DETECTION OF DISEASES IN RICE PLANTS USING
PATTERN RECOGNISATION TECHNIQUES
V.Girish(1210316860) Dept of CSE, GITAM Deemed to be University, Visakhapatnam, India,.
Abstract:
In agricultural field, paddy cultivation plays a
vital role. But their growths are affected by
various diseases. There will be decrease in the
production, if the diseases are not identified at
an early stage. The main goal of this work is to
define a software prototype for rice disease
detection based on the infected images of
various rice plants. Images of the infected rice
plants are captured by digital camera and
processed using image growing, image
segmentation techniques to detect infected
parts of the plants, these images of the
infected part of the leaf has been used for the
classification purpose using neural network.
These methods evolved in the system are both
image processing and soft computing
technique applied on a number of diseased
plants and the results are obtained.
Index Terms -- leaf blast (Magnaporthe grisea),
Fractional Zooming, brown spot (Cochiobolus
Miyabeanus), rice diseases detection, SOM,
HIS model.
INTODUCTION:
India is agriculture-based country that has
many people working in the agriculture
industry. The agricultural sector plays an
important role in economic development by
providing rural employment. Paddy is one of
the nation’s most important products as it is
considered to be one of India’s staple food and
cereal crops and because of that, many efforts
have taken to ensure its safety, one of them is
crop management of paddy plants. Paddy
plants are affected by various fungal and
bacterial diseases. I have restricted this
particular work within the rice diseases only
and considered the most two common
diseases in the North East India, namely Leaf
Blast (Magnaporthe grisea) and Brown Spot
(Cochiobolus Miyabeanus). The proper
detection and recognition of disease is very
important in applying required fertilizer.
Brown Spot Detection is one of the most
common fungal disease and most damaging
paddy plant disease. They are at first little,
roundabout and dim chestnut to purple cocoa.
Completely created sores are round to oval
with a light cocoa to dark focus, encompassed
by a rosy chestnut edge brought on by the
poison delivered by the organisms. LBD is
caused by fungus Magnaporthe oryzae.
Starting indications show up as white to dim
green injuries or spots, with dull green
outskirts. However, the International Rice
Research Institute (IRRI) has developed a
software named Rice Doctor that suggest
name of the diseases by taking the manually
observed information about the infected plant.
In a recent project a prototype system of
automatic identification of cotton insect pest
and intelligent decision based on machine
vision has been developed while remote
sensing technique is being used experimentally
for farm pest management.
The paper has been divided into five sections.
Section II describes the methods of feature
extraction that include region segmentation,
spot detection and boundary detection of
images. Section III narrates the proposed
zooming algorithm used to interpolate points
in the detected spots, which are subsequently
applied for classification of diseases along with
the results illustrated in section IV. Conclusions
are summarized in section V.
 II. FEATURE EXTRACTION
The infected rice leaves samples have been
collected from different parts of India and
mostly from East Midnapur, a major rice
producing district of South Bengal, India. The
images were captured using Nikon COOLPIX P4
digital camera in macro mode. The samples of
the rice leaves affected by Blast and Brown
spots are shown in the Fig.1. This section of
feature extraction has three sub-sections.
A. Segmentation:
These images which are the acquired
through enhancing by increasing
brightness and contrast and then the
images are transformed to Hue Intensity
Saturation (HIS) model for segmentation.
Entropy based bilevel thresholding
method has been invoked for segmenting
the images to facilitate identifying the
infected parts of the leave. The segmented
images of the figure-1 are presented in
Fig.2.
B. Boundary Detection:
Boundary detection algorithm using 8-
connectivy method has been applied
on the segmented images of figure
2(a) and (b), and the respective
outputs are shown in Fig.3.
C. Spot Detection:
Boundaries of the segmented images
are obtained with different image sizes
300×300 to 1500×1500 for different
images. The ratio of the average value
of the green components and intensity
of the pixel lies in between 1.15 to 1.40
respectively and this ratio obtained is
used to identify the infected spots. The
reason for selecting the green
component is because the observation
that when the infection occurs in the
leaf, the greenness of the leaf is
affected most. Instead of considering
only the green value we have taken the
ratio of green component with the
intensity to make the spot detection
invariant of the brightness of the
images and the age of the leave and
also for the reason that all the values
should be normalized before the
pattern recognization. The values have
been determined heuristically based
on the experiments of the hundreds of
infected leave, shown in Fig. 4.
 amongst the spot sizes, the missing
points are often more than the original
points, which may cause a major effect
in the classification process, if the first
method is opted. Thus, interpolation
method has been applied for fractional
zooming to normalize the spots size.
The interpolation algorithm works as
follow:

Step 1.
Determine the scaling factor
sx=100/m and sy=80/n where m × n is
size of the original image.

Step 2.
(i) All points of the original image
located at (i,j) position are placed in
the ((i*sx – sx/2), (j*sy - sy/2)) position
as shown by yellow points, with gray
values a, b, c and d ( figure 5).

(ii) Scan first row wise, then column

wise and calculate the gray value of
intermediate points, like x (shown in
figure 5) by using equation
(a*d1+b*d2)/(d1+d2) where d1 and d2
is the distance from a to x and x to b
respectively.
III. ZOOMING ALGORITHMS
(iii) Scan row wise and any points with
A neural network called Self-
gray value y, which not computed yet
organizing map (SOM) neural network
(figure 5) is calculated as the average
is being used for the classification of
value of its four neighbor. Note that,
images where grey value of the pixels
only the points whose gray values are
of spot images are given as an input to
already computed in the previous
the neural network. The segmented
steps, used to compute the gray value
spots have been normalized to the
of y.
100×80, which is determined by the
average size of the spots, detected in
the previous section to apply it in SOM.
We can make the input size vectors
same by two methods. First one is by
padding of zeros in the input vector
and the second one is by the
interpolation of missing points to
make the input vector of same size.
Because there is a huge difference

classification.
the spot.
the 50% spot.
the 50% rotating spots.
IV. CLASSIFICATION USING
SOM
When the spots are obtained that of
uniform size, unsupervised learning
technique called SOM is used. Here we
have used 100×80 input nodes and 2
output nodes, 300 training patterns
and 50 epochs to train the network.
Here the gray values of the pixels are
applied as the input vector. After
performing training with the above
mention datasets and 50 epochs, we
have tested the system with 300 input
datasets for each of the cases
separately, given below:
Case 1: Use the RGB of the spots for
Case 2: Use the Fourier transform of
Case 3: Use the Arbitrary rotation of
Case 4: Use the Fourier transform of
The result has been shown below
obtained after applying. TABLE I
CLASSIFICATION RESULTS
Category Percentage of
Successful
classification
Case 1 92%
Case 2 82%
Case 3 84%
Case 4 70%
V. CONCLUSION
In the paper, the diseased rice images
are classified using SOM neural
network where train images are
obtained by extracting features of the
infected parts of the leave while four
different types of images are applied
for testing purposes. The zooming
algorithm extracts features of the
images using simple computationally
efficient technique, which results
satisfactory classification for test
images. It has been observed that the
transformation of image in the
frequency domain does not yield a
better classification compare to the
original image.