Detection of Pest and Disease in Banana Leaf using Convolution Random Forest

Abstract

In crop production, pest and disease detection is considered as one of the difficult tasks for the farmers. This paper aims to design a real time pest and disease detection system to recognize the pests in early stage by integrating Image processing techniques with Internet of Things (IoT) in banana plant. In this approach, the images are segmented using K-Mean clustering technique that identifies the pests. Subsequently, the category of pest is identified and is classified using convolution random forest. The various features of the pest and disease are used to train the convolution random forest to classify the pest pixel and disease pixels. Based on disease the organic pesticide is suggested using intelligent system of chatbot. The proposed methodology improvises accuracy and it assist the farmers in safeguarding the crop from damage by sending an alert message.

Full text

March- April 2020
ISSN: 0193-4120 Page No. 3727 - 3735
3727
Published by: The Mattingley Publishing Co., Inc.
Detection of Pest and Disease in Banana Leaf using
Convolution Random Forest
T.Sangeetha1,, G.Lavanya2, D.Jeyabharathi3, * T.Rajesh Kumar4,K.Mythili5
1,2,3,5Assitant Professor, Department of Information Technology, Sri Krishna College of Technology, Coimbatore,
India
4Associate Professor, Department of Information Technology, Sri Krishna College of Technology, Coimbatore, India
*rajeshkumarprofessor@gmail.com
Article Info
Volume 83
Page Number: 3727 - 3735
Publication Issue:
March - April 2020
Article History
Article Received: 24 July 2019
Revised: 12 September 2019
Accepted: 15 February 2020
Publication: 23 March 2020
Abstract:
In crop production, pest and disease detection is considered as one of the difficult
tasks for the farmers. This paper aims to design a real time pest and disease
detection system to recognize the pests in early stage by integrating Image
processing techniques with Internet of Things (IoT) in banana plant. In this
approach, the images are segmented using K-Mean clustering technique that
identifies the pests. Subsequently, the category of pest is identified and is classified
using convolution random forest. The various features of the pest and disease are
used to train the convolution random forest to classify the pest pixel and disease
pixels. Based on disease the organic pesticide is suggested using intelligent system
of chatbot. The proposed methodology improvises accuracy and it assist the farmers
in safeguarding the crop from damage by sending an alert message.
Keywords:Pest detection, Internet of Things, Convolution random forest
I INTRODUCTION
The most important source for human
livelihood on earth is crop production. It plays a
major vital role in the country’s economy.
Farmer's economic growth relies upon at the nice
of the products that they produce, which relies on
the plant's boom and the yield they get. One of the
foremost threats to the growth of the crops are the
pests. They affect the healthy yield of crops and
there by minimize the production. It is a matter of
concern to protect these crops as agriculture is
essential part of the country.
Detection of pests in plants plays an
instrumental function. Pest management is tedious
and a hectic process which needs continuous
monitoring of the crops. Manual revealing of pest
need more manpower and is also time consuming.
Hence, it is essential to develop automated
computational methods which will make the
progression of disease detection and classification
easier.
Figure 1. Pest affected leaf samples
Internet of Things (IoT) is a network of
interconnected gadgets that can transfer statistics
efficiently without human involvement. IoT plays
a vital role in increasing the productivity,
obtaining the global market, idea about recent
trends of crops. With the recent advancements in
the technology, it can be used with agriculture to

March- April 2020
ISSN: 0193-4120 Page No. 3727 - 3735
3728
Published by: The Mattingley Publishing Co., Inc.
make the work easier for the farmers; early stage
of recognition the pest is a vital point for crop
management. Improved strategies in safe-guarding
the crops can prevent such loss and damage, can
increase production and make a considerable
effect to food security.
In this paper, we focus on emergence of pest in
plants. This implies to continuous observation of
the plants. Images are acquired using cameras.
Then the acquired images are sent to the
raspberry-pi. Then the image processing
techniques are apply to elucidate the contents of
the image.
II RELATEDWORK
Crop safety is a bought as agriculture due to the
fact there has continually been a want to maintain
the plants free from attack a good way to growth
the yield of the healthy plant life. There are
number of method proposed so far for pest control
in agriculture. In this part of the paper we can
evaluate the distinct varieties of proposed
strategies and methodology presently used for the
early detection of pests and compares their
relative execs and cons.
Paul Boissard et.al [3] described the method of
using static images for the reason of pest
detection. In this method the images are captured
with the assistance of scanner. After image
acquisition, the advance step is to perform image
processing on the acquired image to detect the
pests. This method has good accurate results but
the main disadvantage of this technique is to use
scanner for image acquisition. Also, this technique
is time consuming; it requires time in hours to
generate the results. Since pests do not remain
static, while the images are scanned there is a
possibility for the pests to fly away which leads to
the blurring of the image. Also, there is a chance
of attractive to certain pests. This is a peculiar way
to reduce pests, it will not help in detecting those
pests which cannot fly. The downside of this
method can be overcome by using a pan tilt
camera with whiz. The camera is constantly
moving as it captures the image. Since the digital
camera is flying there is no problem with the pests
in motion and subsequently there are no false
records.
Tapping [4] is a sampling method. This
technique uses a soap solution or oil with water to
gather arthropods at the base and stalk of the rice
when the rice is tapped. After tapping, the
contents in the collecting pan are analyzed
arthropods are identified and counted immediately
in order to find the pest population. The naked-eye
including and data recording can be done on field
but it subjects to human error and leads to high
labor cost and is time consuming.
According to the survey by Saeed Azfar [6]
there are many sensors available for pest detection
namely Acoustic Sensors, Low-power Image
Sensors etc., The low-power image sensor is a
wireless automated monitoring system that is used
for pest detection. Placed in a single trap, the
wireless sensor captures images of the catch
contents from time to time and sends them
remotely to a control station .Sent images are then
used for determination of the number of pests
found at each trap. Based on insect population
number, a farmer can plan when to start with crop
protection and in which field areas.
Johnny L. Miranda et.al [4] proposed an
innovative technique to detect the pests at the
early stage. This concept was very efficient
towards the finding of pests in crops, but it
detects only the whiteflies, a specific pest on
paddy plants. It is not applicable for other type of
pest.
III PROPOSED SYSTEM
Automatic detection of pest and disease is a
very effective way which uses image processing
techniques for the detection of pests from the
crops and using the different properties of the

March- April 2020
ISSN: 0193-4120 Page No. 3727 - 3735
3729
Published by: The Mattingley Publishing Co., Inc.
images the pest is classified
Figure 2. Block diagram of proposed system
Table 1
Rules for cluster formation
Category of
the pest
Rule
Cluster 1(Small)
Cluster 2(Medium)
Cluster
3(large)
White Fly
Rule1
ROI
healthy region
healthy region
Aphids
Rule2
ROI with disease
region
healthy region
healthy region
Beetle
Rule3
ROI with disease
region
healthy region
disease region
Weevils
Rule4
ROI with disease
region
disease region
healthy region
Thrips
Rule5
ROI with disease
region
disease region
healthy region
Caterpillar
Rule6
ROI with disease
region
healthy region
disease region
Moth
Rule7
ROI with disease
region
healthy region
healthy region
using Convolution Random Forest Detection
(CRFD).In this methodology the images of leaves
from the crop fields are taken from the agriculture
field and then transferred to machine using
Raspberry-pi and stored in the database. The
image is preprocessed and segmented into various
clusters stand on the rule. Table 1 demonstrates
the rules to form the clusters. Features are extort
from each cluster which forms trained dataset.
Input image is preprocessed and feature is
extracted. Based on the key image Trained dataset
consist of two stage of classification. In first stage

March- April 2020
ISSN: 0193-4120 Page No. 3727 - 3735
3730
Published by: The Mattingley Publishing Co., Inc.
the input image is classified as healthy and
unhealthy. If the output is healthy there is no pest
in the leaf and no chance to infect the plant. If the
output unhealthy system is to find the category of
pest using the CRFD. With the aid of pest
identification the possibility of the disease to
spread in the banana leaf is identified. Both pest
and disease identification is send to the cultivator
as a SMS message through GSM Module. By
identifying the category of pest, the disease to be
affected to the plant is predicted and a message is
sent to the farmers which help them in taking
appropriate measures and thereby protecting the
crops. The overall methodology is described
through a brief block diagram given in fig 2
The proposed system works on the basis of
some rules framed in table1.Each pest is classed
into three clusters support on the size as small,
medium and large. The cluster1 represent small
size pest up to 1mm, cluster2 represent medium
size from 1mm to 2mm and cluster3 represent
large more than 2.5mm.The sub region of images
are processed by using ROI.
A. Image acquisition
Image acquisition is the foremost step of image
processing. The images are capture using a high
resolution camera with equal illumination to the
object. The captured images are saved in the same
format such as JPEG, TIF, BMP, PNG etc. The
digital camera is interfaced with the Raspberry- pi
which makes use of the captured photo as an enter
to the system.
B. Image pre-processing
Image preprocessing is done to improve the
image facts that contains unwanted in torsion and
to enhance the features of the image for further
processing. Image preprocessing creates an
enhanced photo that is more useful for buying a
clean remark. Whenever the camera captures a
cluster of leaves, background image (excluding
the pestiferous leaf) will be blurred as the
foremost step. Then the image of the pestiferous
leaf will be cropped out. Finally the RGB image
will be converted into gray image for the
identification of pest and disease.
The steps involved in this system are:
1) Conversion of RGB image to gray image
2) Resizing of the image
3) Filtering of the image.
Figure 3. Steps involved in image pre –processing
1)
Conversion of RGB to gray image.
In RGB color model, pixel color is combination
of three colors Red, Green, and Blue (RGB). The
RGB image is a 24-bit color image that supports
around 16,777,216 different colors, whereas a
grey scale image is of 8 bits. The pixel value
ranges from 0 to 244.To find the edges based on
luminance and chrominance, the conversion to
grey scale image is an essential step. The formula
to convert RGB to gray is given in equation (1).
(, ) = 0.2989 + 0.4870 + 0.11401)
The information possessed by gray scale image
is enough for our methodology so we convert
RGB image to gray scale image because the RGB
Conversation of RGB
image to grey image
Resizing of the image
Filtering of the image

March- April 2020
ISSN: 0193-4120 Page No. 3727 - 3735
3731
Published by: The Mattingley Publishing Co., Inc.
image requires more memory space.
2)
Resizing of the image
The image obtained is resized according to the
need of the system. There are numerous methods
available for picture resizing, Nearest-neighbor
interpolation, bilinear, and bicubic. In Nearest-
neighbor interpolation the factor that falls in the
cost of the pixel is assigned to the output pixel. No
different pixels are considered. In bilinear
interpolation the output pixel fee is a weighted
common of pixels in thennearest - by way of-two
neighbourhood. In bicubic interpolation the output
pixel price is a weighted average of pixels in the
nearest 4 by means of four neighborhoods. Hence
we're the use of bicubic interpolation in our
gadget because it generates greater correct results
than any other approach.
3)
Filtering of the image
Filtering in image processing is a system of
removing the unwanted data or noise. It also
allows scrupulous highlighting of particular
records. There are numerous strategies to be had
to clear out the picture and the first-class
alternative depends at the photo and the way it is
used. Both the analog and virtual photo processing
calls for filtering to yield ausable and appealing
cease result. There are extraordinary styles of
filters such as low bypass filters, excessive bypass
filters, suggest filters and many others. In our
system we are using smoothening filter out that is
to reduce noise and improve the visible best of the
photo. Spatial filters are carried out to both static
and dynamic photos, wherein as temporal pics are
implemented best to dynamic snap shots. Here we
use an average clear out, it's miles used for
smoothing the photograph as well as to lessen the
noise within the photo. In this type of filter out
every pixel price is calculated with the suggest
price of its 8 neighborhood pixels.
4)
Image Segmentation
Image segmentation is the procedure of
conversion of digital image into several segments
and furnishes an image into something for easier
analysis. It is used for identifying the objects and
bounding line of that image .we have used K-
means clustering method for segmenting the
image, where the images are partitioned into
clusters in which at least one part of cluster
contain image with major area of diseased part.
The Algorithm step is given below
1. Identify the amount of cluster k
2. Initialize centroids by first reorder the dataset
and then arbitrarily selecting K data points for
the centroids without
alternate.
3. Repeat till there is no alternate to the centroids
i.e. challenge of information factors to clusters
isn’t converting.
• Calculate the sum of the squared distance
between information factors and all centroids.
• Allocate each data point to the closest cluster
(centroid).
• Compute the centroids for the clusters by
taking the common of the all facts points that
belong to every cluster.
A. Feature extraction
Highlight extraction is the procedure where the
ideal component vectors; for example, shading,
surface, morphology and structure are separated.
After division the district of intrigue (Region of
interest) chosen which having better picture
information utilizing highlight extraction systems.
Number of properties, as an example, assessment,
correlation, suggest, eccentricity, general
deviation, homogeneity and so forth are obtained
by using Gray level co-occasion framework
(GLCM) for texture analysis and texture functions
are calculated from statistical distribution of

March- April 2020
ISSN: 0193-4120 Page No. 3727 - 3735
3732
Published by: The Mattingley Publishing Co., Inc.
located depth mixtures at a particular role relative
to others.
B. Convolution random forest for
Classification
It is a supervised algorithm, create a forest with
N number of decision tree by some way and make
it random. Initially the algorithm checks the leaf is
healthy or not. Features are extracted from the
each cluster as ST.
ST=  
  
  
N number of the random sub-tree is created and
the outcome of the each sub-tree is mapped into
convolution matrix. Based on the class label each
output is mapped with the matrix. The output is
converted into n*n matrix based on the number of
label. The Maximum value in the output is used to
identify the pest. Depend on the size of the pest
the disease is identified which is detail described
in the algorithm. The input of the pest and disease
are passed into the intelligent system to know
about the organic pesticide.
   
   
   
   
   
* 




→





→
   
   
   
   
N=Total number of decision tree
C=Total number of class label
O=output of decision tree*class labels
O= N*N (covert the output matrix into n*n based
on number of class labels)
Accuracy=Max (   

 

 )/(Number of
rows*100)
Algorithm
For N=1….do
Label=createtree(s,f)
C<-Group the output into Convolution Matrix
End For
P<-predict(C)
Return the result
Repeat the process for disease identification
Function createtree(s,f)
C<-find the class label based on sample and
feature
return class
End function
Function predict(C)
R<-FIND THE HIGHEST NUMBER OF
LABEL USING CONVOLUTION METHOD
Return R
End Function
IV EXPERIMENT AND RESULT
The main aim of the model is to develop a
system which recognizes banana leaf pest and
disease .Based on that the intelligence system of
Chabot suggest the organic pesticides due to the
TP.CRFD improves the accuracy of the pest
detection and identification when compare to the
classifiers of SVM, Random Forest And Neural
Network. TheFigure.4 and Fig.5 Shows the
detection of the pest and disease.

March- April 2020
ISSN: 0193-4120 Page No. 3727 - 3735
3733
Published by: The Mattingley Publishing Co., Inc.
Table 2
Experimental results
Figure 4.Pest Detection
Figure 5. Pest Detection and Disease
identification
This methodology is applied to various Pest and
we have determined the level of accuracy of the
system in each pest which the represented in the
form of graph in Fig.5
Figure 6. Experimental result of the CRFD
V CONCLUSION AND FUTURE WORK
The proposed methodology is based on image-
processing with IoT and convolution random
forest algorithm for banana plant. The result
presented in this paper is promising. From the
results we inferred that, wider the plant surface
larger is the accuracy of pest detection. The results
obtained are as expected but few improvements
need to be made on both materials and methods in
order to achieve the requirements of fully
automated pest management system, which
involves pest detection, extraction and
identification. In future an automated spraying
Insect
SVM
Rando
m
Forest
Neural
Networ
k
CRF
D
White
Fly
0.759
3
0.7618
0.7601
0.780
1
Aphids
0.812
3
0.8217
0.8299
0.831
1
Beetle
0.829
8
0.8312
0.8423
0.851
1
Weevils
0.839
9
0.8312
3
0.83
0.862
3
Thrips
0.841
1
0.8567
0.8512
0.864
5
Caterpill
ar
0.852
6
0.8589
0.8678
0.873
Moth
0.882
3
0.8791
0.8534
0.879
1

March- April 2020
ISSN: 0193-4120 Page No. 3727 - 3735
3734
Published by: The Mattingley Publishing Co., Inc.
system will be designed and integrated along with
this system that is after the detection of the pest,
the appropriate pesticide will be chosen and
sprayed on the affected part of the plant. The
enhanced algorithm provides the better result
comparing to the remaining existing algorithms.
VI REFERENCES
[1] Muhammad HafeezJaved, M Humair Noor, Babar
YaqoobKhan,Nazish Noor and
TayyabaArshad,”K-means Based Automatic Pests
Detection and Classification for Pesticides
Spraying”,) International Journal of Advanced
Computer Science and Applications, Vol. 8 No.
11, 2017
[2] K. Venugoban and A. Ramanan, “Image
classification of paddy field insect pests using
gradient-based features,” International Journal of
Machine Learning and Computing, vol. 4, no. 1,
p. 1, 2014.
[3] Paul Boissarda, Vincent Martin, “A cognitive
vision approach to early pest detection in
greenhouse crops”. computers and electronics in
agriculture 6 2 (2 0 0 8) 81–93
[4] Johnny L. Miranda, Bobby D. Gerardo, and
Bartolome T. TanguiligIII,”Pest Detection and
Extraction Using Image Processing
Techniques”,International Journal of Computer
and Communication Engineering, Vol. 3, No. 3,
May 2014.
[5] Rupesh G. Mundada1, Dr. V. V.
Gohokar,”Detection and Classification of Pests in
Greenhouse Using Image Processing”,IOSR
Journal of Electronics and Communication
[6] Saeed Azfar, Adnan Nadeem, Abdul Basit “Pest
detection and control techniques using wireless
sensor network: A review” Journal of Entomology
and Zoology Studies 2015; 3 (2): 92-99 E-ISSN:
2320-7078 P-ISSN: 2349-6800(Mar. - Apr. 2015)
[7] Engineering (IOSR-JECE) e-ISSN: 2278-2834,p-
ISSN: 2278-8735.Volume 5, Issue 6 (Mar. - Apr.
2013), PP 57-63 www.iosrjournals.org
[8] Vincent Martin and Sabine Moisan, “Early Pest
Detection in Greenhouses”. INRIA Sophia
Antipolis M´editerrann´ee, PULSAR team 2004
route des Lucioles, BP9.
[9] K. Thenmozhi and U. Srinivasulu Reddy,” Image
Processing Techniques for Insect Shape Detection
in Field Crops “, Proceedings of the International
Conference on Inventive Computing and
Informatics (ICICI 2017).
[10] E. Yousefi, Y.Baleghi, S.M.Sakhaei, “Rotation
invariant wavelet descriptors, a new set of
features to enhance plant leaves classification”,
Computers and Electronics in Agriculture, vol.
140, pp. 70–76, August 2017.
[11] A. Clement, T. Verfaille, C. Lormel, and B.
Jaloux, “A new colour ´ vision system to quantify
automatically foliar discolouration caused by
insect pests feeding on leaf cells,” Biosystems
Engineering, vol. 133, pp. 128–140, 2015.
[12] Pest Control Techniques, Encyclopedia,
2014.http://en.wikipedia.org/w/index.php?oldid=6
14646055 Accessed
[13] Shimoda M, Honda Ki. Insect reaction to light
and its application to pest management,
ApplEntomol and Zool, 2013.
[14] J. Patil and V. Mytri, “A prediction model for
population dynamics of cotton pest
(thripstabacilinde) using multilayer-perceptron
neural network,” International Journal of
Computer Applications, vol. 67, no. 4, 2013.
[15] Farmers Knowledge base, 2014.
http://www.pakissan.com/english/allabout/crop/co
tton. Accessed March 2014
[16] Punithavathani D.S., Sankaranarayanan
K,2009’’IPv4/IPv6 transition mechanisms’
European Journal of Scientific Research , vol.
34,pp.110-124.
[17] JameerBasha A., Palanisamy V., Purusothaman
.T.,2011, Journal of Computer Science, volume
7(5),pp.698706.
[18] Bhuvaneswari K., Rauf H.A.2009, Edgelet based
human detection and tracking by combined
segmentation and soft decision, International
Conference on Control Automation,
Communication and Energy Conservation,
5204487
[19] Rajesh Kumar,T, Geetha, K, RemmiyaDevi,G,
BarkathNisha,S, “Process Efficient Artificial
Neural Network-Based Approach for Channel
Selection and Classification of Seizures”,
Computational Network Application Tools for

March- April 2020
ISSN: 0193-4120 Page No. 3727 - 3735
3735
Published by: The Mattingley Publishing Co., Inc.
Performance Management, Asset Analytics,
doi.org/10.1007/978-981-32-9585-8_12/
[20] Rajesh Kumar Thangavel.,SveadhaAthithan,
Sarumathi, S, Aruna, M, Nithila, B, “Blackspot
Alert and Accident Prevention System” 2019 10th
International Conference on Computing,
Communication and Networking Technologies
(ICCCNT), IEEE Xplore,
10.1109/ICCCNT45670.2019.8944412.
[21] Rajesh kumar,TSamsudeen, S, Sangeetha, S,
Sudha Rani, “ Enhanced Visual Attendance
System by Face Recognition USING K-Nearest
Neighbor Algorithm” Journal of Advanced
Research in Dynamical & Control Systems, Vol.
11, 06-Special Issue, 2019
[22] Rajesh kumar,T., Remmiyadevi,G., Abinaya,K.,
Deepika ,N N., Priyadharshini,S., “An Integrated
Density Based Traffic Load Balancing System in
a Cloud Environment”, Pak. J. Biotechnol. Vol.
14 (4) 623-627 (2017), ISSN Print: 1812-1837,
ISSN Online: 2312-7791
[23] Rajesh kumar, T., Preethi, S., Siva Rubini, R.,
Yamini, V., “SPEED DETECTING AND
REPORTING SYSTEM USING GPS/GPRS
AND GSM” International Journal of Pure and
Applied Mathematics, Volume 118 No. 20 2018,
73-79, ISSN Online: 1314-3395
[24] Rajesh kumar, T.,Janani, A, Karthika, S,
Raghunath, R, Samsudeen, U, Sai Vishnu, S,
Krishna RikachandNapolean,” Automatic Seizure
Detection in EEG signal using PSO and SVM
“International Journal of Pure and Applied
Mathematics, Volume 118 No. 20 2018, 239-245,
ISSN Online: 1314-3395
[25] R.Saranava Ram, M.Vinoth Kumar,
S.Ramamoorthy, B.SaravanaBalaji and Rajesh
Kumar,T.,“An Efficient Hybrid Computing
Environment to Develop a Confidential and
Authenticated IoT Service Model” Wireless
Personal
Communications,doi.org/10.1007/s11277-020-
07056-0, ISSN online: 1572-834X, Print ISSN :
0929-6212
[26] Pathak, A. and Mathew, K.J., 2017. Need for care
to caregivers: Psychological distress and its socio-
demographic correlates among the relatives of
persons with mental illness. International Journal
of Psychosocial Rehabilitation. Vol 21 (2) 3, 12.
[27] Srinath, R. and Sendilvelan, S., 2017. Behavioral
Problems Perceived by the Alcoholic and His
Family A Study among Males in Rural
Areas. International Journal of Psychosocial
Rehabilitation. Vol 21 (2) 13, 19.
[28] Sari, S.P., Dwidiyanti, M., Wijayanti, D.Y. and
Sarjana, W., 2017. Prevalence, demographic,
clinical features and its association of comorbid
depressive symptoms in patients with
schizophrenia. International Journal of
Psychosocial Rehabilitation, 21(2).
[29] Fuadi, L., 2017. Influence of Behavioral
Counseling Techniques, Token Economy and
Parent's Parenting Class of Behaviour Prosocial X
Syamsulhude Tegallinggah. International Journal
of Psychosocial Rehabilitation, 21(2).