Transcript Basics:

Basics:
Notation:
Qi  Pixelrow i ,col  j band Q
k
Sum:
i  Q Q Q
i 1
0,
1,
2.
Qk
PARAMETERS
k
MEAN:

Q
i 1
k
* the statistical
average
i
 Q  
k
Sample
Variance:
Standard
Deviation:
s
S
2
 Q 

Q
2

i 1
i
k 1

2
Q
* the central
tendency
2

st
.
d
.
Q
* the spread of the
values about the
mean
Covariance
* measures the tendencies of data file values for the
same pixel, but in different bands, to vary with each
other in relation to the means of their respective
bands.
 Q   R  
k
C
QR

i 1
i
Q
k
i
R

Dimensionality
N = the number of bands = dimensions
…. an (n) dimensional data (feature) space
Measurement
Vector
 v1 
 
v 2 
v 
 3

 
v n 
Mean
Vector
 
 1
 2
 
  3
  
 
  n 
Feature Space - 2dimensions
190
85
Band B
Band A
Spectral Distance
* a number that allows two measurement vectors to be
compared
D
 d i ei 
n
2
i 1
i  a band (dimension)
d  valueof pixeld in band i
e  valueof pixele in band i
i
i
terms
• Parametric = based upon statistical
parameters (mean & standard deviation)
• Non-Parametric = based upon objects
(polygons) in feature space
• Decision Rules = rules for sorting pixels
into classes
Clustering
Minimum Spectral Distance - unsupervised
ISODATA
I - iterative
S - self
O - organizing
D - data
A - analysis
T - technique
A - (application)?
Band B
Band A
Band B
Band A
1st iteration cluster mean
2nd iteration cluster mean
Classification Decision Rules
• If the non-parametric test results in one
unique class, the pixel will be assigned
to that class.
• if the non-parametric test results in zero
classes (outside the decision
boundaries) the the “unclassified rule
applies … either left unclassified or
classified by the parametric rule
• if the pixel falls into more than one
class the overlap rule applies … left
unclassified, use the parametric rule, or
processing order
Non-Parametric
•parallelepiped
•feature space
Unclassified Options
•parametric rule
•unclassified
Overlap Options
•parametric rule
•by order
•unclassified
Parametric
•minimum distance
•Mahalanobis distance
•maximum likelihood
Parallelepiped

Maximum likelihood
(bayesian)
B
•probability
•Bayesian, a prior (weights)
Band B

A
Band A
Minimum Distance
SDxyc 
 
n
i 1
ci
 X xyi

2
c  class
X xyi  value of pixel x, y in i class
Band B
 ci  mean of valuesin i for samplefor class c
Band A
cluster mean
Candidate pixel
GeoStatistics
•Univariate
•Bivariate
•Spatial Description
Univariate
•One Variable
•Frequency (table)
•Histogram (graph)
•Do the same thing (i.e count of observations
in intervals or classes
•Cumulative Frequency (total “below” cutoffs)
Summary of a histogram
• Measurements of location (center
of distribution
n
– mean (m µ x )
– median
– mode

i 1
n
2
st. d. 
• Measurements of shape
(symmetry & length
– coefficient of skewness
– coefficient of variation
i
  1 / n  xi  
n
• Measurements of spread
(variability)
– variance
– standard deviation
– interquartile range
x
CS 
i 1
2
3
1 n
xi   


n i 1

2
2
IQR 
Q Q
3

CV 

1
Bivariate
p 
Scatterplots
Yin
p 
Correlation
  x    y   
n
1
n
p
i 1
i
i
x

x
y
X in
Linear Regression
y
y  ax  b
slope
constant
a

p

y
x
b   y  a  x
Spatial Description
- Data Postings = symbol maps
(if only 2 classes = indicator map
- Contour Maps
- Moving Windows => “heteroscedasticity”
(values in some region are more variable than in others)
- Spatial Continuity
(h-scatterplots
* Xj,Yj
Spatial lag = h = (0,1) = same x, y+1
h=(0,0) h=(0,3) h=(0,5)
tj
hij=tj-ti
* Xi,Yi
correlation coefficient
(i.e the correlogram, relationship of p with h
*
(0,0)
ti
•Correlogram = p(h) = the relationship of the correlation
coefficient of an h-scatterplot and h (the spatial lag)
•Covariance = C(h) = the relationship of thecoefficient of
variation of an h-scatterplot and h
•Semivariogram = variogram =  (h) = moment of inertia
moment of inertia =
1 n

2n i 1
x  y 
i
2
i
OR: half the average sum difference between the x and y pair
of the h-scatterplot
OR: for a h(0,0) all points fall on a line x=y
OR: as |h| points drift away from x=y