions of intensive development
Generate road features.
Extract road network
Attribute road network
Procedure Overview
Date 1
Geo-coded
Date 2
Geo-coded
Radiometric Correction
Radiometric
Normalization
Process
Radiometric Change Detection
Date 2 Geocoded and
normalized to
Date 1
Categorical
Process
Categorical
Process
Date 2
Categorized
Image
Date 1
Categorized
Image
Categorical
Change
Process
Radiometric
Change Detection
Process
Change
Magnitude
and Change
Direction
Categorical Processing
Categorical
Change
Detection
Image
Categorical Change Detection
Hybrid Change Detection
Hybrid Change
Detection
Process
Hybrid
Change
Product
Acquire Data …
Acquire 2 dates of LANDSAT data
Summer season
Cloud free
Same time of year
Mid-Michigan on June 8, 1986
Mid-Michigan on June 6, 2000
Categorize Both Dates…
Label resultant clusters into water, vegetation, bare
ground, and urban areas, as appropriate.
Water
Bare ground
Vegetation
Urban
Unsupervised clustering of Landsat
Thematic Mapper image over portion of
Michigan on June 8, 1986.
Unsupervised clustering of Landsat
Thematic Mapper image over portion of
Michigan on June 6, 2000
Categorical Change …
Recode categorized files to urban/non-urban.
Water
Bare ground
Vegetation
Urban
Urban
No data
Categorical Change…
Combine binary files from both dates to determine
where urban changes have occurred.
Date1
Date2
Urban on date 1,
not on date 2
Urban on both
dates
Urban on date 2,
not on date 1
No data
Radiometric Change Detection
This change magnitude channel
shows differences in two dates of
Landsat imagery for a region in
Michigan.
Brighter areas indicate higher
magnitudes of change. Often a
threshold from this channel is
established so that only changes
above a certain magnitude will be
considered when extracting
changes of interest.
Increasing difference between pixel values from
date 1 to date 2 input images.
Radiometric Change Detection
Band 2
Color
Band 3
Sector Code
0
1
2
3
4
5
6
7
The sector code channel provides information on the “direction”
or nature of change. Each color corresponds to a sector code.
Each sector code relates to a specific combination of changes
observed in image bands as shown in the table above. For
example, sector code 6, shown in orange in the image to the left,
shows areas that have increased spectral reflectance in bands 2
and 3, and decreased spectral reflectance in band 4.
Blue
Green Red
Near
IR
Band 4
Radiometric Change Detection…
Create a change image composition (CIC) and
determine sector codes that best represent
urban areas.
Hybrid Change
Advantages are:
1.
2.
Labels from categorization
Reduction in false categorical change from CVA
Hybrid urban change product of Delta
Township in Michigan. Changed areas
are annotated in yellow over a Landsat
Thematic Mapper False color
composite
Feature Extraction and Attribution
Identify geographic locations of localized
regions in LANDSAT change product where
intensive development has occurred
Generate road features
Extract road network
Attribute road network
Identify Geographic Locations
Identify areas in the Landsat hybrid change
product where urban change has occurred
and order IKONOS data
Order and Receive Data
Acquire IKONOS data over area of interest
IKONOS natural color composite
with blue band displayed in blue,
green band displayed in green,
and red band displayed in red.
IKONOS false color composite
with green band displayed in blue,
red band displayed in green, and
near infrared band displayed in
red.
IKONOS panchromatic band
Generate Road Features…
This scatterplot illustrates how different
landcover materials can be separated in
2-dimensional space (2 spectral bands).
The arrow shows a direction that can be
described as a linear combination of
these two bands. The dashed line
indicates that both concrete and asphalt
can be separated from the other
materials with this 2-dimensional
feature. Often features are created by
using multiple bands ( > than 2
dimensions)
Generate Road Features…
This plot illustrates how well a specific
4-band spectral feature will work in
isolating certain landcover material from
other materials in the image. Natural
materials are projected towards a
categorical value of 1, while man made
materials are projected towards a
categorical value of 2. The vertical
dashed line between these two
categories illustrates that this equation
will work in separating these 2
categories. In the feature created, man
made materials will appear as the
brightest objects and natural materials
will appear as darker objects. Level
slicing the feature at around 150 will
separate the two.
Generate Road Features…
Apply coefficients of spectral feature to data
and produce road feature.
[(Band 1 * -.0256) + (Band 2 * .0915) + (Band 3 * .1346) + (Band 4 * -.2241)] + 148
Weighted average of satellite raw bands
Adjusts data values
into 0-255 range for
unsigned 8-bit output
Generate Road Features…
False color composite of IKONOS data displayed
with green band in blue, red band in green, and
near infrared band in red
IKONOS road feature derived from a linear
combination of the raw bands
Extract Road Network
Use road feature as input to LOC toolkit and
semi-automatically extract roads.
Convert to vector coverage.
Extract Road Network…
Extract Road Network…
Attribute Road Network
3-lane roads
2-lane roads
2-lane roads
Process Summary
Landsat imagery provides broad spatial and temporal
coverage over which to observe land changes
Hybrid change detection offers advantages over
traditional post-classification change detection in that it
also incorporates important radiometric change
information and allows “thresholding” of changes
IKONOS imagery provides high spatial resolution to
identify the specific transportation features that
constitute the changes observed in Landsat imagery
Using a “Road Feature” helps maximize the
differentiability of roads and background classes in the
imagery
Semi-automated extraction and labeling tools facilitate
the process of developing GIS database layers from
these remote sensing sources
Questions?
Please contact:
Chris Chiesa
Veridian Systems
4400 East Broadway, Suite 116
Tucson, AZ 85711
(520)326-7005 ext. 106
[email protected]
www.veridian.com