Remote Sensing I-2011 2

Module 2, Remote Sensing 1 1st semester Lecture Notes of Prof. Dr. M.
Hahn
MSc Course PHOTOGRAMMETRY AND GEOINFORMATICS
Module 2 Photogrammetry and Remote Sensing

Lecture Notes Prof. Dr. M. Hahn WS 2011/2012
Topic: Remote Sensing (Part 1)
Last printed 20 September 2011
Page 1 of
56
Module 2, Remote Sensing 1 1st semester Lecture Notes of Prof. Dr. M. Hahn
Remote Sensing
(Part 1)
Table of Contents
1. Basic Principles of Remote Sensing
1.1. 1.2. 1.3. 1.4. 1.5. 1.6. 1.7. Definitions, overall Remote Sensing process Electromagnetic radiation The electromagnetic spectrum Interaction of electromagnetic radiation with the atmosphere Interaction of electromagnetic radiation with Earth-surface material Energy sources and sensing Satellite images and visualization
2. Preprocessing of remotely-sensed data

2.1. 2.2. 2.3. 2.4. Removal of data errors Registration and geometric correction Atmospheric correction Sensor calibration
3. Classification 3.1. 3.2. 3.3. 3.4. Concept of supervised and unsupervised classification Scatterplot and decision making Supervised classification Unsupervised classification
Page 2 of
56
Chapter 1 Basic principles of Remote Sensing

1.1 Definitions, Overall Remote Sensing process
Remote Sensing is the science (and to some extent: art) of acquiring information about the Earths surface without actually being in contract with it. This is done by sensing and recording reflected or emitted energy and processing, analysing and applying that information. Source: Canada Centre for Remote Sensing, CCRS Tutorial Remote Sensing: the science and art of obtaining useful information about an object, area, or phenomenon through the analysis of data acquired by a device that is not in contact with the object, area or phenomenon under investigation. T.M. Lillesand and R.W. Kiefer Remote Sensing and Image Interpretation, Wiley book Remote Sensing may be broadly defined as the collection of the information of natural sources and environmental information about an object without being in physical contact with the object. The term Remote Sensing is restricted to methods that employ electromagnetic energy as the means of detecting and measuring target characteristics. F.F. Sabins, Remote Sensing principles and interpretation, Freeman book
Page 3 of
56
The science of Remote Sensing consists of the interpretation of measurements of electromagnetic energy reflected or emitted by a target from a vantage-point that is distant from the target. Earth observation (EO) by Remote Sensing is the interpretation and understanding of measurements P.M. Mather Computer processing of Remotely-sensed images, Wiley book Aircraft and satellites are the common platform from which Remote Sensing observations are made. F.F. Sabins Remote Sensing principle and interpretation, Freeman book
Elements of an overall Remote Sensing process
Energy Source The first requirement for remote sensing is to have an energy source which illuminates or provides electromagnetic energy to the target of interest. Radiation and the Atmosphere As the energy travels from its source to the target, it will interact with the atmosphere it passes through.
Page 4 of
56
This interaction may take place a second time as the energy travels from the target to the sensor.
Source: CCRS Tutorial
Interaction with the Target The energy (electromagnetic radiation) interacts with the target depending on the properties of both the target and the radiation.
Recording of Energy by the Sensor After the energy has been scattered by, or emitted from the target, a sensor is required to collect and record the electromagnetic radiation.
Page 5 of
56
Transmission, Reception, and Processing The energy recorded by the sensor has to be transmitted, often in electronic form, to a receiving and processing station where the data are processed into an image (hardcopy and/or digital).
Interpretation and Analysis The processed image is interpreted, visually and/or digitally (image analysis), to extract information about the target which was illuminated.
Application The extracted information assists to solve a particular problem.
1.2 Electromagnetic radiation

Electromagnetic energy /radiation is the means by which information is transmitted from an object (target) to a sensor.
1.2.1 Basic terminology Energy - the capacity to do work, expressed in J (Joules) Radiant energy - the energy associated with electromagnetic radiation (EMR)
Flux of energy
- the rate of transfer of energy from one
Page 6 of
56
place to another (Latin word, meaning = flow) - is measured in W (Watts) Radiant flux density - to understand the interaction between electromagnetic Radiant flux is the rate of transfer radiation and surfaces of radiant
(=electromagnetic) energy Density implies variability over the two-dimensional
surface on which the radiant energy falls.
Radiant flux density is the
magnitude of the radiant flux
that is incident upon or, conversely, is emitted by a surface of unit area (measured in W/m2) (if radiant energy falls upon a surface)
== Irradiance
== Radiant emittance ( if the energy flow is away from surface)
Example
thermal energy emitted by the Earth
solar energy reflected by the Earth
Radiance
- is the radiant flux density transmitted from a unit
area on the Earths surface as viewed through a unit solid (3D) angle.
is measured in steradians = 3 D equiv. of the radian
Page 7 of
56
Assume: Diffuse reflectance = radiance incident
upon the surface is back scattered in all upward directions
surface
Then:
(based on diffuse reflection)
A proportion of the radiant flux might be measured per unit solid viewing angle --this proportion is the radiance flux
surface normal solid (3D ) angle
source area
Radiance is measured in watts per square meter per steradiant [ W/(m2*sr) ]
Page 8 of
56
Reflectance --- is the ratio between the radiant emittance of an object.
irradiance
and the
Diffuse reflectance ( see above) Specular reflectance: = angle of incidence and angle of reflectance
are equal and no
scattering
occurs at the surface.
angle of incidence
angle of reflection
Remarks: 1 When remotely-sensed images collected over a time period (= multi-temporal images) are to be compared it is common radiance values recorded by the sensor
practice to convert the into reflectance factors
in order to eliminate the effects of
variable irradiance 2
over the seasons of the year. particular wavebands
All these described quantities refer to
rather than to the whole electromagnetic spectrum. Precede the terms by the adjective spectral Spectral reflectance, spectral irradiance,etc.
Page 9 of
56
1.2.2 Nature of electromagnetic radiation (or the view of quantum mechanics)
Controversy in physics over the past 250 years EMR: wave theory corpuscular theory
considers radiation as a wave form (wave-like form of energy)
considers radiation as a stream of particles so-called photons
Importance to remote sensing
the wave-like characteristics of EMR allows the distinction with regard to wavelength e.g. microwave, infrared radiation
in order to understand the interactions between EMR and the Earths atmosphere and surface.
Todays view of quantum mechanics: EMR is both a wave and a stream of particles.
1.2.3 Wave characteristics of electromagnetic radiation EMR is travelling at a velocity c (=speed of light) equal to 3*108 m/s in a sinusoidal, harmonic fashion. EMR consists of an electrical (E) field and a magnetic (M) field
Page 10 of 56
Source: CCRS tutorial
Characterisation of electromagnetic waves
Wavelength ---- length of one wave cycle = distance between successive wave crests
Frequency ---- number of cycles (or crests) of a wave passing a fixed f point per unit of time ( = 1 sec)
Wavelength and frequency are related according to c= * f c --- speed of light (is essentially constant) = 3*108m/s (in a vacuum) --- unit is [m] or mm, f --- unit is [Hz=cycle/sec] m, m inverse is period T = 1/f
= time elapsed in seconds per cycle

Last printed 20 September 2011 Page 11 of 56
Examples:
Source: CCRS tutorial
Explanation i) 2.5 cycles per second f = 2.5Hz period = 0.4 sec per cycle ii) iii) 4 Hz 1.5 Hz
Example 1: Given f = 4 Hz, calculate : with c = *f = 3*108 m /sec = 300 000 km/sec and f = 4 Hz = c/f = (3*108m/sec)*(1/4)sec = 75 000 km
Page 12 of 56
Example 2: Given wavelength in micrometers, = 1m, calculate f: = 1m (Near infrared) = 10-6m f = (3*108m/sec) / (1/10-6m) = 3*1014 Hz
These examples demand for a closer look to the EM spectrum.
1.2.4 Corpuscular characteristics of electromagnetic radiation
In the particle description, electromagnetic energy travels in quanta (discrete units) of energy.
The energy of a quantum is given as Q= h * f Q = energy of a quantum h = Plancks constant f = frequency (Joules J) h=6.26*10-34 J*sec (Hz=1/sec)
Energy Q is delivered to a target. Note: delivery is on a probabilistic basis not in such a
way that it is evenly spread over the wave
Page 13 of 56
Relate the wave model to the quantum model
Substituting
f = c/
into
Q = h*f
yields
Q = h*c/
h*c = constant
Conclusion: The shorter the wavelength , the higher the energy content
and vice visa
shorter wavelengths are easier to sense.
Page 14 of 56
The Electromagnetic Spectrum

The electromagnetic spectrum represents the continuum of electromagnetic energy from extremely to extremely short wavelengths (cosmic and gamma rays)
long
wavelengths (radio and television waves)
Page 15 of 56
The names assigned to regions of the spectrum make a discussion of the spectrum more convenient. In each of the regions wavelengths behave similarly mechanisms. However the division between UV and visible or microwave and thermal infrared is not hard. The regions blur into each other. adjacent
or are generated by similar
Three regions are of particular importance for RS:
a) The visible spectrum (visible light) is so called because it is detected by the eyes, whereas other forms of EMR are invisible to the unaided eye. The spectrum range of visible light is 0.4-0.7 um wavebands are perceived as particular colours:
waveband violet -- blue -- green -- yellow - orange -- red
0.40
0.46
0.50
0.58 0.60 0.62
0.70 m
Blue, green and red are the primary colours or wavelengths of the visible spectrum. All other colours can be formed by combining R-G-B.
Page 16 of 56
b) The infrared spectrum (infrared = beyond the red) covers the wavelength range from approximately 0.7 m to 100 m
Regions:
m Radiation property
Near IR
0.7
Mid IR
1.3 3.0
Thermal/Far IR
100
reflective reflected energy visible light
emissive, radiative, thermal emitted from Earths surface in the form of heat
Note: different definitions/boundaries are found in the literature.
c) The microwave spectrum ranges from submillimetre to 1 (to 3) metres further subdivided in bands : K, X, C, S, L, P band
Some microwave sensors can detect small amounts of radiation at those wavelengths that are emitted by the Earth. passive sensors
But the important RS microwave sensors are all active systems Generation, transmission and recording of the reflected radiation
Page 17 of 56
Question: which spectral bands can be used most effectively in RS?
Figure: Wavelengths that can be used most effectively
depends on interaction with the Earths atmosphere (particles and gases in the atmosphere) obviously absorption (cf. figure above) happens not everywhere and not to the same degree in the spectrum visible area and other high spectral transmission in the atmospheric windows energy level of the sun has its peak in the visible area all passive RS sensor systems have to take these two aspects (transmission and energy) into account. the heat energy emitted by the Earth corresponds to a windows around 10 m (max energy) in the thermal IR
Page 18 of 56
1.4 Interaction of Electromagnetic Radiation with the Atmosphere

As the energy travels from a source (the sun) to the target (the Earths surface) it interacts on its travel with the Earths atmosphere. Interaction with atmosphere
particles and gas molecules
in the
The total amount of radiation that strikes an object is equal to reflected radiation absorbed radiation transmitted radiation incident radiation
reflected off the object
absorbed by the object
transmitted through the object
Two mechanisms of interaction: scattering absorption
Page 19 of 56
1.4.1 Scattering
Particles or the EMR to be
large gas molecules redirected
present in the atmosphere cause
from its original path.
How much scattering takes place depends on the wavelength of the radiation the abundance of the particles or gases the distance the radiation travels through the atmosphere Three types of scattering take place:
a)
Rayleigh scattering occurs when particles are very small compared to the wavelength of the EMR particles: small specks of dust or nitrogen and oxygen molecules shorter wavelengths of energy are much more scattered than longer wavelengths Rayleigh scattering is the mechanism dominant scattering
(0.1 m and less)
in the upper atmosphere
Page 20 of 56
Remarks: Blue sky phenomenon (during the day) ---- stronger scattering of the blue wavelength of the sunlight the blue light seems to reach our eyes from all directions Sunrise and sunset ---- the scattering of the shorter
wavelengths is more complete (longer distance through atmosphere) longer wavelengths (orange, red) penetrate
b) Mie scattering occurs when particles are particles: dust, pollen, smoke Mie scattering tends to affect
just above the same size
as the wavelength of the radiation (industrial or domestic pollution),
water vapour, salt particles from oceanic evaporation longer wavelengths than those
affected by Rayleigh scattering.
Page 21 of 56
occurs mostly in the
lower portions of the atmosphere
where larger particles are more abundant. dominates when cloud conditions are overcast.
(a) and (b) are selective scattering processes i.e. scattering affects specific wavelengths of energy.
c)
Nonselective scattering occurs when the particles are of the radiation (above 10 m) particles: water droplets, ice fragments, large dust particles all wavelengths are scattered about equally causes fog and clouds to appear white to our eyes. All wavelengths are scattered (by the water droplets) in approximately equal quantities. ( white light)
much larger than the wavelength
Page 22 of 56
1.4.2 Absorption
Molecules present in the atmosphere absorb energy at various wavelengths.
The three main constituents (gases) which absorb radiation are ozone carbon dioxide water vapour
Ozone absorbs the harmful UV radiation from the sun. protective layer in the atmosphere avoids skin burn Carbon dioxide the greenhouse gas tends to absorb radiation in the the spectrum thermal heating far infrared portion of
and serves to trap this heat inside the atmosphere.
Water vapour absorbs much of the microwave radiation.
longwave IR and shortwave
The presence of water vapour in the lower
atmosphere greatly varies from location to location and at different times of the year. (Little water vapour above deserts but high humidity in the tropics)
Page 23 of 56
These gases absorb EM energy in very specific regions of the spectrum (called absorption bands ). Those areas which are not severely atmospheric windows. .
influenced by absorption are called
(= areas which are useful for RS purposes). cf. figure in section 1.3 transmission
Page 24 of 56
1.5
Interaction of EMR with Earth surface materials
Electromagnetic energy that is not absorbed or scattered in the atmosphere can reach and interact with the Earths surface.
Three forms of interaction: Absorption Transmission Reflection
The total incident energy will interact with the target in one or more of these three ways. The proportions of each will depend on The wavelength of the EMR The material and condition of the target
Target dependency: There will be a variation of the interaction from time to time during the year. Example: vegetation, from leafing stage to maturity.
Page 25 of 56
A: absorption: T: transmission:
radiation is absorbed into the target radiation passes through a target radiation bounces of the target, is redirected travels upwards through the atmosphere
R: reflection:
reflected energy that part which
(interaction with the atmosphere) enters the field of view of the sensor is
detected and recorded by the sensor most interest in RS is in measuring radiation reflected from targets
Amount and distribution of reflected energy are used in RS to infer the nature of the reflecting surface.
Background: basic assumption made in remote sensing is that specific targets (soils, rocks, vegetation, water, ) have an
individual and characteristic manner of interacting with incident radiation.
Page 26 of 56
spectral response
Distinction between two types of reflection that occur at a surface:
specular reflection
diffuse reflection
incidence angle
reflection angle
mirror-like directed away in a single direction (no scattering)
reflected almost uniformly in all directions (scattered in all directions)
i = r
like a piece of paper
- specular and diffuse reflection represent the two extreme ends of the way in which energy is reflected - most Earth surface features are located somewhere between
perfectly diffuse or perfectly specular
Page 27 of 56
- but: in the visible part of the spectrum terrestrial targets diffuse reflectors calm water specular reflector
diffuse reflector
--
rough smooth
surface surface
specular reflector --
Rough/smooth is defined by
surface variations or particle sizes
that make up the surface in comparison to the wavelength of the incoming radiation
Example: fine-grained sand would appear fairly smooth to microwaves (long wavelength) but quite rough to the visible (short wavelength).
Page 28 of 56
Examples: target interactions
--- leaves/vegetation and water
--- visible and infrared wavelength
Spectral response curve / pattern (sometimes also called the

reflectance % 30
spectral signature
20
vegetation
10
water (2) water (1)
0.4
0.5
0.6
0.7
0.8
(m)
A spectral reflection curve describes the spectral response of a target for a certain region e.g. 0.4 - 2.5 m.
Note: A satellite sensor operating in the visible and NIR region does not observe and detect all reflected energy FOV. To make use of such measurements, the distribution of radiance all possible observation and illumination angles, called the
bi-directional reflectance distribution function (or BRDF) must be taken into consideration
Page 29 of 56
Spectral response curses for about 2000 materials can be found in the JPL ASTER library: e.g. 9 different ones for water/ snow/ ice 1350 minerals etc.
Reminder violet 0.4 Example 1. leaves: lower reflection = higher absorption in B, R higher reflection in G very higher reflection in IR ---- not plotted 2. water: lower reflection = higher absorption in R, NiR darker if viewed in R, NiR higher reflection in B, G water looks blue, green blue 0.46 0.50 green yellow orange 0.58 0.60 0.62 red 0.70
Page 30 of 56
Example Target interaction with leaves
Chlorophyll absorbs radiation in B, R wavelengths but reflects G
Summer Autumn
: chlorophyll content is at its maximum greenest : less chlorophyll
healthy leaves: internal structure of leaves act as all excellent diffuse reflector of NiR wavelengths extremely bright ( but not visible to our eyes )
measure + monitor the near-iR reflectance to determine healthiness of vegetation
Target interaction with water longer wavelength radiation R, NiR is than shorter wavelength water looks blue or blue-green due to stronger reflectance absorbed more by water
of B,G and darker if viewed in R, NiR
Page 31 of 56
If
sediment
is present in the upper layers of the water body.
slight shift to longer wavelengths ( green, yellow ) because of a better reflectivity brighter appearance of water. Chlorophyll (algae) absorbs blue, reflects green water will appear in a more green colour Topography of the water surface (rough, smooth, floating material, etc.) can also lead to complications for water-related interpretations due to problems of specular reflection or other influences on colour or brightness. spectral response can be quite variable, even for the same target type, and can also vary with time and location. important to know where to look spectrally
Water and vegetation completely different in NiR
is similar in the visible area
but
Page 32 of 56
1.6
Energy Sources and Sensing
The sun is the most obvious source of the electromagnetic energy measured in remote sensing. The suns energy is either reflected (visible, reflected iR) or absorbed and re-emitted (thermal iR).
EM energy that is naturally available comes from a passive source. The RS instruments which detect the naturally available energy are called passive sensors.
Passive sensors: Can only be used to detect energy when natural energy is available. reflected energy: -- requires illumination of the Earth (daytime) re-emitted energy: -- can be detected day or night, as long as the amount of energy is large enough.
Passive source: solar energy Visible iR (include thermal) UV, X- and Gamma-ray
Page 33 of 56
Sensing
Recording with passive sensors: Reflected energy is mainly recorded by instruments which travel in sun-synchronous orbits (the satellite
travels southwards over the illuminated side of the Earth and crosses the Equator at the same local Sun time on each orbit). Data are recorded only on the way from North pole to South pole because the other
half of the orbit is in the Earth shadow.
Active sensors provide their The sensor emits radiation
own energy source
for illumination. target.
which is directed toward the
The reflected radiation is detected and recorded by the sensor. Man-made energy source
An active system requires the generation of a fairly large amount of energy to adequately illuminate targets.
Active sensors can be used to examine wavelengths that are not sufficiently provided by the Sun, such as microwaves.
Microwave imaging radar (synthetic aperture radar, SAR) and laser scanner (airborne platform) are examples of active sensors.
Page 34 of 56
1.7
Satellite Images and Colour Display

in channels or bands
Satellite-borne sensors record digital images
which represent reflected radiation of specific wavebands.
Example: SPOT 1, 2 (2 HRV-instruments) 1986, 1990 XS-bands 0.50 0.59 um 0.61 0.68 um 0.79 0.89 um M (= pan) band 0.51 0.73um 1998 1.58 1.75 um (mid iR additionally) SPOT 4
Note: SPOT3, launched in 1993, failed.
Data formats for digital satellite imagery Unfortunately, no world-wide standard for storage and transfer of remotely-sensed data has been agreed upon specific procedures for reading satellite image data requested CEOS (Committee on Earth Observation Satellites) tries to standardise.
Display of satellite images Display of one band: Display of three bands: grey scale image pseudo colour image
Page 35 of 56
Example: (Honolulu data) 6 bands, wavelength between 0.4 and 12 m
Note: different waveband width of the individual bands.
Inspection of three images of Honolulu (cf. the corresponding pictures) Band 2 Band 7 Band 11 0.45 0.52 um 0.76 0.90 um 0.5 14.0 um (blue-green (NiR (thermal iR 0.460.50.58 m) 0.7 1.0/1.3 m) 3 15 m)
Observation: For different regions of Honolulu different brightness levels can be observed in different wavebands value of obtaining multiple images at different wavelength.
Discussion: 1) Region indicated by (a) in band 2 blue wavelength: along the coast line. 2) Region (b) in band 2 and band 7 blue: rain forest appears fairly dark NiR: rain forest (vegetation) appears quite bright. reflective nature of chlorophyll. one can see through the shallow water
Page 36 of 56
3) Region (c) in band 11 Thermal iR: dark patches are clouds, which in the thermal iR are cold. Blue, NiR: at shorter wavelengths the high reflectivity of the water droplets leads to bright patches. 4) Region (d) in band 11 Bright areas in the thermal band that are dark in the other bands. Areas (d) include parts of the airport runways which are facing the sun are warmer than the average scene and so are bright.
Page 37 of 56
Chapter 2 Preprocessing of the Remotely-sensed Data

The raw remotely-sensed image data generally contain flaws or
deficiencies. Removal of flaws and correction of deficiencies are termed preprocessing. Some corrections are carried out at the receiving station.. ground Nevertheless there is often a need on the users
part for some further preprocessing.
Preprocessing may include: Corrections for geometric, radiometric and atmospheric deficiencies Removal of flaws (data errors)
Note: not all operations will be applied in all cases.
2.1 Removal of data error

Defects in the data can be due to errors in the equipment or in the data. transmission or scanning or sampling of image recording
a) Partially or entirely missing scan lines are normally seen as horizontal black (0) and white (255) lines on the images.
Page 38 of 56
b) Horizontal banding pattern electro-mechanical scanners (Landsats MSS and TM) have several (a small number of) detectors that are used in the scanning process. The imbalance in the six (MSS) detectors shows up by strips (banding pattern) in the image.
Missing scan lines and banding patterns can be considered to be a cosmetic defect that interferes with the visual appreciation of the patterns on the image. It might be even more problematic for statistical/pattern analysis of images.
a) Missing scan lines

There is no means of knowing which values should be present at missing scan lines. Solution: Estimate the values by looking at the data values of
the scan lines above and below.
Background: Spatial autocorrelation have similar values. Therefore, neighboring pixels of objects will strongly correlate = points that are close geographically tend to
Page 39 of 56
Processing for replacement of missing scan lines Option 1: Replace a missing pixel value by the value of the corresponding pixel on the immediately preceding scan line. Option 2: Replace the missing value by the average of the neighbouring pixels on the scan lines above and below of a defective line.
I (i, k) = (I (i, k-1) + I (i, k+1))/2 Note: read as ''pixel i on scan line k'' Option 3: Use the neighboring bands of multi-spectral imagery.
For instance, the Landsat (1 to 3) MSS band 4 (green) and 5 (red) are normally highly correlated. In general, bands in the same region of the spectrum are highly correlated and can be used to correct missing scan lines. I ( i, k, b) = b/r*( I( i, k, r) - ( I(i, k+1, r) + I(i, k-1, r) )/2 ) + ( I(i, k+1, b)+ I(i, k-1, b) )/2
k
b, r -- bands
band b
band r
Detection of missing scan lines is a tedious task if such lines are located interactively (by visual
examination)
Page 40 of 56
the auto-correlation property can be used for semi-automatic E.g. by comparison of the average grey values of search along
localization.
neighbouring scan lines. In case of large differences these scan lines for unexpected sequences
of values (strings of either
0's or 255's). Mark the suspected sequence and display it for inspection by a operator.
b) De-striping methods A horizontal banding pattern is sometimes seen on Landsats MSS and TM data. (electro-mechanical scanners). This pattern is more apparent when seen against a areas. dark, low-radiance background such as water
The MSS has six detectors for each band (MSS: 4 spectral
bands) why the banding pattern is known as sixth-line banding in Landsat MSS images. TM has 16 detectors per band and produces seven bands of imagery.
The underlying idea of de-striping is based upon the assumption that each detector sees a categories similar distribution of all the land cover
that are present in the imaged area. In consequence,
the histograms generated for a given band from the pixel values produced by all n detectors should be identical. This implies that
the mean and standard deviation of the data from each detector should be the same. To get rid of the stripping effects the means and standard deviations are calculated from lines 1, 7, 13, 19, (histogram 1),
Page 41 of 56
lines 2, 8, 14, 20, (histogram 2) situation).
and so on
(MSS six detector
All n histograms are equalized by forcing mean and standard deviation to be equal to the corresponding average values of mean and standard deviation of all of the pixels in the image.
2.2 Registration and geometric correction

Registration is the fitting of the coordinate system of one image to that of a second image (or map) of the same area. Geometric correction or rectification is a related technique. An image is transformed so that it has the map. The integration of information extracted from remotely-sensed images with map data into a GIS requires registration. Image scale and projection properties of a
registration, also called rubber-sheeting, is typically defined by a polynomial transformation of an image to a set of control points.
For presentation of RS images in a map-like form rectification (geometric correction) has to be carried out. Rectified images can be overlaid with maps or used to locate features of interest on the map and the image. Rectification may also be used to bring registration adjacent images into
or to overlay images of the same area acquired by
different sensors. Rectification procedures of photogrammetry range from simple plane rectification to the more complex process of generating digital orthophotos.
Page 42 of 56
Sources of geometric errors in digital satellite imagery are instrument errors
- distortions of the optical system, non-linearity of the scanning mechanism and non-uniform sampling rates. panoramic distortion
is a function of the angular field of view and affects instruments with wide AFOV (such as AVHRR) more than those with a narrow AFOV (Landsat MSS + TM, SPOT HRV) Earth rotation During the movement of a satellite southwards above the earths surface the Earth moves eastwards thus the effect of Earth rotation is to skew the image.
scan lines at time t1
Potential scan lines at time t2 without Earth rotation scan lines at time t2 with Earth rotation Satellites ground track
Skew angle at latitude L: = 900- arccos (sinEquator / cos L )
Page 43 of 56
platform instabilities Include variations in altitude and attitude . The information needed to correct for the variations is not generally available (modern enough satellites carry GPS, INS, star sensors, ) or for correction of the image data.
not precise
Therefore a correction band on
nominal orbital parameters must be replaced by a transformation using ground control points. Instead of the attempt to define the sources of error and their effects an alternative method is to look at the problem from the opposite end the differences between the positions of points recorded on image and map can be used to estimate the distortions present in the image. ,
Processing: a) Relate the image and map coordinate system by an empirical transformation. --- commonly polynomials of second or third order are used for locate gcps
map-to-image (image-to-map) coordinate transformation. b) Locate suitable ground control points by using GPS or on the map
and measure its corresponding image coordinates.
Note: gcp chips (19*19 pixels) of existing image maps may also be used. c) Estimate the transformation parameter by least squares and d) Determine the pixel values of the rectified image by resampling (gray scale interpolation)
Page 44 of 56
Third-order polynomial for mapping (x, y) map coordinates to (r, c) image coordinates (and vice versa). X = a00+a10*c+a01*r+a20*c2+a11*c*r+a02*r2+a30*c3+a21*c2*r+a12*c*r2+a03*r3 Y = b00+b10*c+b01*r+b20*c2+b11*c*r+b02*r2+b30*c3+b21*c2*r+b12*c*r2+b03*r3 The unknowns are the parameters aij, bij of the transformation. First order polynomial: Second order polynomial: Third order polynomial: 6 unknowns ( 3 or more gcps) 12 unknowns ( 6 or more gcps) 20 unknowns ( 10 or more gcps)
To have reasonable redundancy significantly more gcps should be used.
Experience: Around (<) 10 gcps give acceptable result for a first-order fit with a small image area of up to 10242 pixels. More gcps will be needed in area of moderate relief (where a secondorder polynomial may be required)
Page 45 of 56
2.3 Atmospheric correction

From Chapter 1 it is already known that a value recorded at a given pixel location is not a recording of the true ground-leaving radiance at
that point. Scattering redirects some of the incoming EM energy and some of the reflected EM energy within the atmosphere into the field of view of the sensor.
sensor
ground
Relationship for estimating atmospheric effects on multi-spectral images in the 0.4 2.4 m reflective solar region: LS = Htotal * * T + LP
(2.1)
Htotal
is the total downwelling radiance in a specific spectral band

Page 46 of 56
is the
reflectance
of the target.
(ratio: downwelling/upwelling or irradiance/radiant emittance) T is the transmittance given by the transmission curves as a
function of the wavelength. LP is the atmospheric path radiance.
The reflectance relates to the the transmittance to the
interaction of EMR to the target,
interaction with the atmosphere.
Model (2.1) is a simplified model which does not explicitly take account of the following aspects: reflectance of a surface will vary with the view angle as well as with the solar illumination angle (particularly important for wide FOV and offnadir viewing) the slope of the ground and the disposition of topographic features.
More complex models are developed but operational use of these models is limited by the need to supply data relating to the condition of
the atmosphere at the time of imaging. The costs of such data-collection activities is considerable, hence reliance is placed upon the use of standard atmospheres such as mid-latitude summer. In this case horizontal visibility
only a very small number of parameters, e.g. the in kilometres have to be supplied.
Page 47 of 56
Atmospheric correction might be beneficial in three cases: 1) If ratios of the values in two bands of a multi-spectral image are computed e.g. the normalised difference vegetation index. NDVi=(NiR-R)/(NiR+R) study vegetation patterns A simple technique for compensation of atmosphere path radiance might be sufficient.
NiR
R Estimate of path radiance
2)
If upwelling radiance from a surface is related to some property of that surface in terms of a physically based model => the atmospheric component must be estimated and removed.
3)
If results found at one time are to be compared with results achieved at a later date => the state of the atmosphere will undoubtedly vary from time 1 to time 2.
Page 48 of 56
2.4 Sensor calibration

Sensor calibration, combined with atmospheric and view angle correction, aims at the estimation of target reflectance. A number of
methodologies for the calibration of the Landsat TM optical bands, SPOT HRV, other optical sensors and Radar are proposed.
The relationship between radiance and pixel value (PV) can be defined for spectral band as Ln* = a0 + a1 *PV Where a0 and a1 are offset and gain coefficients and Ln* is apparent radiance at the sensor. (Measured in units of mW/(m*sr*m))
Spot provides gain values ai in the header of the XS image. The apparent radiance of a given pixel is calculated from L = PV/ ai Given the value of radiance L it is usual to convert to apparent reflectance by = (*L*d2)/(ES*cos(S)) d = relative Earth - Sun distance Es = exoatmospheric solar irradiance s = solar zenith angle
(Reference: Floyd F. Sabins, Remote Sensing: Principles and Interpretation, 3rd edition, W.H.Freeman and company, 1997 Note: is not corrected for atmospheric effects
Page 49 of 56
Chapter 3 Classification
3.1 Supervised and unsupervised classification
Image classification can be decided into two categories: Supervised and unsupervised classification.
Supervised classification refers to the process of measuring characteristic features of the entities (or objects) one wants to classify by using training sets of known objects or object classes and use them to determine the class membership of all other pixels in an image.
Unsupervised classification is a at the determination of the
clustering process
which aims
number of distinct,
naturally occurring
groups and the allocation of pixels to these groups (or classes). In this respect it can be considered as a segmentation technique which
aims at subdividing an image into meaningful regions.
In both cases the properties (features) of the pixel to be classified are used to label that pixel. In the simplest case, a pixel is
characterised by a
vector whose elements are its grey levels
in each spectral band. This feature vector (also called pattern) represents the spectral properties of that pixel. Further features such as
texture or context may be included in the feature vector.
Page 50 of 56
Note: If classification does not lead to proper results: i. try to find a more sophisticated classification scheme not recommended ii. try to find better or more features, e.g. from other sensors (Laser, Radar, in addition to optical sensors) or existing databases (DTMs,...)
3.2 Scatterplot (scattergram) and decision making

A scatterplot is one of the easiest ways to perceive values the distribution of measured on two features. One feature is plotted against the
other for each pixel and the vector (feature 1, feature 2) determines the position of that pixel in the two-dimensional Euclidean space.
Example Pixel Water body Vigorous vegetation

Feature 2 NiR, 255
Feature 1 / R low Low
Feature 2 / NiR low high
0 0 Last printed 20 September 2011 Feature 1 R, 255 Page 51 of 56
The
position of a point in the feature space
is directly
related to the values of the two features. Obviously points belonging to the same class tend to cluster and points belonging to different classes tend to be separated. This is the underlying assumption
of any classification scheme.
Example 2 R Water body Shadow region low low NiR low low MidiR low Any grey value possible
Adding a third feature leads to a three-dimensional scattergram. The problem of N-dimensional feature spaces is that they can not be visualised properly. 3D 1+2 4D 1+2+3 5D 1+2+3+4 3 6 10 two-dimensional scatterplots two-dimensional scatterplots two-dimensional scatterplots use several 2D scatterplots ?
Decision making: Given a scatterplot (cf. figure 3.2 ) one can recognise Two district clusters The compactness of each cluster The distance in feature space (example: d12, d23 ) A linear decision boundary (boundary between two clusters/classes)
Page 52 of 56
Feature 2
Feature 1 Figure 3.2 Decision making
3.3 Supervised classification

Supervised classification methods require external knowledge of the area shown in the image. This knowledge may be derived from fieldwork, analysis of aerial photos, maps or other sources like reports. Most statistical methods assume that the type of the distribution of features in each class is known and only parameters derived from sample data ( training samples ) have to be estimated before using them to
make classification decisions. (Parametric decision making or parametric classification methods.)
Training samples (learning phase) Task: Determine statistical characteristics of each class (the number of classes must be known)
feature value 1 feature value 2
If X is a feature vector:
X =
feature value n Last printed 20 September 2011 Page 53 of 56
Determine: mean X extreme values: min, max, for each feature in each class variances and covariances > variancecovariance matrix
Supervised classification methods

1) Parallelepiped or box classifier 2) all points within min-max region (box) class i all other points are unclassified
K-means or centroid method calculate mean/centre X of each training class calculate Euclidean distance from each unknown pixel
to the centre of each class. The pixel is given the label of the centre to which its distance is smallest. (nearest centre decision)
X2
X3
X1
X3
X4
Page 54 of 56
In n-dimensional feature space the decision boundary corresponds to a hyper plane which is perpendicular to the line connecting the two centroids X 1 and X 2 .
3. Maximum likelihood method If one of two neighbouring clusters is much smaller than the other one, it makes sense to move the boundary between them closer to the
centroid of the smaller one. Similarly, if the clusters are elongated in a certain direction, the boundary should be tilted toward the direction of their elongation. model of probability distribution (Gaussian normal distribution)
variance covariance matrix

X1
X2
equi-probability contours Probability that P belongs to X 2 . P belongs to

X1
is higher than probability that
Page 55 of 56
3.4 Unsupervised classification

Occasionally points will form such distinct clusters that automated
means of discovering which points belong together with be successful. This process is referred to unsupervised learning.
Principles of automatic cluster formation 1. Each point is considered a separate embryonic cluster. In an iterative process points (clusters) are merged together if they are closer
than any other two points. The iteration stops either when the expected number of clusters has been found or when the next points to be added to a clusters is more than some threshold distance away. K-means clustering
2.
Initially the whole collection of points is considered to be one huge cluster. Iteratively, existing clusters are split along lines of
weakness in two clusters. The splitting is repeated until some limits (max number of expected clusters) are passed. Splitting combined with merging to improve the results. ISODATA algorithm (Iterative Self-Organising Data Analysis Technique) can be
Page 56 of 56

Remote Sensing I-2011 2

Transféré par

Informations du document

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Remote Sensing I-2011 2

Transféré par

Droits d'auteur :

Formats disponibles

Module 2, Remote Sensing 1 1st semester Lecture Notes of Prof. Dr. M.

MSc Course PHOTOGRAMMETRY AND GEOINFORMATICS

Module 2 Photogrammetry and Remote Sensing

Topic: Remote Sensing (Part 1)

Last printed 20 September 2011

MSc Course PHOTOGRAMMETRY AND GEOINFORMATICS

2. Preprocessing of remotely-sensed data

Last printed 20 September 2011

MSc Course PHOTOGRAMMETRY AND GEOINFORMATICS

Chapter 1 Basic principles of Remote Sensing

Last printed 20 September 2011

MSc Course PHOTOGRAMMETRY AND GEOINFORMATICS

Elements of an overall Remote Sensing process

Last printed 20 September 2011

MSc Course PHOTOGRAMMETRY AND GEOINFORMATICS

Source: CCRS Tutorial

Last printed 20 September 2011

MSc Course PHOTOGRAMMETRY AND GEOINFORMATICS

Application The extracted information assists to solve a particular problem.

1.2 Electromagnetic radiation

- the rate of transfer of energy from one

Last printed 20 September 2011

MSc Course PHOTOGRAMMETRY AND GEOINFORMATICS

(=electromagnetic) energy Density implies variability over the two-dimensional

surface on which the radiant energy falls.

Radiant flux density is the

magnitude of the radiant flux

== Radiant emittance ( if the energy flow is away from surface)

thermal energy emitted by the Earth

solar energy reflected by the Earth

- is the radiant flux density transmitted from a unit

Last printed 20 September 2011

MSc Course PHOTOGRAMMETRY AND GEOINFORMATICS

Assume: Diffuse reflectance = radiance incident

upon the surface is back scattered in all upward directions

(based on diffuse reflection)

Radiance is measured in watts per square meter per steradiant [ W/(m2*sr) ]

Last printed 20 September 2011

MSc Course PHOTOGRAMMETRY AND GEOINFORMATICS

Reflectance --- is the ratio between the radiant emittance of an object.

are equal and no

occurs at the surface.

practice to convert the into reflectance factors

in order to eliminate the effects of

over the seasons of the year. particular wavebands

All these described quantities refer to

Last printed 20 September 2011

MSc Course PHOTOGRAMMETRY AND GEOINFORMATICS

1.2.2 Nature of electromagnetic radiation (or the view of quantum mechanics)

considers radiation as a wave form (wave-like form of energy)

considers radiation as a stream of particles so-called photons

Importance to remote sensing

Last printed 20 September 2011

MSc Course PHOTOGRAMMETRY AND GEOINFORMATICS

Source: CCRS tutorial

Characterisation of electromagnetic waves

= time elapsed in seconds per cycle

MSc Course PHOTOGRAMMETRY AND GEOINFORMATICS

Source: CCRS tutorial

Last printed 20 September 2011

MSc Course PHOTOGRAMMETRY AND GEOINFORMATICS

These examples demand for a closer look to the EM spectrum.

1.2.4 Corpuscular characteristics of electromagnetic radiation

Energy Q is delivered to a target. Note: delivery is on a probabilistic basis not in such a