CLASSIFICATION OF AVIRIS DATA FOR CROPS
MAPPING USING HYPERSPECTRAL APPROACH
HYPERSPECTRAL REMOTE
SENSING
ECV 5511
PREPARED BY:
FAHMI NASRULLAH
NUR HIDAYAH BT ISHAK HIZAM
SUZALINA BT KAMARUDDIN
PRESENTATION OUTLINE
This presentation consists of:
a) Introduction
b) Material and method
- Study area
- Satellite and ancillary data
- Digital image processing
c) Accuracy assessment
d) Conclusion
INTRODUCTIO
N
INTRODUCTION
Hyperspectral remote sensing is one of technique for
acquiring image but differs from multispectral imaging
Multispectral imaging produce images with few
relatively
broad
wavelength
bands
while
hyperspectral
imaging,
collect
image
data
simultaneously in dozens or hundreds of narrow,
adjacent spectral bands.
Images produced from hyperspectral sensors contain
more data to be interpreted compared to
multispectral sensors.
Example: Multispectral image can be used to map
agriculture area while hyperspectral image can be
used to map agriculture species within the
agriculture area.
INTRODUCTION
In vegetation studies, hyperspectral data can be
used to characterize, model, classify and map
agricultural crops and natural vegetation specifically
in study of species composition, vegetation crop
type, biophysical properties, biochemical properties,
disease and stress, nutrients, moisture and others
( Thenkabail et al.2000)
Classification of surface features in satellite imagery
is one of the most important applications of remote
sensing. There are two types of classification which
unsupervised and supervised methods.
Thenkabail P.S., Smith, R.B., and De-Pauw, E. (2000) Hyperspectral vegetation
indices for determining agricultural crop characteristics. Remote sensing of
Environment. 71:158-182.
INTRODUCTION
However, the main problem with supervised methods is
that the learning process depends heavily on the quality of
the training data set and the input space dimensionality.
While, unsupervised method is not sensitive to the number
of labelled samples since they work on the whole image.
Traditionally, agricultural crops are identified using
broadband satellite imagery by the classification of
satellite imagery with statistical classifiers such as the
Maximum Likelihood (ML) classifier (Philipp and Rath,
2002).
Philipp I, Rath T (2002) Improving plant discrimination in image processing by use of
different colour space transformations. Comput Electron Agric 35(1):115
INTRODUCTION
Recent advances in sensor technology have led to the
development of hyperspectral remote sensing imaging
devices which can obtain high-spectral resolution
radiance data for each location (pixel) within the field
of view (Chen et al. 1999).
It is expected that such detailed spectral data will
permit the unique identification of most surface types
of rocks, soils and vegetation, provided that the spatial
resolution of the data is sufficient to represent a single
surface type for each spectrum.
Chen JM, Leblanc SG, Miller JR, Freemantle J, Loechel SE,Walthall CL, Innanen KA,
White HP (1999) Compact airborne spectrographic imager (CASI) used for mapping
biophysical parameters of boreal forests. J Geophys Res 104:2794527958
MATERIAL &
METHOD
STUDY AREA
The Flevopolder (52 30 N and 5 28 E)
with total area of 970 km2 located in
Flevoland, Netherlands.
SATELLITE DATA
On 5 July 1991 AVIRIS
acquired data over the
Flevoland polder.
AVIRIS records data in
224 spectral bands of
10 nm width.
The flight height is 20
km and the size of the
pixels is approximately
20 m 20 m.
The 224 spectral
bands cover the
spectral range of 400 2400 nm.
Data are available as
unsigned 16-bit data.
ANCILLARY DATA
Agricultural land use in
the Flevopolder was
surveyed during the
AVIRIS overflight and
comprised of potatoes,
sugar beets, wheat,
maize, peas, beans,
flax, onions and grass.
This is the location of
different farms and
agricultural fields in the
Flevopolder where the
numbers refer to the
different farms and farms
owners in 1991.
Beans
Flax
Maize
Onions
Sugar beet
Wheat
Grass
Potato
Clover
ANCILLARY DATA
This is parcel division of different
agricultural fields and the key to the crops.
METHODOLOGY
DATA PREPARATION
PRE PROCESSING
Radiometric correction
Convert DN to
reflectance
DATA CLASSIFICATION
Hyperspectral Analysis
Maximum Likelihood
Select training area
Supervised classification
ACCURACY
ASSESSMENT
Minimum Noise Fraction
Pixel Purity Index
N D Visualization
Endmember Extraction
Spectral Angle Mapper
PRE - PROCESSING
The pre processing involved two main
stages:
a) Radiometric correction
b) Convert DN to reflectance
(a)Radiometric correction
The pre - processing, consisting of dark
current correction, vignetting and
radiometric calibration of the sensors
themselves, is performed by the Jet
Propulsion Laboratory (JPL) in Pasadena,
California.
PRE - PROCESSING
(b)Convert DN to reflectance
The observed digital numbers (DNs) are linearly
related to radiances (in mW/cm2/m/sr) according:
Radiance = DN / 200
Reflectance can be calculated using formula:
Reflectance = offset + gain radiance
where offset and gain for the calibration of AVIRIS
has been given.
DATA CLASSIFICATION
The data classifications were performed using classifications
approaches:
a) Maximum Likelihood
b) Hyperspectral Analysis Approach
Maximum Likelihood
Two steps were carried out in the extraction of crop types:
i) Select training area
ii) Supervised classification
Hyperspectral Analysis Approach
Four steps were carried out in the extraction of crop types:
i) Minimum Noise Fraction
ii) Pixel Purity Index
iii) N D Visualization and Endmember
Extraction
iv) Spectral Angle Mapper
Maximum
Likelihood
Maximum Likelihood
Classifier
Maximum Likelihood classifier was
employed as one of the common widely
used classifier to classify the satellite data.
It served as comparison to hyperspectral
approach examined in this study.
The classification was performed using
supervised approach where all the three
bands are the main feature input using
designated training area.
Hyperspectral
Analysis
(i) Minimum Noise
Fraction
The MNF transform was applied to further
segregate noise thus found in the data
(Green et al. 1988).
MNF Transformation Bands
Eigenvalues of MNF
Transformed Bands versus
band number
(i) Minimum Noise
Fraction
MNF Band 3 2 1
MNF Band 7 4 2
(ii) Pixel Purity Index
Within the feature space, the noise free
data occurred as a continuous class from
the purest to variety of mixtures. The
purest was referred to endmembers while
mixtures were considered as sub
compositions due to pixel attributes.
Pixel purity index
image
Pixel purity index
plot
(ii) Pixel Purity Index
PPI image
enhanced using
density slice
Image showing
pure or extreme
pixel
(iii) n - D Visualization
and Endmember
Extraction
The distribution of the AVIRIS data in n
space was used to estimate the number of
spectral endmembers and their pure
spectral signatures and to help understand
the spectral characteristics of the material
which make up the signature.
Image generated from PPI was used as the
input in this session. Spectral library for
crop types of study area was built using ROI
export from n D visualization.
Different classes generated from n D
Visualization were compared to spectral
(iii) n - D Visualization
and Endmember
Extraction
The n dimensional visualizer and spectral class for each endmembers
(iv) Classification using
Spectral Angle Mapper (SAM)
The SAM is an automated method for
comparing image spectra to individual
spectra or spectral library (Kruse et al.,
1993).
It assumes that the data to be classified
have been reduced to apparent reflectance.
A simplified explanation of this can be
given by considering a reference spectrum
and an unknown spectrum from two - band
data.
RESULTS
The classification result of AVIRIS image using (a) Maximum
Likelihood (b) Spectral Angle Mapper is shown in the figure below:
a)
b)
ACCURACY
ASSESSMENT
ACCURACY ASSESSMENT
a) Maximum Likelihood
Accuracy = 60.57 %
Kappa coefficient = 0.52
Class
Producers Accuracy
(%)
User Accuracy (%)
Water
100
100
Wheat
77.69
100
Onion
76.19
5.23
Sugar beet
Clover
Bean
51.21
79.70
Potato
100
84.19
Grass
66.29
21.22
Peas
Maize
ACCURACY ASSESSMENT
b) Spectral Angle Mapper
Accuracy = 47.59 %
Kappa coefficient = 0.37
Class
Producers Accuracy
(%)
User Accuracy (%)
Water
1.45
100
Wheat
61.26
100
Onion
25
5.23
Maize
Sugar beet
21.70
29.68
Clover
Bean
50.97
48.84
Potato
75.83
84.72
Grass
1.12
0.31
Peas
11.63
3.94
CONCLUSION
THANK
YOU
