Introduction to Multisensors

1
Introduction to Multisensors
2
Fundamental Steps for Using Multisensor Data

• Definition of information needs
• - accuracy, time, cost, format
• Data collection
• - data specification, techniques and
  technologies
• Data analysis
• - measurement, classification, estimation

3
Fundamental Steps for Using Multisensor Data
(cont.)

• Verification of the analysis results
• - data quality, accuracy
• Reporting the results
• - format
• Taking action
• - information for decision-making, user format

4
Satellite Systems

• Multispectral Scanner (MSS)

Picture taken from the space shuttle (note
atmospheric scattering)
MSS color composite
5
Satellite Systems

• Landsat 1, 2 and 3
• Earth Resources Technology Satellite (ERTS-1),
  later renamed Landsat-1, launched in 1972 as an
  experimental system to test the feasibility of
  collecting earth resources data from satellites
• Data made publicly available world-wide Open
  Skies Policy
• Carried a multispectral scanner (MSS), imaged a
  185 km swath in 4 wavebands, 2 in visible, 2 in
  near-infrared, spatial resolution of 80 m,
  sun-synchronous orbit, repeat cycle of 18 days
• Each scene contains 7.5 million pixels 30
  million data values

6
Information Extraction Principles for
Hyperspectral Data
7
Brief History
REMOTE SENSING OF THE EARTH Atmosphere - Oceans -
Land
1957 - Sputnik
1958 - National Space Act - NASA formed
1960 - TIROS I
1960 - 1980 Some 40 Earth Observational
Satellites Flown
8
Image Pixels
Thematic Mapper Image
9
Three Generations of Sensors
10
Systems View
11
Scene Effects on Pixel
12
Data Representations

• Spectral Space - Relate to Physical Basis for
  Response

• Feature Space - For Use in Pattern Analysis

13
Data Classes
14
Scatter Plot for Typical Data
15
Bhattacharyya Distance
Mean Difference Term
Covariance Term
16
Vegetation in Spectral Space
Laboratory Data Two classes of vegetation
17
Scatter Plots of Reflectance
18
Vegetation in Feature Space
19
Hughes Effect
G.F. Hughes, "On the mean accuracy of statistical
pattern recognizers," IEEE Trans. Inform.
Theory., Vol IT-14, pp. 55-63, 1968.
20
A Simple Measurement Complexity Example
21
Classifiers of Varying Complexity
22
Classifier Complexity - Continued

• Other types - Nonparametric
• Parzen Window Estimators
• Fuzzy Set - based
• Neural Network implementations
• K Nearest Neighbor - K-NN
• etc.

23
Covariance Coefficients to be Estimated

• Assume a 5 class problem in 6 dimensions

Common Covariance d c d c c d c c c d c c c
c d c c c c c d

• Normal maximum likelihood - estimate
  coefficients a and b

• Ignore correlation between bands - estimate
  coefficients b

• Assume common covariance - estimate coefficients
  c and d

• Ignore correlation between bands - estimate
  coefficients d

24
Example Sources of Classification Error
25
Intuition and Higher Dimensional Space
Borsuks Conjecture If you break a stick in two,
both pieces are shorter than the original.
Kellers Conjecture It is possible to use cubes
(hypercubes) of equal size to fill an
n-dimensional space, leaving no overlaps nor
underlaps.
Counter-examples to both have been found for
higher dimensional spaces.
Science, Vol. 259, 1 Jan 1993, pp 26-27
26
The Geometry of High Dimensional Space
27
Some Implications

• High dimensional space is mostly empty.
• Data in high dimensional space is mostly in a
  lower dimensional structure.

Normally distributed data will have a tendency to
concentrate in the tails Uniformly distributed
data will concentrate in the corners.
28
How can that be?
29
How can that be? (continued)
30
MORE ON GEOMETRY

• The diagonals in high dimensional spaces become
  nearly orthogonal to all coordinate axes

Implication The projection of any cluster onto
any diagonal, e.g., by averaging features could
destroy information
31
Still More Geometry

• The number of labeled samples needed for
  supervised classification increases rapidly with
  dimensionality

In a specific instance, it has been shown that
the samples required for a linear classifier
increases linearly, as the square for a quadratic
classifier. It has been estimated that the number
increases exponentially for a non-parametric
classifier.

• For most high dimensional data sets, lower
  dimensional linear projections tend to be normal
  or a combination of normals.

32
A Hyperspectral Data Analysis Scheme
200 Dimensional Data
Class Conditional Feature Extraction
Feature Selection
Classifier/Analyzer
Class-Specific Information
33
Finding Optimal Feature Subspaces

• Feature Selection (FS)

• Discriminant Analysis Feature Extraction (DAFE)

• Decision Boundary Feature Extraction (DBFE)

• Projection Pursuit (PP)

Available in MultiSpec via WWW at
http//dynamo.ecn.purdue.edu/biehl/MultiSpec/ Ad
ditional documentation via WWW at
http//dynamo.ecn.purdue.edu/landgreb/publication
s.html
34
Hyperspectral Image of DC Mall
HYDICE Airborne System 1208 Scan Lines, 307
Pixels/Scan Line 210 Spectral Bands in 0.4-2.4
µm Region 155 Megabytes of Data (Not yet
Geometrically Corrected)
35
Define Desired Classes
Training areas designated by polygons outlined in
white
36
Thematic Map of DC Mall
Legend
Operation CPU Time (sec.) Analyst Time Display
Image 18 Define Classes lt 20 min. Feature
Extraction 12 Reformat 67 Initial
Classification 34 Inspect and Mod. Training
5 min. Final Classification 33 Total 164 sec
2.7 min. 25 min.
Roofs Streets Grass Trees Paths Water Shadows
(No preprocessing involved)
37
Hyperspectral Potential - Simply Stated

• Assume 10 bit data in a 100 dimensional space.

• That is (1024)100 10300 discrete locations

Even for a data set of 106 pixels, the
probability
of any two pixels lying in the same discrete
location
is vanishing small.
38
Summary - Limiting Factors

• Scene - The most complex and dynamic part

• Sensor - Also not under analysts control

• Processing System - Analysts choices

39
Limiting Factors
Scene - Varies from hour to hour and sq. km to
sq. km
Sensor - Spatial Resolution, Spectral Bands, S/N
Processing System -
- Informational Value
- Separable

• Classes to be labeled

- Exhaustive

• Number of samples to define the classes

• Features to be used

• Complexity of the Classifier

40
Source of Ancillary Input
Possibilities
- From the Ground

• Ground Observations

- Of the Ground

• Imaging Spectroscopy

• Previously Gather Spectra

• End Members

41
Use of Ancillary Input
A Key Point

• Ancillary input is used to label training samples.

• Training samples are then used to compute class
  quantitative descriptions

Result

• This reduces or eliminates the need for many
  types of preprocessing by normalizing out the
  difference between class descriptions and the data

42
Satellite Systems

• Landsat 4 5
• Landsat 4 deactivated shortly after launch, but
  remains in orbit
• Landsat 5 carries a multispectral scanner (MSS)
  a Thematic Mapper (TM), imaged a 185 km swath
• 7 wavebands from the visible blue to the thermal
  infrared, spatial resolution of 30 m except the
  thermal band (120 m), sun-synchronous orbit,
  repeat cycle of 16 days
• each scene contains about 36 million pixels 250
  million data values

43
Landsat 5 - 7 wavebands
44
Satellite Systems

• Landsat 1 operated from 1972 to 1978
• Landsat 2 from 1975 to 1983
• Landsat 3 from 1978 to 1983
• Landsat 4 deactivated shortly after launch
• Landsat 5 launched in 1984 still in use
• Landsat 6 launched in 1993 but did not achieve
  orbit - the first underwater satellite
• Landsat data is available from EOSAT (Earth
  Observation Satellite Company)
• Landsat 7 now in orbit with 15 m resolution.
  Operated by NOAA, data is provided at cost.

45
Satellite Systems

• SPOT
• French satellite launched in 1986
• Repeat cycle of 26 days
• Swath width of 117 km
• Sun-synchronous orbit, carries 2 pointable
  scanners
• Panchromatic mode with spatial resolution of 10 m
  and multispectral mode with spatial resolution of
  20 m images in 3 wavebands
• Potential to view a location from adjacent
  satellite paths stereoscopic imaging (useful
  for topographic mapping)

46
SPOTs Steerable Mirror
47
Uses of Landsat SPOT Data

• Geology - used for mapping in mineral and
  petroleum exploration
• Agriculture - used to estimate crop quantities,
  monitor condition of crops
• Forestry - to estimate forest losses caused by
  fires, clear cutting disease to provide forest
  inventory data used for comparative forest land
  valuation

48
Uses of Landsat SPOT Data

• Land use planning - mapping current land cover,
  change detection, route location planning
• High resolution satellite imagery is being used
  as a substitute for high-altitude aerial
  photography

Blue-water Green-forest Yellow-suburban
Red-urban
49
Uses of Landsat SPOT Data

• For monitoring rangeland condition, wildlife
  habitat, identify water pollution, identify
  flooded areas, to aid in the assessment of damage
  caused by natural disasters

50
(No Transcript)
51
(No Transcript)
52
(No Transcript)
53
Post Image Processing
54
Ocean Monitoring Satellites

• Oceans are important natural resource - difficult
  to map monitor over large areas or for long
  time periods
• Satellites provide complete coverage at regular
  intervals
• Landsat SPOT data used extensively to monitor
  sediment and chlorophyll concentrations,
  phytoplankton and pollution in marine and
  fresh-water environments also to map water
  depths

55
Ocean Monitoring Satellites

• What is phytoplankton?
• single-celled plants living in surface waters
  that form the base of the marine food chain -
  critical to the biological productivity of the
  ocean, including the production of commercial
  fish and shellfish

CZCS Data of Atlantic Ocean
56
Ocean Monitoring Satellites

• The Coastal Zone Color Scanner
• Launched by the US in 1978, operated till 1986
• 1600 km swath width, spatial resolution of 825 m,
  images in 6 wavebands (4 in visible, 1 in
  near-infrared and 1 in thermal infrared)
• Measures ocean color and temperature
• Monitors changes in ocean current location,
  position of upwelling areas
• Estimates the sediment content in coastal waters
• Also used for the detection of acid waste
  pollution

57
Meteorological Satellites

• Provides more often coverage but at lower
  resolution (also means less expensive than
  Landsat SPOT)
• Designed primarily to collect weather data, but
  often used for natural resource monitoring over
  large areas
• NOAA Satellites
• Advanced Very High Resolution Radiometer (AVHRR)

58
AVHRR

• Spatial resolution of 1.1 km
• Swath width of 2400 km
• Provides daily global coverage
• Used to monitor snow cover, assess snow depths,
  monitor floods, detect map forest fires,
  monitor crop conditions, monitor dust and sand
  storms, identify geologic activities like
  volcanic eruptions, for mapping regional drainage
  networks, physiography geology

59
AVHRR Thermal Channel
60
Different Sensors and Resolutions
sensor spatial
spectral
radiometric temporal -------------------------
--------------------------------------------------
------------------------------------- AVHRR
1.1 and 4 KM 4 or 5 bands 10
bit 12 hours 2400 Km
.58-.68, .725-1.1, 3.55-3.93
(0-1023) (1 day, 1 night)
10.3-11.3, 11.5-12.5
(micrometers) Landsat MSS 80 meters
4 bands 6 bit
16 days 185 Km
.5-.6, .6-.7, .7-.8, .8-1.1
(0-63) Landsat TM 30 meters 7
bands 8 bit 14
days 185 Km .45-.52,
.52-.6, .63-.69, (0-255)
.76-.9, 1.55-1.75,
10.4-12.5, 2.08-2.3 um SPOT P
10 meters 1 band
8 bit 26 days
60 Km .51-.73 um
(0-255) (2 out of 5) SPOT X 20
meters 3 bands 8 bit
26 days 60 Km
.5-.59, .61-.68, .79-.89 um (0-255)
(2 out of 5)
61
Spatial and Temporal Resolution

As spatial resolution increases, the revisit time
is also increased, as are the applications that
are appropriate and the cost
62
Active Remote Sensing Systems

• Passive optical - deals with reflected sunlight
  and re-emitted thermal radiation, from the very
  short wavelength ultra-violet to the longer
  infra-red region
• Active systems - a pulse of energy is sent from
  the sensor towards the target the energy
  reflected back to the sensor is measured and used
  to produce images

63
Active (Radar) Systems
Data are processed into useful information Origina
lly developed for military purposes
64
Active Remote Sensing Systems

• RADAR - Radio Detection and Ranging technology
  was developed during WWII to detect enemy
  aircraft
• Because the system provides its own source of
  illumination, it has day-and-night imaging
  capability.
• The microwave wavelengths used are not blocked by
  clouds, hence it can provide all-weather coverage.

65
RADAR Imaging Systems

• The strength of reflected energy depends on
• Surface roughness
• Orientation of the terrain
• Electrical properties of the surface material
• Note metal reflects better than soil, wet soil
  reflects better than dry soil
• Routinely used to map areas with persistent cloud
  cover (tropical/sub-tropical regions)
• Stereo imaging used to map terrain relief

66
Remote Sensing Analysis

• 3 categories measurement, classification
  estimation
• Measurement Analyses
• Use remotely sensed data to measure features or
  phenomena on the earths surface
• Surface temperature, elevation from stereo
  images, vegetation biomass condition
• Normalized difference vegetation index (NDVI)
  comparing vegetation reflectance in the red and
  near-infrared bands

67
Remote Sensing Analysis

• Classification Analyses
• Identify and map areas with similar
  characteristics
• Conditions as varied as soil types, crop types,
  forest species composition, geologic strata, and
  land cover are routinely assessed by visually
  interpreting remotely sensed images
• Digital image processing provides an automated
  classification method to complement the results
  of visual interpretation to classify large
  volumes of data at high speed

68
Remote Sensing Analysis

• Classification Analyses
• Monitor water quality, assess soil erosion
  potential, quantify environmental effects,
  monitor urban sprawl post-development
  conditions
• Estimation Analyses
• Estimates quantity
• Expected crop production for a region, forest
  resource inventory
• Generally uses a classification of the data
• Accuracy of classification affects estimation

69
Remote Sensing Analysis

• Field data collection is an integral part of most
  analyses, but field data collection is expensive
  and time-consuming
• Remotely sensed data can reduce the quantity of
  field data and decrease the amount of time needed
  to produce the estimate
• All analyses must be checked against ground
  truth

70
12 Steps of Digital Image Processing
1) State the objective of the project 2) Acquire
data 3) Assess data quality 4) Correct the
data 5) Enhance the data 6) Correlate the data
with ancillary data
71
12 Steps of Digital Image Processing (Cont.)
7) Select training area 8) Determine the spectral
characteristics of the training area 9) Classify
the data 10) Evaluate the results of the
classification 11) Refine training areas and
classifications if necessary 12) Communicate the
results
72
Digital Image Processing

• Our eyes actually do image processing, analyzing
  shape, size, pattern, shadow, tone/color, texture
  and context
• Computerized image processing let's us see beyond
  the limits of our eyes, integrating spectral
  channels, dividing them by one another, etc.
• But remember, the computer can't think (a
  positive and a negative attribute)
• The human eye can recognize same feature under
  different illumination levels - best pattern
  recognition device

73
Change Detection
The ability to monitor change is one of the
benefits of remote sensing We can monitor human
and natural changes in the landscape
74
Change detection-Mississippi flood
St Louis, MO (1993)
75
1988 Yellowstone Fires
76
Fire Management
Before After
Yellowstone National Park fires, 1988 Nearly 50
of the park burned Over 321,000 Ha
burned areas in pink
77
Remote Sensing Applications

• There are many ways remote sensing is used
• Geostationary weather monitoring
• Cartography and mapping
• Natural resource management
• Disaster management warning - fire,
  earthquakes, etc.
• New hi-resolution systems are equal to aerial
  photos
• Data for Geographic Information Systems (GIS)
• Atmospheric and Marine

78
Other Satellite Applications

• Remote sensing of other planets
• Distance learning
• Telecommunications
• Telemedicine
• Global Positioning Systems (Navstar/Glonass)
• Telemetry systems (Argos, Doris)
• Search and Rescue (Cospas/Sarsat)

79
Cospas-Sarsat
Cospas-Sarsat is a satellite system designed to
provide distress alert and location data to
assist search and rescue (SAR) operations, using
spacecraft and ground facilities to detect and
locate the signals of distress beacons operating
on 406 Megahertz (MHz) or 121.5 MHz. The
position of the distress and other related
information is forwarded by the responsible
Cospas-Sarsat Mission Control Center (MCC) to the
appropriate SAR authorities. Its objective is to
support all organizations in the world with
responsibility for SAR operations, whether at
sea, in the air or on land.
80
ARGOS, ALTIMETRY and DORIS (CLS is responsible,
on behalf of the French space agency CNES, for
the day-to-day operation of the DORIS instruments
in service on SPOT 2, SPOT 4 and TOPEX/POSEIDON.)
An ARGOS transmitter, connected to a tide gauge,
enables continuous in situ measurement of sea
surface height.
DORIS provides reference altimetrydata and
measures verticalcoastal movements. Doris
determines the location of a satellite or ground
location beacon with centimeter accuracy.
Satellite altimetry highlights global sea level
variability.