Principles of Remote sensing

Chapter 9

Digital Image classification

By: Kinfe W.
 Chapter outlines
• Introduction
 Digital Image classification

• Supervised classification
• popular classifier
• Minimum distance
• Maximum likelihood
• Parallel piped

• Unsupervised classification
• Sequential clustering
• Isodata clustering
Introduction
• Use of mage classification

• Direct uses
• Produce a map of land-use/land-cover

• Indirect use
• Classification is an intermediate step, and may form only one of several data
layers in a GIS
• Water map vs water quality GIS
 Why classify?
Make sense of a landscape
Place landscape into categories (classes)
Forest, Agriculture, Water, etc
Classification scheme = structure of classes
Depends on interest of users

what is a classified image

• Map of the different land features or map of atmosphere

• Categories are defined by the intended use of the map

• Can be few or many categories, depending on the purpose of the map and available resources
Image Classification
 Image Classification
Classification

Supervised Classification Unsupervised Classification

(Clustering)

Distribution Free Statistical Techniques

Euclidean classifier No extensive prior knowledge required
K-nearest neighbour Unknown, but distinct, spectral classes are
generated
Minimum distance
based on probability Limited control over classes and identities
Decision Tree No detailed information
distribution models, which
may be parametric
or nonparametric
 Cont.…
Why Digital image classification is needed?

I. To translate continuous variability of image data in to map pattern that provide

meaningful information to the user.

II. To obtain awareness in the data with respect to ground cover and surface characteristics.

III. Cost efficient in the analysis of large data set.

IV. Effective analysis of complex multi-bands (spectral) interrelationship.

Image Classification

black = water
yellow = open/field
dark green = dense forest
light green = sparse forest
bronze = mixed urban
red = dense urban
 Image classification
• Image classification: is the process by which pixels which have similar spectral characteristics and
which are consequently assumed to belong to the same class are identified and assigned a unique colour.

• Classification is the process of sorting pixels into a finite number of individual classes, or categories of
data based on their data file values.

•It is uses to smooth out small, insignificant variations and simplify an image into a thematic map of
land cover.

• For the first part of the classification process, the computer system must be trained to recognize
patterns in the data.

• Training is the process of defining the criteria by which these patterns are recognized.

9
 Multispectral classification
• Assigning each pixel in a remote sensed image a label describing real world object.
• That is automatically categorizing all pixels in an image in to land cover classes or themes.
• Output is a classified map a form of digital thematic map
• Grouping of similar pixel
• Separation of dissimilar ones
• Assigning class label to pixel
• Resulting in manageable size of classes
 Unsupervised Classification
Unsupervised classification does not use training data as the basis for classification, but
examine unknown pixels in an image and aggregate them into a number of clusters based on
natural groupings present in the image values.
The classification assumes that DN values within a given cover type should be close together
in the spectral space, while data in different classes should be comparatively well separated.
The categories that unsupervised classification identify are not land cover or use classes, but
spectral classes or clusters. Some cover types may have encompass multiple clusters, for
example, agriculture land may have sugar cane, rice, wheat etc. But they look different
spectrally.
The analysts must provide labels to each of the clusters after unsupervised classification using
other sources of data.
 Cont.…
A. Unsupervised classification: it is a technique that groups the pixels into clusters based upon the
distribution of the digital numbers in the image.

• User has controls over

 Maximum number of classes

 Maximum number of iterations

 Threshold value

• The identities of land cover types to be specified as classes within a scene are generally not known a priori
because ground reference information is lacking or surface features within the scene are not well defined.

• The computer is required to group pixel with similar spectral characteristics in to unique clusters according to
some statistically determined criteria

• Analysis then combine and re-labels the spectral clusters in to information classes
Unsupervised classification
 Cont…
• Advantages
• Requires no prior knowledge of the region
• Human error is minimized
• Unique classes are recognized as distinct units

• Disadvantages
• Classes do not necessarily match informational categories of interest
• Limited control of classes and identities
• Spectral properties of classes can change with time
 Isodata clustering
• The iterative self-organizing data analysis technique(ISODATA)
• ISODATA is iterative because it makes a large number of passes through remote sensing dataset
specified results are obtained instead of just two passes
• It dose not allocate it initial mean vector base on the analysis of pixel rather an initial arbitrary
assignment of all Cmax clustering takes place along n-dimensional vector that runs b/n very specific
point in feature space
• maximum no. of clusters to be identify
• maximum % of pixel whose class value are allowed to be unchanged b/n iterations
• max no. of time ISODATA is to classify pixel and recalculate clusters mean vector
• Minimum members in cluster
• Maximum standard deviation for cluster
• Minimum distance b/n cluster means
 Cont`d
• Phase 1: ISODATA cluster building using many passes through the dataset

A B
A. ISODATA initial distribution of five hypothetical mean vectors using + standard deviations in both bands as
beginning and ending points.
B. in the first iteration each candidate pixel is compared to each cluster mean and assigned to the cluster whose
mean is closest in Euclidean distance
 Cont`d
• During the 2nd iteration a new mean calculated for each cluster based on the actual spectral location of pixels
assigned to each cluster instead of the initial arbitrary

• This involves analysis of several parameter to merge or split cluster .

• After the new cluster mean vectors are selected every pixel in the scene is assigned to one of the new
clusters

• This split-merge-assign process continues until there is little change in class assignment b/n iteration (the T
threshold is reached )or the maximum number of iterations is reached(M).
 Example k-means

Band 2

Band 2

Band 2
Band 1 Band 1 Band 1

1. First iteration. The cluster centers 2. Second iteration. The 3. N-th iteration. The centers have
are set at random. Pixels will be centers move to the stabilized.
assigned to the nearest center. mean-center of all pixels
in this cluster.
Sequential Clustering

Y
12
3
4 10
11 13
5 14
2 7
9
15
1 6 8

Sequential Clustering X
B. Supervised classification: initially the operator outlines sample or training areas for each
surface class (from ancillary data or Ground truth).

• Computer then generates statistical parameters from the training areas and compares the
digital numbers of every pixel in the image with these statistical parameters.

• Samples of known identity are use to classify unknown pixels identity

• Multivariate statistical parameters are calculated for the known sample

• Every pixel is evaluated and assigned to the class which it most closely resemble digitally(
in statistics)

• Hard classification – a pixel is assigned to only one class

• The analyst supervises the algorithm by providing training site 20

Supervised Classification

“Training”
Classified Image
Landsat ETM+
Digital color infrared
Acquired: April 21, 2003
Spatial resolution: 30 meters
Landsat TM
Digital color infrared
Acquired: February 17, 1989
Spatial resolution: 30 meters
Landsat MSS
Digital color infrared
Acquired: March 14, 1975
Spatial resolution: 57 meters
Corona
Panchromatic (b/w) film
Acquired: March 2, 1969
Spatial Resolution: 3 meters
Supervised classification
 Cont`d
• Many of the classification tools can also be accessed through the Signature Editor.

• Signature editor: allows to create, manage, evaluate, edit, and classify signatures.

• Threshold dialog: allows you to evaluate the accuracy of the classification process.

• Accuracy assessment dialog: allows to evaluate the accuracy of the classification process.

28
• Classification Steps

1. It is always a good reference to first perform an unsupervised classification.

• This gives a general impression of the classes involved.

2. Define Training Areas

3. Run Supervised Classification

4. Additional Classification by Editing

5. Accuracy Assessment

29
• Supervised Classification Algorithms

i. Minimum Distance to Mean Classifier: every pixel is assigned a class based on its
distance from the mean of each class.

• Also called spectral distance classifier.

ii. Parallelepiped (box) Classifier: the data file values of the candidate pixel are
compared to upper and lower limits. These limits can be either:
 the minimum and maximum data file values of each band in the signature,

 the mean of each band, plus and minus a number of standard deviations, or

 any limits that you specify, based on your knowledge of the data and signatures.

30
 Parallelepiped (supervised)
• For each training region determine the range of values observed in each band.

• These ranges form a spectral box (or parallelepiped) which is used to classify this class type.

• Assign new image pixels to the parallelepiped which it fits into best.

• Pixels outside all boxes can be unclassified or assigned to the closest one.

• Problems with classes that exhibit high correlation between bands. This creates long
‘diagonal’ data-sets that don’t fit well into a box.
 Parallelepiped example

Training classes plotted in spectral space. In this example using 2 bands.

 Parallelepiped example continued
•Each class type defines a spectral box
•Note that some boxes overlap even though the classes are spatially separable.
•This is due to band correlation in some classes.
•Can be overcome by customising boxes.
 Supervised Classification
• Advantages
• Analyst has control over the selected classes tailored to the purpose
• Has specific classes of known identity
• Does not have to match spectral categories on the final map with informational categories
of interest
• Can detect serious errors in classification if training areas are misclassified.
• Disadvantages
• Analyst imposes a classification (may not be natural)
• Training data are usually tied to informational categories and not spectral properties
• Remember diversity
• Training data selected may not be representative
• Selection of training data may be time consuming and expensive
• May not be able to recognize special or unique categories because they are not known or
small
iii. Maximum Likelihood Classifier: assumes a normal distribution for the training areas.

• Also called the Bayesian classifier.

• Probability contours are created around each training area and a pixel assigned to a class
depending upon the value of the probability contours that encompass it.

• The maximum likelihood classifier is generally considered to be the most powerful but is
also considered the most computer intensive.

• Using this algorithm, pixel 1 belongs to class A, pixel 2 to class B and pixel 4 to class C.
Pixel 3 has a higher probability of belonging to class B than class C.

35
 Training samples
• Training samples are sets of pixel that represent what is recognized as a potential class.

• Should be well distributed over the entire scene

• Collected from homogeneous environment.

• Each site is usually composed of many pixels.

• Size: the general rule is if training data is being collected from n bands then 10n pixels of training
data is to be collected from each class but the total sample should be less than 100.

• There are a number of ways to collect training site data

• Using a vector layer

• Defining a polygon in the image

• Using a class from a thematic raster layer from an image file of the same area.
 Select appropriate classification algorithm
• Various supervised classification algorithms may be used to assign an unknown pixel to one of the
classes

• The choice of particular classifier depends on nature of input data and output required

 Parametric

 parametric classification algorithms assume that the observed measurement vector Xc obtained
for each class in each spectral band during the training phase are Gaussian in nature.
Maximum likelihood classifier

 Non parametric

 Classification algorithms make no such assumption

Parallelepiped classifier

Minimum distance to mean classifier

 Maximum likelihood example
• Normal probability distributions are fitted to each training class.
• The lines in the diagram show regions of equal probability.
• Point 1 would be assigned to class ‘pond culture’ as this is most probable.
• Point 2 would generally be unclassified as the probabilities of fitting into one for the classes would be below
threshold.

Equiprobability
contours

2
 Maximum likelihood classifier
• This rule is based on the probability that a pixel belongs to a particular class.

• The probability is equal for all class and the input band have normal distributions.

• If you have a priori knowledge that the probabilities are not equal for all classes, you can
specify weight factors for particular classes
 Cont`d
• Advantage

 Most accurate classifiers (if input is normal distributed)

 Takes variability of classes in to account

• Disadvantage

 An extensive equation takes along time to compute

 Tends to over classify signatures with relatively large value in the covariance matrix
Basic Steps in supervised classification
 Post classification
• Can check non-training regions with more ground truth if available.

• Calculate classification statistics.

• Confusion Matrix: Columns show ground truth, rows show how many pixels are assigned to each
class.
• Overall accuracy: Total correct pixels/total pixels
• Commission errors: Incorrect pixels assigned to a class
• Omission errors: Pixels in class that are assigned a different class

• Visually check to see if any major errors or unwanted features.

Principles of Remote sensing
Chapter 10

Accuracy Assessment

By: Kinfe W.
• Chapter Outline

• Overall accuracy:

• KAPPA(K^)

• Errors matrix

• User and producers accuracy

• Commission errors:

• Omission errors
 Accuracy assessment
• Accuracy assessment is a general term for comparing the classification to
geographical data that are assumed to be true.
• Three types of random pixel selecting can be distinguished:
Random : no rules will be used
Stratified random: the number of points will be stratified to the distribution of
thematic layer classes.
Equalized random: each class will have an equal number of random points.
• From the accuracy assessment cell array, two kinds of reports can be
derived:
The error matrix simply compares the reference points to the classified points in a c*c
matrix, where c is the number of classes.
The accuracy report calculates statistics of the percentages of accuracy, based upon the
results of the error matrix.
45
 Confusion matrix
• A confusion matrix shows the number of correct and incorrect prediction made by the
classification model compare to the actual outcome (target value ) in the data.

• The matrix is estimated from N*N confusion matrix classes.

• E.g. a 2*2 confusion matrix for two classes ( 1&2)

Target
Confusion matrix Class 1 Class 2
Class 1 a b class1 predictive value a/(a+b)
Model
Class 2 c d Class 2 predictive value d/(c+d)

Accuracy= (a+d)/(a+b+c+d)
 Errors matrix
Omission error

 Error of exclusion

 pixel are not assigned to its appropriate class or pixel are omitted from the actual class they
belongs.

 Correspond to non diagonal column elements

 Commission error

 Error of inclusion

 Pixel is assigned to class to which it dose not belong or

 Pixel were improperly included in that category

 Represented by non diagonal row element

 Cont`d
Commission errors (error of inclusion): including an area in a category to which it does not truly
belong.
Omission errors (error of exclusion): excluding an area from a category to which it does truly belong.
Reference data

Grass Agricultura Forest bush Total possible

Classified land l land land land
image
Grass land 25 5 10 3 43 18 out of 43 pixels were
Agricultural 2 50 6 5 63 committed to the grass land
land category
Forest land 3 4 60 5 72
Bush land 2 2 2 100 106
Total 32 61 78 113 284

7 out of 32 were omitted from grass category Grass land category most confused with other land cover
 Cont`d
• Representation of map accuracy:

over all accuracy (of the classification or map) number of correctly classified pixel (sum of the
major diagonal divided by the number of reference pixel.
Grass Agricultura Forest bush Total possible
land l land land land
Grass land 25 5 10 3 43
Agricultural 2 50 6 5 63
land
Forest land 3 4 60 5 72
Bush land 2 2 2 100 106
Total 32 61 78 113 284

OAA = (25+50+60+100)/284*100 = 82.7%

 Cont`d
• Producer`s accuracy (1-omission error)
• Producers accuracy for Grass land category
• PA = divided the number of correctly classified pixel in each category by the number of
reference pixels that were in the category (column total)
• PA = 25/32*100 = 78.1%
• Omission error of class =100-% accuracy of class
Grass Agricultura Forest bush Total possible
land l land land land
Grass land 25 5 10 3 43
Agricultural 2 50 6 5 63
land
Forest land 3 4 60 5 72
Bush land 2 2 2 100 106
Total 32 61 78 113 284
 Cont`d
• Representation of map accuracy:

• User`s accuracy (1-Commission error ): divided the number of correctly classified pixels in each
category by the number of pixels that were classified in that category (row total )

• User`s accuracy for Grass land category

• UA = 25/43*100= 58.1%

Grass Agricultura Forest bush Total possible

land l land land land
Grass land 25 5 10 3 43
Agricultural 2 50 6 5 63
land
Forest land 3 4 60 5 72
Bush land 2 2 2 100 106
Total 32 61 78 113 284
52
 KAPPA (K^)
• Another measure of map accuracy is the Kappa Coefficient which is a measure of the proportional (or
percentage) improvement by the classifier over a purely random assignment to classes.
• Kappa = Shows the difference b/n actual agreement and the agreement expected by chance.
• Kappa of 0.85 means there is 85% better agreement than by chance.
• Kappa = (ND – P)/(𝑵𝟐 – P) or K^ = (observed accuracy – chance agreement)/1-chance agreement
• D – sum of correctly classified class pixels
• P- sum of (product of row total and column total
• N- total number of pixels(reference Pixel)
• Overall accuracy is biased on the correctly classified pixel (the major diagonal ) and excludes the
omission and commission error
• Kappa in addition to the correctly classified pixels, includes the off diagonal elements as a product of
row total and column total
Cont..
cont
Principles of Remote sensing
Chapter 11

Applications of remote sensing

By: Kinfe W.
 Applications of remote sensing
Forestry
1) reconnaissance mapping:
• Forest cover updating, depletion monitoring, and measuring biophysical properties
of forest stands.
• Forest cover type discrimination
• Agroforestry mapping
2) Commercial forestry
• Clear cut mapping / regeneration assessment
• Burn delineation
• Infrastructure mapping / operations support
• Forest inventory
• Biomass estimation
• Species inventory
 Cont`d
3) Environmental monitoring

• Deforestation (rainforest, mangrove colonies)

• Species inventory

• Watershed protection

• Coastal protection

• Forest health
 Land Use/Cover
Land cover refers to the surface cover on the ground, whether vegetation, urban infrastructure,
water, bare soil or other.

• Identifying, delineating and mapping land cover is important for global monitoring studies,
resource management, and planning activities.

Land use refers to the purpose of the land serves, for example, recreation, wildlife habitat, or
agriculture.
 applications of remote sensing on Land use

• Natural resource management

• Wildlife habitat protection

• Baseline mapping for GIS input

• Urban expansion / encroachment

• Routing and logistics planning for seismic / exploration / resource extraction activities

• Damage delineation (wind storm, flooding, volcanic, fire)

• Legal boundaries for tax and property evaluation

• Target detection - identification of landing strips, roads, clearings, bridges, land/water interface
 Agriculture

 Crop type classification

 Crop condition assessment
 Crop yield estimation
 Mapping of soil characteristics
 Mapping of soil management practices
 Compliance monitoring (farming practices)
applications of remote sensing on Geology
• Surficial deposit / bedrock mapping
• Lithological mapping
• Structural mapping
• Sand and gravel (aggregate) survey/ exploitation
• Hydrocarbon exploration
• Geobotany
• Baseline infrastructure
• Sedimentation mapping and monitoring
• Event mapping and monitoring
• Geo-hazard mapping
• Planetary mapping
 hydrological applications of RS
• wetlands mapping and monitoring,
• soil moisture estimation,
• snow pack monitoring / delineation of extent,
• measuring snow thickness,
• determining snow-water equivalent,
• river and lake ice monitoring,
• flood mapping and monitoring,
• River /delta change detection
• Drainage basin mapping and watershed modelling
• Irrigation canal leakage detection
• Irrigation scheduling
 military applications of RS

• maps have been the basic ingredient of all military planning, throughout the
history.

• military strategists use maps to locate opposing forces, plan operations and to
coordinate logistics

• since topography is vital for operations, defense forces prefer topographic maps
over other types of maps
I am will come, If you have question or comment

65