HSI Classification: Analysis

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 07 | July -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 1892
HSI Classification: Analysis
1Priya G. Deshmukh, 2Prof V.S.Ublae
1 Electronics Dept. 2Asst. Prof. Electronics Dept.,
AVCOE, Sangamner
---------------------------------------------------------------------***-----------------------------------------------------------------------
Abstract: Spatial features are extracted for the
classification of Hyperspectral Image (HSI) image. For
features extraction Local Binary Pattern is used (LBP),
Gabor filters used to extract global features. Then two level
fusions are applied like feature level & decision fusion. To get
classification output Extreme learning machine (ELM) is
added. With the use of LBP & ELM we get more efficient&
cost effective results.
Key features: Local binary pattern (LBP), Extreme
learning Machine (ELM) and Feature level fusion
I. INTRODUCTION
The main objective is to develop a technique using
technologies of computational intelligence to classify HSI.
To classify HSI here in our method steps are included like
feature extraction, filtration, & classification. In this paper
an LBP, Gabor filter, and ELM are used. LBP is used to
extract local features, to generate encode image. For global
feature extraction, Gabor filter a type of linear filter is used
and then all features including spectral features are
concatenated. A classifier i.e. an ELM is used to classify HSI
image. [4]
II. PROBLEM STATEMENT
Hyperspectral image processing has been a very dynamic
area in remote sensing and other applications in recent
years. Hyperspectral images provide abundant spectral
information to identify and distinguish spectrally similar
materials for more accurate and detailed information
extraction. Wide range of advanced classification
techniques are available based on spectral information and
spatial information. To improve classification accuracy it is
essential to identify and reduce uncertainties in image
processing chain
Large number of high spatial resolution images is available
through various advances of sensor technology. In
conventional HSI classification systems, classifiers only
consider spectral signatures and ignore the spatial
information at neighboring locations. So we focused on
classification of Hyperspectral images using local binary
patterns
III. POPOSED SOLUTION
In our project we use an unsupervised band selection
method using linear prediction error is used to select
distinctive and informative bands. After that local binary
pattern to extract local features then Gabor filter to extract
global features, concatenate all features like local global
including spectral features using feature level fusion and
classifier i.e. Extreme Learning Machine is apply to classify
image. Decision level fusion is used to individual features
along with classifier.
IV. SYSTEM ANALYSIS
Figure 1 Block Diagram of HSI Classification
Figure 2 Flowchart

V. BAND SELECTION
For unsupervised band selection Linear Prediction Error
(LPE) & Principal Component Analysis (PCA) is used.
LPE is a simple efficient band selection method based on
band similarity measurement
Band Selection Algorithm:
1. Assume two initial bands B1 and B2
2. Then for every other band B, an approximation can be
expressed as
(1)
Where a0, a1, a2 are the parameters to minimize the LPE
error
‫׳‬ . (2)
3. The band which produces the maximum error e is
considered as the most dissimilar band to B1 and B2, and
it will be selected.
4. Using these three bands, a fourth band can be found by
using the same strategy and so on. [1]
VI. LBP
By using LBP, texture or feature extraction can be
performed. It includes various applications like surface
inspection, remote sensing and in biomedical area
For m number of neighbors , the LBP code for is
given by
∑ (3)
Where, =1
Figure 3 Example of LBP binary thresholding(a) Center
pixel tc and its eight circular neighbors {ti} 7i=0 with
radius r=1. (b) 3×3 sample block (c) Binary labels of eight
neighbors
Fig.3 shows an example of binary thresholding process of
(m, r) = (8, 1). LBP divide examine window into the
cells(for e.g., 16x16) For each pixel in the cell, compare the
pixels to each of its eight neighbors ;follow the pixel along
the circle clockwise or counter clockwise If center pixel
value greater than neighbor's value write “0”, otherwise
write “1” gives 8 digit binary number. LBP code is
calculated in clockwise direction i.e. 11001010= 83
Figure 4 Implementation of LBP feature extraction
After band selection, the LBP feature extraction process or
Gabor filtering is applied to each selected band image. Fig.
4 illustrates the implementation of LBP feature extraction.
The input image is from the 63th band of the University of
Pavia data. In Fig. 4, the LBP code is first calculated for the
entire image to form an LBP image, and the LBP features
are then generated for the pixel of interest in its
corresponding local LBP image patch. Note that patch size
is a user-defined parameter.[2]
VII. GABOR FILTER
Figure 5 2-D Gabor kernel with different orientations, from
top to bottom, left to
right:[0,π/8,π/4,3π/8,π/2,5π/8,3π/4,7π/8]
Gabor filter is used to extract global features. A linear band
pass filter, with circular symmetric orientation to consider
all directions called as Gabor filter. This is given by, [3]
( ) ( ( ))
Where,

(4)
δ: wavelength of sinusoidal factor
θ: orientation separation angles (π/8, π/4, π/2 etc)
ψ: phase offset
σ: Standard derivation of Gaussian envelope
γ: Spatial aspect ratio.
With ψ=0 & ψ= π/2 return the real & imaginary parts of
the Gabor filter respectively.
√ (5)
Gabor output image of university of Pavia for single band
selection (e.g. Band number 65)
For each selected bands dimensionality (i.e. no of bins) of
LBP feature is m (m-1) +3
Figure 6 LBP and Gabor Output
VIII. ELM
A feed forward neural network with single layer hidden
nodes is actually an ELM which is a type of classifier. It
determines the output weights by randomly assigning
weights to input node. ELM is extremely fast and efficient
classifier compared to SVM. The training samples and
labels are represented as , where xi ∈ Rd and
yi∈RC, the output function of an ELM with L hidden nodes.
This can be expressed as [6]
∑ , (6)
i=1, 2, 3 …n Where
h(): nonlinear function
βj∈ RC:weight vector connecting hidden node to output
wj∈ Rd:weight vector connecting hidden nodes to input
For n equations, (6) can be written as
(7)
Where
Y= [y1; y2……. yn] ∈ Rn×C, β = [β1; β2... βn] ∈ RL×C
[ ] [ ]
(8)
Where H: hidden layer output matrix [6]
IX. FEATURE LEVEL FUSION
Classification is performed by combining all features in
feature level fusion. All LBP, Gabor & spatial features are
arranged compositely. But this may cause to less efficient
classification. [5]
Figure 7 Feature level fusions
X. CLASSIFICATION RESULTS
Here, three aforementioned features, i.e., LBP features
(local texture), Gabor features (global texture), and
selected bands (spectral features), and their combinations,
such as LBP features + Gabor features + spectral features,
LBP features + spectral features, Gabor features + spectral
features, etc., will be discussed.
Table 1. Optimal Band Selection for classification using
ELM
No. of
Selected
Bands
Patch
Size
BW
University of Pavia
LBP 7 21x21 -
Gabor 10 - 5
Indian Pines
LBP 7 17x17 -
Gabor 7 - 1
Salinas

LBP 8 25x25 -
Gabor 8 - 5
The data employed were acquired using National
Aeronautics and Space Administration’s Airborne Visible/
Infrared Imaging Spectrometer (AVIRIS) sensor and was
collected over northwest Indiana’s Indian Pine test site in
June 1992. The image represents a classification scenario
with 145 × 145 pixels and 220 bands in 0.4- to 2.45-μm
region of visible and infrared spectrum with a spatial
resolution of 20 m. The scenario contains two-thirds
agriculture and one-third forest. In this paper, a total of
202 bands are used after removal of water absorption
bands. There are 16 different land-covers classes, but not
all are mutually exclusive in the designated ground truth
map. The number of training and testing samples is shown
in Table 2
Take the Indian Pines data for example, (m, r) is fixed to be
(8, 1). Cross validation strategy is employed for tuning
these parameters. It can be seen that the accuracy tends to
be maximum with 7 or more selected bands and with 17 ×
17 patch size. Note that, for each selected band, the
dimensionality (i.e., number of bins) of the LBP features
are m (m − 1) + 3. Therefore, more selected bands will
increase the dimensionality of the LBP features and
computational complexity. [7]
Table 2. Indian Pines Datasets
Class Train Test
1 Alfalfa 6 48
2 Corn-notill 144 1290
3 Corn-mintill 84 750
4 Corn 24 210
5 Grass-Pasture 50 447
6 Grass trees 75 672
7 Grass pasture mowed 3 23
8 Hay windrowed 49 440
9 Oats 2 18
10 Soybean Notill 97 871
11 Soybean-mintill 247 2221
12 Soybean-Clean 62 552
13 Wheat 22 190
14 Woods 130 1164
15
Build-Grass-Trees-
Drives
38 342
16 Stone-Steel-Towers 10 85
Total 1043 9323
Indian Pines output with no of bands=7 ,No of
samples=m=1:4 (2,4,6,8),Radius r= 1:3 (1,2,3),shown in
table 3.
Table 3. Accuracy of classifier with different m and r values
m
r
1 2 3
2 0.8981 0.8912 9186
4 0.9827 0.9832 0.9832
6 0.9772 0.9823 0.9826
8 0.9777 0.9795 0.9792
Figure 8 Classifier Accuracy
XI. ADVANTAGES & APLLICATIONS
Advantages:
 An entire spectrum is acquired at each point
 The operator needs no prior knowledge of the
sample
 Post-processing allows all available information
from the dataset to be mined
 Utilizes the spatial relationships among the
various spectra in a neighborhood, thus allowing
more elaborate spectral-spatial models for a more
accurate segmentation and classification of the
image
Applications:
 Agriculture: In agriculture for monitoring the
development and health of crops.
 Geology: In geology for rapidly mapping nearly all
minerals of commercial interest
 Military: In the military to provide a unique
standoff detection, identification and imaging
capability for chemical warfare agents
Overall Accuracy of
Indian Pines
S-ELM
G-ELM
L-ELM

 Food processing: In the food processing industry,
Hyperspectral imaging, combined with intelligent
software, enables digital sorters (also called
optical sorters) to identify and remove defects and
foreign material (FM) that are invisible to
traditional camera and laser sorters.
XII. CONCLUSION AND FUTURE SCOPE
Here in our method we include results with variable
number of bands, number of samples for band selection
method using LPE and feature extraction using LBP to
classify Hyperspectral images using classifier ELM. In our
modification we will concentrate on feature selection
method using LTrP. The overall accuracy of Indian pines
datasets is 0.9795. In our future work we will concentrate
on more sophisticated band selection method or
classification method.
REFERENCES
[1]Q. Du and H. Yang, “Similarity-based unsupervised band
selection forhyperspectral image analysis,” IEEE Geosci.
Remote Sens. Lett., vol. 5,no. 4, pp. 564–568, Oct. 2008.
[2 ]T. Ojala, M. Pietikainen, and T. T. Maenpaa,
“Multiresolution gray-scale and rotation invariant texture
classification with local binary pattern,”IEEE Trans.
Pattern Analysis Mach. Intell., vol. 24, no. 7, pp. 971–987,
Jul. 2002.]
[3] C. Chen, W. Li, H. Su, and K. Liu, “spectral–spatial
classification of hyperspectral image based on kernel
extreme learning machine,” RemoteSens., vol. 6, no. 6, pp.
5795–5814, Jun. 2014.
[4] Y. Baziet al., “Differential evolution extreme learning
machine for the classification of hyperspectral images,”
IEEE Geosci. Remote Sens. Lett.,vol. 11, no. 6, pp. 1066–
1070, Jun. 2014.
[5] M. Pal, A. E. Maxwell, and T. A.Warner, “Kernel-based
extreme learning machine for remote sensing image
classification,” Remote Sens. Lett., vol. 9, no. 4, pp. 852–
862, Jun. 2013.
[6] Z. Guo, L. Zhang, and D. Zhang, “Rotation invariant
texture classification using LBP variance (LBPV) with
global matching,” Pattern Recogn., vol. 43, no. 3, pp. 706–
719, Mar. 2010.
[7] R. Moreno, F. Corona, A. Lendasse, M. Grana, and L. S.
Galvao, “Extreme learning machines for soybean
classification in remote sensing hyperspectral images,”
Neuro computing, vol. 128, no. 27, pp. 207–216,Mar. 2014.

HSI Classification: Analysis

More Related Content

What's hot (20)

Similar to HSI Classification: Analysis (20)

More from IRJET Journal (20)

Recently uploaded (20)

HSI Classification: Analysis