CCS355 Neural Network & Deep Learning UNIT III notes and Question bank .pdf
Digital image processing 1
1. Intro of digital image processingIntro of digital image processing
2. Remote Sensing Raster (Matrix) Data FormatRemote Sensing Raster (Matrix) Data FormatRemote Sensing Raster (Matrix) Data FormatRemote Sensing Raster (Matrix) Data Format
Digital number of column 5,
row 4 at band 2 is expressed
as BV5,4,2 = 105.
3. Image file formatsImage file formats
BSQ (Band Sequential Format):BSQ (Band Sequential Format):
each line of the data followed immediately by the next line in the same spectraleach line of the data followed immediately by the next line in the same spectral
band. This format is optimal for spatial (X, Y) access of any part of a singleband. This format is optimal for spatial (X, Y) access of any part of a single
spectral band. Good for multispectral imagesspectral band. Good for multispectral images
BIP (Band Interleaved by Pixel Format):BIP (Band Interleaved by Pixel Format):
the first pixel for all bands in sequential order, followed by the second pixel forthe first pixel for all bands in sequential order, followed by the second pixel for
all bands, followed by the third pixel for all bands, etc., interleaved up to theall bands, followed by the third pixel for all bands, etc., interleaved up to the
number of pixels. This format provides optimum performance for spectral (Z)number of pixels. This format provides optimum performance for spectral (Z)
access of the image data. Good for hyperspectral imagesaccess of the image data. Good for hyperspectral images
BIL (Band Interleaved by Line Format):BIL (Band Interleaved by Line Format):
the first line of the first band followed by the first line of the second band,the first line of the first band followed by the first line of the second band,
followed by the first line of the third band, interleaved up to the number offollowed by the first line of the third band, interleaved up to the number of
bands. Subsequent lines for each band are interleaved in similar fashion. Thisbands. Subsequent lines for each band are interleaved in similar fashion. This
format provides a compromise in performance between spatial and spectralformat provides a compromise in performance between spatial and spectral
processing and is the recommended file format for most ENVI processingprocessing and is the recommended file format for most ENVI processing
tasks. Good for images with 20-60 bandstasks. Good for images with 20-60 bands
5. Band sequential (BSQ) format storesBand sequential (BSQ) format stores
information for the image one band at ainformation for the image one band at a
time. In other words, data for all pixels fortime. In other words, data for all pixels for
band 1 is stored first, then data for allband 1 is stored first, then data for all
pixels for band 2, and so on.pixels for band 2, and so on.
Value=image(c, r, b)
Band interleaved by pixel (BIP) data isBand interleaved by pixel (BIP) data is
similar to BIL data, except that the data forsimilar to BIL data, except that the data for
each pixel is written band by band. Foreach pixel is written band by band. For
example, with the same three-bandexample, with the same three-band
image, the data for bands 1, 2 and 3 areimage, the data for bands 1, 2 and 3 are
written for the first pixel in column 1; thewritten for the first pixel in column 1; the
data for bands 1, 2 and 3 are written fordata for bands 1, 2 and 3 are written for
the first pixel in column 2; and so on.the first pixel in column 2; and so on.
Value=image(b, c, r)
Band interleaved by line (BIL) data storesBand interleaved by line (BIL) data stores
pixel information band by band for eachpixel information band by band for each
line, or row, of the image. For example,line, or row, of the image. For example,
given a three-band image, all three bandsgiven a three-band image, all three bands
of data are written for row 1, all threeof data are written for row 1, all three
bands of data are written for row 2, and sobands of data are written for row 2, and so
on, until the total number of rows in theon, until the total number of rows in the
image is reached.image is reached.
Value=image(c, b, r)
6. What is image processingWhat is image processing
Is enhancing an image or extractingIs enhancing an image or extracting
information or features from an imageinformation or features from an image
Computerized routines for informationComputerized routines for information
extraction (eg, pattern recognition,extraction (eg, pattern recognition,
classification) from remotely sensedclassification) from remotely sensed
images to obtain categories of informationimages to obtain categories of information
about specific features.about specific features.
Many moreMany more
7. Image Processing IncludesImage Processing Includes
Image quality and statistical evaluationImage quality and statistical evaluation
Radiometric correctionRadiometric correction
Geometric correctionGeometric correction
Image enhancement and sharpeningImage enhancement and sharpening
Image classificationImage classification
Pixel basedPixel based
Object-oriented basedObject-oriented based
Accuracy assessment of classificationAccuracy assessment of classification
Post-classification and GISPost-classification and GIS
Change detectionChange detection
GEO5083: Remote Sensing Image Processing and Analysis, spring
8. Image QualityImage Quality
Many remote sensing datasets contain high-quality,Many remote sensing datasets contain high-quality,
accurate data. Unfortunately, sometimes error (oraccurate data. Unfortunately, sometimes error (or
noise) is introduced into the remote sensor data by:noise) is introduced into the remote sensor data by:
the environmentthe environment (e.g., atmospheric scattering,(e.g., atmospheric scattering,
cloud),cloud),
random or systematic malfunctionrandom or systematic malfunction of the remoteof the remote
sensing system (e.g., an uncalibrated detectorsensing system (e.g., an uncalibrated detector
creates striping), orcreates striping), or
improper pre-processingimproper pre-processing of the remote sensorof the remote sensor
data prior to actual data analysis (e.g., inaccuratedata prior to actual data analysis (e.g., inaccurate
analog-to-digital conversion).analog-to-digital conversion).
11. Striping Noise and RemovalStriping Noise and Removal
CPCACPCA
Combined PrincipleCombined Principle
Component AnalysisComponent Analysis
Xie et al. 2004
12. Speckle Noise andSpeckle Noise and
RemovalRemoval
G-MAPG-MAP
Blurred objectsBlurred objects
and boundaryand boundary
Gamma Maximum
A Posteriori Filter
13. Univariate descriptive image statisticsUnivariate descriptive image statistics
TheThe modemode is the value thatis the value that
occurs most frequently in aoccurs most frequently in a
distribution and is usually thedistribution and is usually the
highest point on the curvehighest point on the curve
(histogram). It is common,(histogram). It is common,
however, to encounter more thanhowever, to encounter more than
one mode in a remote sensingone mode in a remote sensing
dataset.dataset.
TheThe medianmedian is the value midwayis the value midway
in the frequency distribution.in the frequency distribution.
One-half of the area below theOne-half of the area below the
distribution curve is to the right ofdistribution curve is to the right of
the median, and one-half is to thethe median, and one-half is to the
leftleft
TheThe meanmean is the arithmeticis the arithmetic
average and is defined as theaverage and is defined as the
sum of all brightness valuesum of all brightness value
observations divided by theobservations divided by the
number of observations.number of observations.
n
BV
n
i
ik
k
∑=
= 1
µ
14. Cont’Cont’
MinMin
MaxMax
VarianceVariance
Standard deviationStandard deviation
Coefficient ofCoefficient of
variation (CV)variation (CV)
SkewnessSkewness
KurtosisKurtosis
MomentMoment
( )
1
var 1
2
−
−
=
∑=
n
BV
n
i
kik
k
µ
kkks var== σ
k
k
CV
µ
σ
=
15.
16.
17. Multivariate Image StatisticsMultivariate Image Statistics
Remote sensing research is often concernedRemote sensing research is often concerned
with the measurement of how much radiant fluxwith the measurement of how much radiant flux
is reflected or emitted from an object in moreis reflected or emitted from an object in more
than one band. It is useful to computethan one band. It is useful to compute
multivariatemultivariate statistical measures such asstatistical measures such as
covariancecovariance andand correlationcorrelation among the severalamong the several
bands to determine how the measurementsbands to determine how the measurements
covary. Variance–covariance and correlationcovary. Variance–covariance and correlation
matrices are used in remote sensingmatrices are used in remote sensing principalprincipal
components analysiscomponents analysis (PCA),(PCA), featurefeature
selectionselection,, classification and accuracyclassification and accuracy
assessmentassessment..
18. CovarianceCovariance
The different remote-sensing-derived spectral measurementsThe different remote-sensing-derived spectral measurements
for each pixel often change together in some predictablefor each pixel often change together in some predictable
fashion. If there is no relationship between the brightnessfashion. If there is no relationship between the brightness
value in one band and that of another for a given pixel, thevalue in one band and that of another for a given pixel, the
values are mutually independent; that is, an increase orvalues are mutually independent; that is, an increase or
decrease in one band’s brightness value is not accompanieddecrease in one band’s brightness value is not accompanied
by a predictable change in another band’s brightness value.by a predictable change in another band’s brightness value.
Because spectral measurements of individual pixels may notBecause spectral measurements of individual pixels may not
be independent, some measure of their mutual interaction isbe independent, some measure of their mutual interaction is
needed. This measure, called theneeded. This measure, called the covariancecovariance, is the joint, is the joint
variation of two variables about their common mean.variation of two variables about their common mean.
( )
n
BVBV
BVBVSP
n
i
n
i
ilikn
i
ilikkl
∑ ∑
∑ = =
=
−×= 1 1
1 1
cov
−
=
n
SPkl
kl
19. CorrelationCorrelation
To estimate the degree of interrelation between variables in a manner not
influenced by measurement units, the correlation coefficient, is
commonly used. The correlation between two bands of remotely sensed
data, rkl, is the ratio of their covariance (covkl) to the product of their
standard deviations (sksl); thus:
To estimate the degree of interrelation between variables in a manner not
influenced by measurement units, the correlation coefficient, is
commonly used. The correlation between two bands of remotely sensed
data, rkl, is the ratio of their covariance (covkl) to the product of their
standard deviations (sksl); thus:
lk
kl
kl
ss
r
cov
=
If we square the correlation coefficient (rkl), we obtain the sample coefficient of
determination (r2
), which expresses the proportion of the total variation in the values of
“band l” that can be accounted for or explained by a linear relationship with the values
of the random variable “band k.” Thus a correlation coefficient (rkl) of 0.70 results in an
r2
value of 0.49, meaning that 49% of the total variation of the values of “band l” in the
sample is accounted for by a linear relationship with values of “band k”.
If we square the correlation coefficient (rkl), we obtain the sample coefficient of
determination (r2
), which expresses the proportion of the total variation in the values of
“band l” that can be accounted for or explained by a linear relationship with the values
of the random variable “band k.” Thus a correlation coefficient (rkl) of 0.70 results in an
r2
value of 0.49, meaning that 49% of the total variation of the values of “band l” in the
sample is accounted for by a linear relationship with values of “band k”.
20. exampleexample
Band 1Band 1 (Band 1 x Band(Band 1 x Band
2)2)
Band 2Band 2
130130 7,4107,410 5757
165165 5,7755,775 3535
100100 2,5002,500 2525
135135 6,7506,750 5050
145145 9,4259,425 6565
675675 31,86031,860 232232
( )( )
135
4
540
cov
5
232675
)860,31(
12
12
==
−=SP
PixelPixel Band 1Band 1
(green)(green)
Band 2Band 2
(red)(red)
Band 3Band 3
(ni)(ni)
Band 4Band 4
(ni)(ni)
(1,1)(1,1) 130130 5757 180180 205205
(1,2)(1,2) 165165 3535 215215 255255
(1,3)(1,3) 100100 2525 135135 195195
(1,4)(1,4) 135135 5050 200200 220220
(1,5)(1,5) 145145 6565 205205 235235
21. Band 1Band 1 Band 2Band 2 Band 3Band 3 Band 4Band 4
Mean (Mean (µµkk)) 135135 46.4046.40 187187 222222
Variance (Variance (varvarkk)) 562.50562.50 264.80264.80 10071007 570570
((sskk)) 23.7123.71 16.2716.27 31.431.4 23.8723.87
((minminkk)) 100100 2525 135135 195195
((maxmaxkk)) 165165 6565 215215 255255
Range (Range (BVBVrr)) 6565 4040 8080 6060
Band 1Band 1 Band 2Band 2 Band 3Band 3 Band 4Band 4
Band 1Band 1 562.2562.2
55
-- -- --
Band 2Band 2 135135 264.8264.8
00
-- --
Band 3Band 3 718.75718.75 275.25275.25 1007.1007.
5050
--
Band 4Band 4 537.50537.50 6464 663.75663.75 570570
Univariate statistics
covariance
BandBand
11
BandBand
22
Band 3Band 3 BandBand
44
Band 1Band 1 -- -- -- --
Band 2Band 2 0.350.35 -- -- --
Band 3Band 3 0.950.95 0.530.53 -- --
Band 4Band 4 0.940.94 0.160.16 0.870.87 --
Covariance Correlation coefficient
22. Types of radiometric correctionTypes of radiometric correction
Detector error or sensor error (internalDetector error or sensor error (internal
error)error)
Atmospheric error (external error)Atmospheric error (external error)
Topographic error (external error)Topographic error (external error)
23. Atmospheric correctionAtmospheric correction
There are several waysThere are several ways
to atmospherically correctto atmospherically correct
remotely sensed data.remotely sensed data.
Some are relativelySome are relatively
straightforward whilestraightforward while
others are complex,others are complex,
being founded onbeing founded on
physical principles andphysical principles and
requiring a significantrequiring a significant
amount of information toamount of information to
function properly. Thisfunction properly. This
discussion will focus ondiscussion will focus on
two major types oftwo major types of
atmospheric correction:atmospheric correction:
Absolute atmosphericAbsolute atmospheric
correctioncorrection, and, and
Relative atmosphericRelative atmospheric
correctioncorrection..
Solar
irradiance
Reflectance from
study area,
Various Paths of
Satellite Received Radiance
Diffuse sky
irradiance
Total radiance
at the sensor
L L
L
Reflectance from
neighboring area,
1
2
3
Remote
sensor
detector
Atmosphere
5
4
1,3,5
θ
θ
E
L
90Þ
θ0
T
θ v
T
0
0
v
p T
S
I
r λr
Ed
Solar
irradiance
Reflectance from
study area,
Various Paths of
Satellite Received Radiance
Diffuse sky
irradiance
Total radiance
at the sensor
L L
L
Reflectance from
neighboring area,
1
2
3
Remote
sensor
detector
Atmosphere
5
4
1,3,5
θ
θ
E
L
90Þ
θ0
T
θ v
T
0
0
v
p T
S
I
λ n
r λr
Ed
60 miles
or
100km
Scattering, Absorption
Refraction, Reflection
24. Absolute atmospheric correctionAbsolute atmospheric correction
Solar radiation is largely unaffected as it travels through theSolar radiation is largely unaffected as it travels through the
vacuum of space. When it interacts with the Earth’s atmosphere,vacuum of space. When it interacts with the Earth’s atmosphere,
however, it is selectivelyhowever, it is selectively scattered and absorbedscattered and absorbed . The sum of. The sum of
these two forms of energy loss is calledthese two forms of energy loss is called atmospheric attenuationatmospheric attenuation..
Atmospheric attenuation may 1) make it difficult to relate hand-heldAtmospheric attenuation may 1) make it difficult to relate hand-held
in situin situ spectroradiometer measurements with remotespectroradiometer measurements with remote
measurements, 2) make it difficult to extend spectral signaturesmeasurements, 2) make it difficult to extend spectral signatures
through space and time, and (3) have an impact on classificationthrough space and time, and (3) have an impact on classification
accuracy within a scene if atmospheric attenuation variesaccuracy within a scene if atmospheric attenuation varies
significantly throughout the image.significantly throughout the image.
The general goal ofThe general goal of absolute radiometric correctionabsolute radiometric correction is to turnis to turn
the digital brightness values (or DN) recorded by a remote sensingthe digital brightness values (or DN) recorded by a remote sensing
system intosystem into scaled surface reflectancescaled surface reflectance values. Thesevalues. These valuesvalues
can then be compared or used in conjunction with scaled surfacecan then be compared or used in conjunction with scaled surface
reflectance values obtained anywhere else on the planet.reflectance values obtained anywhere else on the planet.
25. a) Image containing substantial haze prior to atmospheric correction. b) Image after
atmospheric correction using ATCOR (Courtesy Leica Geosystems and DLR, the
German Aerospace Centre).
a) Image containing substantial haze prior to atmospheric correction. b) Image after
atmospheric correction using ATCOR (Courtesy Leica Geosystems and DLR, the
German Aerospace Centre).
26. relative radiometric correctionrelative radiometric correction
When required data is not available forWhen required data is not available for
absolute radiometric correction, we canabsolute radiometric correction, we can
do relative radiometric correctiondo relative radiometric correction
Relative radiometric correction may beRelative radiometric correction may be
used toused to
Single-image normalization using histogramSingle-image normalization using histogram
adjustmentadjustment
Multiple-data image normalization usingMultiple-data image normalization using
regressionregression
27. Single-image normalization usingSingle-image normalization using
histogram adjustmenthistogram adjustment
The method is based on the fact that infraredThe method is based on the fact that infrared
data (>0.7data (>0.7 µµm) is free of atmosphericm) is free of atmospheric
scattering effects, whereas the visible regionscattering effects, whereas the visible region
(0.4-0.7(0.4-0.7 µµm) is strongly influenced by them.m) is strongly influenced by them.
UseUse Dark SubtractDark Subtract to apply atmosphericto apply atmospheric
scattering corrections to the image data. Thescattering corrections to the image data. The
digital number to subtract from each band candigital number to subtract from each band can
be either thebe either the band minimum, an averageband minimum, an average
based upon a user defined region of interest,based upon a user defined region of interest,
oror a specific valuea specific value
29. Topographic correctionTopographic correction
Topographic slope and aspect also introduceTopographic slope and aspect also introduce
radiometric distortion (for example, areas inradiometric distortion (for example, areas in
shadow)shadow)
The goal of a slope-aspect correction is toThe goal of a slope-aspect correction is to
remove topographically induced illuminationremove topographically induced illumination
variation so that two objects having the samevariation so that two objects having the same
reflectance properties show the samereflectance properties show the same
brightness value (or DN) in the image despitebrightness value (or DN) in the image despite
their different orientation to the Sun’s positiontheir different orientation to the Sun’s position
Based on DEM, sun-elevationBased on DEM, sun-elevation
30. Conceptions of geometric correctionConceptions of geometric correction
Geocoding:Geocoding: geographical referencinggeographical referencing
Registration:Registration: geographically or nongeographically (no coordination system)geographically or nongeographically (no coordination system)
Image to Map (or Ground Geocorrection)Image to Map (or Ground Geocorrection)
The correction of digital images to ground coordinates using ground controlThe correction of digital images to ground coordinates using ground control
points collected from maps (Topographic map, DLG) or ground GPS points.points collected from maps (Topographic map, DLG) or ground GPS points.
Image to Image GeocorrectionImage to Image Geocorrection
Image to Image correction involves matching the coordinate systems or columnImage to Image correction involves matching the coordinate systems or column
and row systems of two digital images with one image acting as a referenceand row systems of two digital images with one image acting as a reference
image and the other as the image to be rectified.image and the other as the image to be rectified.
Spatial interpolation:Spatial interpolation: from input position to output position or coordinates.from input position to output position or coordinates.
RST (rotation, scale, and transformation), Polynomial, TriangulationRST (rotation, scale, and transformation), Polynomial, Triangulation
Root Mean Square Error (RMS):Root Mean Square Error (RMS): The RMS is the error term used toThe RMS is the error term used to
determine the accuracy of the transformation from one system to another. It isdetermine the accuracy of the transformation from one system to another. It is
the difference between the desired output coordinate for a GCP and the actual.the difference between the desired output coordinate for a GCP and the actual.
Intensity (or pixel value) interpolation (also called resampling):Intensity (or pixel value) interpolation (also called resampling): The process ofThe process of
extrapolating data values to a new grid, and is the step in rectifying an image thatextrapolating data values to a new grid, and is the step in rectifying an image that
calculates pixel values for the rectified grid from the original data grid.calculates pixel values for the rectified grid from the original data grid.
Nearest neighbor, Bilinear, CubicNearest neighbor, Bilinear, Cubic
32. Purposes of image classificationPurposes of image classification
Land use and land cover (LULC)Land use and land cover (LULC)
Vegetation typesVegetation types
Geologic terrainsGeologic terrains
Mineral explorationMineral exploration
Alteration mappingAlteration mapping
…………..
33. What is image classificationWhat is image classification
oror
pattern recognitionpattern recognition
Is a process of classifying multispectral (hyperspectral) images intoIs a process of classifying multispectral (hyperspectral) images into
patterns of varying gray or assigned colorspatterns of varying gray or assigned colors that represent eitherthat represent either
clustersclusters of statistically different sets of multiband data, some of which canof statistically different sets of multiband data, some of which can
be correlated with separable classes/features/materials. This is the resultbe correlated with separable classes/features/materials. This is the result
ofof Unsupervised ClassificationUnsupervised Classification, or, or
numerical discriminatorsnumerical discriminators composed of these sets of data that have beencomposed of these sets of data that have been
grouped and specified by associating each with a particulargrouped and specified by associating each with a particular classclass, etc., etc.
whose identity is known independently and which has representativewhose identity is known independently and which has representative
areas (training sites) within the image where that class is located. This isareas (training sites) within the image where that class is located. This is
the result ofthe result of Supervised ClassificationSupervised Classification..
Spectral classesSpectral classes are those that are inherent in the remote sensorare those that are inherent in the remote sensor
data and must be identified and then labeled by the analyst.data and must be identified and then labeled by the analyst.
Information classesInformation classes are those that human beings define.are those that human beings define.
34. supervised classification. Identify known a priori
through a combination of fieldwork, map
analysis, and personal experience as training
sites; the spectral characteristics of these sites are
used to train the classification algorithm for
eventual land-cover mapping of the remainder of
the image. Every pixel both within and outside
the training sites is then evaluated and assigned to
the class of which it has the highest likelihood of
being a member.
unsupervised classification, The
computer or algorithm automatically
group pixels with similar spectral
characteristics (means, standard
deviations, covariance matrices,
correlation matrices, etc.) into unique
clusters according to some statistically
determined criteria. The analyst then
re-labels and combines the spectral
clusters into information classes.
35. Hard vs. Fuzzy classificationHard vs. Fuzzy classification
SupervisedSupervised andand unsupervisedunsupervised classificationclassification
algorithms typically usealgorithms typically use hard classificationhard classification logiclogic
to produce a classification map that consists ofto produce a classification map that consists of
hard, discrete categories (e.g., forest,hard, discrete categories (e.g., forest,
agriculture).agriculture).
Conversely, it is also possible to useConversely, it is also possible to use fuzzy setfuzzy set
classificationclassification logiclogic, which takes into account the, which takes into account the
heterogeneous and imprecise nature (mixheterogeneous and imprecise nature (mix
pixels) of the real world. Proportion of the mpixels) of the real world. Proportion of the m
classes within a pixel (e.g., 10% bare soil, 10%classes within a pixel (e.g., 10% bare soil, 10%
shrub, 80% forest). Fuzzy classificationshrub, 80% forest). Fuzzy classification
schemes are not currently standardized.schemes are not currently standardized.
36.
37. Pixel-based vs. Object-orientedPixel-based vs. Object-oriented
classificationclassification
In the past, most digital image classification was based onIn the past, most digital image classification was based on
processing the entire scene pixel by pixel. This is commonlyprocessing the entire scene pixel by pixel. This is commonly
referred to asreferred to as per-pixel (pixel-based) classificationper-pixel (pixel-based) classification ..
Object-oriented classificationObject-oriented classification techniques allow thetechniques allow the
analyst to decompose the scene into many relativelyanalyst to decompose the scene into many relatively
homogenous imagehomogenous image objectsobjects (referred to as(referred to as patches orpatches or
segmentssegments) using a multi-resolution image segmentation) using a multi-resolution image segmentation
process. The various statistical characteristics of theseprocess. The various statistical characteristics of these
homogeneous image objects in the scene are then subjectedhomogeneous image objects in the scene are then subjected
to traditional statistical or fuzzy logic classification. Object-to traditional statistical or fuzzy logic classification. Object-
oriented classification based on image segmentation is oftenoriented classification based on image segmentation is often
used for the analysis of high-spatial-resolution imagery (e.g.,used for the analysis of high-spatial-resolution imagery (e.g.,
1 1 ×× 1 m Space Imaging IKONOS and 0.61 1 m Space Imaging IKONOS and 0.61 ×× 0.61 m Digital 0.61 m Digital
Globe QuickBird).Globe QuickBird).
38. Unsupervised classificationUnsupervised classification
UsesUses statistical techniquesstatistical techniques to group n-dimensional data into their naturalto group n-dimensional data into their natural
spectral clusters, and uses thespectral clusters, and uses the iterative proceduresiterative procedures
label certain clusters as specific information classeslabel certain clusters as specific information classes
K-mean and ISODATAK-mean and ISODATA
For the first iteration arbitraryFor the first iteration arbitrary starting valuesstarting values (i.e., the cluster properties) have(i.e., the cluster properties) have
to be selected. Theseto be selected. These initial valuesinitial values can influence the outcome of thecan influence the outcome of the
classification.classification.
In general, both methods assign first arbitrary initial cluster values. TheIn general, both methods assign first arbitrary initial cluster values. The
second step classifies each pixel to the closest cluster. In the third step thesecond step classifies each pixel to the closest cluster. In the third step the
new cluster mean vectors are calculated based on all the pixels in onenew cluster mean vectors are calculated based on all the pixels in one
cluster. The second and third steps are repeated until the "change" betweencluster. The second and third steps are repeated until the "change" between
the iteration is small. The "change" can be defined in several different ways,the iteration is small. The "change" can be defined in several different ways,
either by measuring the distances of the mean cluster vector have changedeither by measuring the distances of the mean cluster vector have changed
from one iteration to another or by the percentage of pixels that havefrom one iteration to another or by the percentage of pixels that have
changed between iterations.changed between iterations.
TheThe ISODATA algorithm has some further refinementsISODATA algorithm has some further refinements by splitting andby splitting and
merging of clusters. Clusters are merged if either the number of membersmerging of clusters. Clusters are merged if either the number of members
(pixel) in a cluster is less than a certain threshold or if the centers of two(pixel) in a cluster is less than a certain threshold or if the centers of two
clusters are closer than a certain threshold. Clusters are split into twoclusters are closer than a certain threshold. Clusters are split into two
different clusters if the cluster standard deviation exceeds a predefined valuedifferent clusters if the cluster standard deviation exceeds a predefined value
and the number of members (pixels) is twice the threshold for the minimumand the number of members (pixels) is twice the threshold for the minimum
number of members.number of members.
39.
40. Supervised classification:Supervised classification:
training sites selectiontraining sites selection
Based on known a priori through a combination of fieldwork,Based on known a priori through a combination of fieldwork,
map analysis, and personal experiencemap analysis, and personal experience
on-screen selectionon-screen selection of polygonal training data (ROI),of polygonal training data (ROI), and/orand/or
on-screen seedingon-screen seeding of training data (ENVI does not haveof training data (ENVI does not have
this, Erdas Imagine does).this, Erdas Imagine does).
TheThe seedseed programprogram begins at a singlebegins at a single x, yx, y location and evaluateslocation and evaluates
neighboring pixel values in all bands of interest. Using criterianeighboring pixel values in all bands of interest. Using criteria
specified by the analyst, the seed algorithm expands outward likespecified by the analyst, the seed algorithm expands outward like
an amoeba as long as it finds pixels with spectral characteristicsan amoeba as long as it finds pixels with spectral characteristics
similar to the original seed pixel. This is a very effective way ofsimilar to the original seed pixel. This is a very effective way of
collecting homogeneous training information.collecting homogeneous training information.
FromFrom spectral libraryspectral library of field measurementsof field measurements
42. Supervised classification methodsSupervised classification methods
Various supervised classification algorithms may be used to assign an unknown pixel to oneVarious supervised classification algorithms may be used to assign an unknown pixel to one
ofof mm possible classes. The choice of a particular classifier or decision rule depends on thepossible classes. The choice of a particular classifier or decision rule depends on the
nature of the input data and the desired output.nature of the input data and the desired output. ParametricParametric classification algorithmsclassification algorithms
assumes that the observed measurement vectorsassumes that the observed measurement vectors XXcc obtained for each class in each spectralobtained for each class in each spectral
band during the training phase of the supervised classification areband during the training phase of the supervised classification are GaussianGaussian; that is, they are; that is, they are
normally distributed.normally distributed. NonparametricNonparametric classification algorithms make no such assumption.classification algorithms make no such assumption.
Several widely adopted nonparametric classification algorithms include:Several widely adopted nonparametric classification algorithms include:
one-dimensionalone-dimensional density slicingdensity slicing
parallepipedparallepiped,,
minimum distanceminimum distance,,
nearest-neighbornearest-neighbor, and, and
neural networkneural network andand expert system analysisexpert system analysis..
The most widely adopted parametric classification algorithms is the:The most widely adopted parametric classification algorithms is the:
maximum likelihoodmaximum likelihood..
Hyperspectral classification methodsHyperspectral classification methods
Binary EncodingBinary Encoding
Spectral Angle MapperSpectral Angle Mapper
Matched FilteringMatched Filtering
Spectral Feature FittingSpectral Feature Fitting
Linear Spectral UnmixingLinear Spectral Unmixing
44. Accuracy assessment ofAccuracy assessment of
classificationclassification
Remote sensing-derived thematic information areRemote sensing-derived thematic information are
becoming increasingly important. Unfortunately, theybecoming increasingly important. Unfortunately, they
contain errors.contain errors.
Errors come from 5 sources:Errors come from 5 sources:
Geometric error still thereGeometric error still there
None of atmospheric correction is perfectNone of atmospheric correction is perfect
Clusters incorrectly labeled after unsupervised classificationClusters incorrectly labeled after unsupervised classification
Training sites incorrectly labeled before supervisedTraining sites incorrectly labeled before supervised
classificationclassification
None of classification method is perfectNone of classification method is perfect
We should identify the sources of the error, minimize it,We should identify the sources of the error, minimize it,
do accuracy assessment, create metadata before beingdo accuracy assessment, create metadata before being
used in scientific investigations and policy decisions.used in scientific investigations and policy decisions.
We usually need GIS layers to assist our classification.We usually need GIS layers to assist our classification.
47. Change detectionChange detection
Change detect involves the use of multi-temporal datasets toChange detect involves the use of multi-temporal datasets to
discriminate areas of land cover change between dates of imaging.discriminate areas of land cover change between dates of imaging.
Ideally, it requiresIdeally, it requires
Same or similar sensor, resolution, viewing geometry, spectral bands,Same or similar sensor, resolution, viewing geometry, spectral bands,
radiomatric resolution, acquisition time of data, and anniversary datesradiomatric resolution, acquisition time of data, and anniversary dates
Accurate spatial registration (less than 0.5 pixel error)Accurate spatial registration (less than 0.5 pixel error)
MethodsMethods
Independently classified and registered, then compare themIndependently classified and registered, then compare them
Classification of combined multi-temporal datasets,Classification of combined multi-temporal datasets,
Principal components analysis of combined multi-temporal datasetsPrincipal components analysis of combined multi-temporal datasets
Image differencing (subtracting), (needs to find change/no change threshold,Image differencing (subtracting), (needs to find change/no change threshold,
change area will be in the tails of the histogram distribution)change area will be in the tails of the histogram distribution)
Image ratioing (dividing), (needs to find change/no change threshold, changeImage ratioing (dividing), (needs to find change/no change threshold, change
area will be in the tails of the histogram distribution)area will be in the tails of the histogram distribution)
Change vector analysisChange vector analysis
Delta transformationDelta transformation