See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/271305520

Pest Detection and Extraction Using Image Processing Techniques

Article · January 2014

DOI: 10.7763/IJCCE.2014.V3.317

CITATIONS READS

9 938

3 authors:

Johnny L. Miranda Bobby Dioquino Gerardo

Pampanga State Agricultural University West Visayas State University
1 PUBLICATION   9 CITATIONS    73 PUBLICATIONS   133 CITATIONS   

SEE PROFILE SEE PROFILE

Bartolome T. Tanguilig Iii

AMA Education System, Philippines
23 PUBLICATIONS   40 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Association Rule Mining Project View project

Telematics and Data Mining View project

All content following this page was uploaded by Bartolome T. Tanguilig Iii on 27 May 2016.

The user has requested enhancement of the downloaded file.

 International Journal of Computer and Communication Engineering, Vol. 3, No. 3, May 2014

Pest Detection and Extraction Using Image Processing

Techniques
Johnny L. Miranda, Bobby D. Gerardo, and Bartolome T. Tanguilig III

 that can lead to poor decisions on rice pest management.

Abstract—Detection of pests in the paddy fields is a major Due to the rapid development of digital technology, there
challenge in the field of agriculture, therefore effective is an opportunity for image processing technology to be used
measures should be developed to fight the infestation while in the field of agricultural research which could help the
minimizing the use of pesticides. The techniques of image
researcher to solve a complex problem. Image analysis
analysis are extensively applied to agricultural science, and it
provides maximum protection to crops, which can ultimately provides a realistic opportunity for the automation of insect
lead to better crop management and production. Monitoring of pest detection. This study extends the implementation of
pests infestation relies on manpower, however automatic image processing techniques to estimate pest densities in rice
monitoring has been advancing in order to minimize human fields by establishing an automated detection system.
efforts and errors. This study extends the implementation of Through this system, crop technicians can easily count the
different image processing techniques to detect and extract
pests from the collected specimens, and right pests’
insect pests by establishing an automated detection and
extraction system for estimating pest densities in paddy fields. management can be applied to increase both the quantity and
Experiment results shows that the proposed system provides a quality of rice production. Using the automated system, crop
simple, efficient and fast solution in detecting pests in the rice technicians can make the monitoring process easier. Rice
fields. infestation may be easily detected and monitored with the use
of a camera.
Index Terms—Automated pests management, image analysis
object detection, object extraction.
II. REVIEW OF RELATED LITERATURES AND STUDIES
I. INTRODUCTION A. Current Methods Used by the Crop Technicians in
Rice is the most important and a primary source of food in Sampling Insect Pests in the Paddy Fields
Asia especially in the Philippines. However, rice may lose its Currently, the population of insect pests and the damage on
quantity and quality when rice is attacked by different insect rice are generally randomly distributed that later aggregate in
pests. Therefore, it is a top priority to find effective methods clumps [2]. That is why the sampling method was used by the
to reduce the level of their infestation in the paddy fields. In farmers to achieve this goal because they used this to collect,
agriculture, pest control has always been considered as the count or inspect a small part of the pest population. It actually
most challenging task for farmers [1]. Most of the farmers determines the trends in population of organism. Sampling
used the traditional pest management methods which is the method has two fundamental reasons; the research and the
regular spray program based on schedules rather than the pest management decision-making, which normally requires
presence of insect pests on the paddy fields. These chemicals a considerable time, efforts and costs as well. There are no
kill useful insects which eradicate pests in crops. universal sampling methods but this is carried out by using a
Assessing the density of the rice pest population in paddy variety of techniques and devices depending on the objective
fields is very important for pest forecasting decisions. Sticky of the work. One of this is the effective use of pest
traps are widely used to trap the insect pests. The trapped surveillance and monitoring system which may result to
insects are brought to the laboratory for counting and identify efficient timing of interventions and reduce cost of
manually. Usually, crop technicians identify and segregate production.
the insects manually according to their species and count the In the study of Carino, Kenmore and Dyck [3] there are
major pests separately. The resulting counts are used to several sampling techniques and devices for pest
estimate the pest density in the paddy fields. However, management decision-making; the light trap, that involves
multiple site and frequent counting of rice pests is time varying size sample which is good for comparing seasonal
consuming and tedious for a crop technician. This can lead to and yearly catches of insects, but catches are subject to
low count accuracy and delays in obtaining accurate counts changes in insect behavior and do not catch none flying
insects; the sweep net (catching insect using fishnet), is a fast
Manuscript received January 25, 2014; revised March 13, 2014.
method, very economical, and good for sampling arthropods
Johnny Miranda is with the College of Resource Engineering and staying in canopy of rice, but it has human error due to
Agricultural Mechanization, Pampanga State Agricultural University, variability and poor catch of arthropods at the base of the
Magalang, Pampanga, Philippines (e-mail: [email protected]).
Bobby D. Gerardo is with the Institute of Information and
plant; tapping the rice, this is a sampling method that utilize a
Communications Technology, West Visayas State University, Lapaz, Iloilo collecting pan with soap solution or oil with water to collect
City, Philippines (e-mail: [email protected]). arthropods at the base and stem of the rice. After tapping,
Bartolome T. Tanguilig III is with the College of Information Technology arthropods are identified and counted immediately in the
Education, Technological Institute of the Philippines, Quezon City,
Philippines (e-mail: [email protected]). field; the visual counting and data recording can be done on

DOI: 10.7763/IJCCE.2014.V3.317 189

 International Journal of Computer and Communication Engineering, Vol. 3, No. 3, May 2014

field but also subject to human error and very costly in labor; novel method to identify insects. They used integrated region
sticky trap is also economical, it measures insect movement matching and dual tree complex wavelet transform for image
and colonization but does not catch non flying insect; yellow matching. The collected images of lepidopteran insects are
pan trap is also economical; it measures insect migration, reduced into 248×200 pixels and filtered using mean shift
easy sorting and counting of samples, but the attraction is due algorithm. The processed image is segmented in region based
to color stimulus and does not catch none flying insects. on k-mean algorithm using color feature. They used a
pixel-wise segmentation to classify pixels into region
B. Feature Extraction and Image Processing
according to their color properties.
Image processing is the analysis and manipulation of
graphical images from sources such as photographs and
videos. There are three main steps in image processing; first, III. MATERIALS AND METHODS
is the conversion of captured images into binary values that a
computer can process; second, is the image enhancement and A. Image Acquisition
data compression; and the third is the output step that consists The experiments on pest detection were carried out in the
of the display or printing of the processed image. Image paddy fields. The authors setup a network of wireless
processing is used in such applications as satellite weather cameras (protected against water projection and direct sun
mapping, machine vision, and computer-based pattern light) together with the sticky traps to capture the insect pests.
recognition. The novel feature extraction presented by The cameras used are CISCO Linksys Wireless-G Internet
Phinyomark, Limsakul and Phukpattaranont [4] for Home Monitoring Camera which can capture 10 frames per
Electromyography signal used white gaussian noise to seconds at 8 megapixel resolution. The captured images were
represent interference. Two novel mean and median processed using a local machine equipped with Intel i3
frequencies were presented for robust feature extraction. The processor and 4 GB RAM. The architectural design of the
paper by Alsmadin, Omar, Noa and Almarashdeh [5], proposed system is shown on Fig. 1.
presented the development of a system that can recognize
isolated pattern of interest on image based on the
combination of robust features extraction. The system was
used as classifier for fish image recognition and able to
classify and categorize fish into poison or non-poison family.
C. Real Life Application of Image Processing in the Fields
of Agricultural Research
Several attempts have been made to build automatic insect
identification systems based on image analysis. The study
conducted by Samantha and Ghosh [6] is concentrated to
eight major insect pests based on the records of tea gardens of
North Bengal Districts of India. The authors apply correlation Fig. 1. Global architectural design.
based feature selection for the feature extraction and
reduction, and incremental back propagation neural network B. Image Pre-processing
as the neural network algorithm used for classifications. Image pre-processing creates an enhanced image that is
Do, Harp and Norris [7] designed a computerized pattern more useful in processing the still image. In RGB color
recognition system for non-specialist in recognizing arachids model, each color appears in its primary spectral components
and other arthropods to make the specimen identification of red, green, and blue. The color of a pixel is made up of
easier and accurate. The researchers promote early insect pest three components; red, green, and blue (RGB), described by
detection in greenhouse crops in order to reduce pesticide use. their corresponding intensities. RGB color image require
The target application of this system is the detection of pests large space to store and consume much time to process. In
on plant organs such as leaves. The goal of their work is to image processing it needs to process the three different
define an innovative decision support system for in situ early channels so it consumes large time. In this study, grayscale
pest detection based on video analysis and scene image is enough for the method so the authors convert the
interpretation from multi-camera data. In this research, they RGB image into grayscale image with the following formula:
used apriori algorithm to detect white flies and aphids, and
follow a generic approach to design a system that can easily I(x, y) = 0.2989×R + 0.5870×G + 0.1140×B (1)
adapt different categories of bioagressors.
Al-Saqer [8] developed a neural network-based C. Detection of Pests in the Image
identification system for pecan weevils. They used image The detection mechanism used to detect the insect pests in
descriptors as input in the neural network to recognize the the image is simple and yet efficient. The authors compared
pecan weevil. The authors collected different images of the image pixel values of the successive captured images
pecan weevil and other insects found in the paddy fields. from the camera. Two images are used in detecting the
After collecting the images they converted them into binary difference. The first image served as the reference image that
and resized to 114×134 pixels. To process the data, they used represents the reference pixel values for comparison purposes,
different image processing techniques such as Regional while the second image served as the input image. The two
Properties and Zernike Moments. images were compared to each other and the differences in
The paper presented by Zhu and Zhang [9] proposed a pixel values were determined. If the inputted image pixel

190
 International Journal of Computer and Communication Engineering, Vol. 3, No. 3, May 2014

values are not equal to the reference image pixel values, the
inputted image pixel will be saved as the output image pixel.
If the pixel value of the input image is equal to the reference
image then the background will be white (pixel value is 255).
Fig. 2 shows the difference between reference and input
image. After determining the difference of the reference and
input image, the input image will now be used as the
reference image and the output image will be the extract and
served as the input on the next phase.
To determine the difference of the reference and input Fig. 3. Calculates the median value of a pixel neighborhood. The selected
image, the following equation will be used: pixel value of 126 is replaced with median value of 122 because it is
unrepresentative of the surrounding pixels.

255, 𝑖𝑓 𝑅(𝑥, 𝑦) = 𝐼 𝑥, 𝑦 E. Extraction of the Detected Pests

𝑂𝑅𝐼 𝑥, 𝑦 = (2)
𝐼 𝑥, 𝑦 , 𝑖𝑓 𝑅(𝑥, 𝑦) ≠ 𝐼(𝑥, 𝑦)
This phase is the extraction of the detected insect pest from
the image. The output image which was obtained at the end of
where I (x, y) is the input image and R(x, y) in the reference
the previous phase was used in this phase. The image pixel
image in coordinate (x, y).
values of the output image will be scanned both horizontally
and vertically to determine the coordinates of each insect in
the image. The width and height of the extracted image was
determined by using its starting and ending coordinates. Once
the start and end coordinates of the objects are determined,
the width and height of the matrix as defined by those
coordinates will take note and the matrix will be saved.
The method used in extracting has two steps. First, the
value of the output image in pixels is scanned horizontally
which starts at the first x-direction, then the pixel values are
summed up to the corresponding columns. The coordinate
value in x-direction is incremented by one and the total pixel
Fig. 2. Determining the difference of the reference and input image.
value in the next column is calculated. The process is
To demonstrate the difference between the reference R(x, y) repeated until reaching the last value in the x-direction. As a
and input image I(x, y), assume that we have two 4x4 result, the total pixel values of each column are calculated. In
grayscale images. The two images are represented in the order to determine the x-coordinate where an object starts or
following matrices: ends within the image, each total value is compared to a
certain threshold value. Second, it will vertically repeat the
230 230 230 230 horizontal scanning method; therefore the total pixel value in
230 230 230 230 each row is calculated. Then, to determine the y coordinates,
𝑅 𝑥, 𝑦 = the thresholding will apply where objects start or end within
230 230 230 230
230 230 230 230 the image. The mathematical description of the object
extraction method is shown in the following equations.
230 230 230 230 Starting at coordinates (x=0, y=0) until coordinate (x= 𝑛𝑥 ,
230 30 22 230 y=𝑛𝑦 ), the total pixel value of the column and row are defined
𝐼 𝑥, 𝑦 =
230 24 23 230 as:
230 230 230 230 𝑛
𝑇𝑜𝑡𝑎𝑙𝑌𝑥 = 𝑦 𝑦=0 𝑂(𝑥, 𝑦) (3)
255 255 255 255
𝑛𝑥
255 30 22 255 𝑇𝑜𝑡𝑎𝑙𝑋𝑦 = 𝑥 =0 𝑂(𝑥, 𝑦) (4)
𝑂 𝑥, 𝑦 =
255 24 23 255
The comparison rules are defined as follows:
255 255 255 255
𝑆𝑥 = 𝑦 𝑖𝑓 𝑇𝑜𝑡𝑎𝑙𝑌𝑥 ≤ 𝜃 (5)
where the O(x, y) is the output image in coordinate (x, y).
𝐸𝑥 = 𝑦 𝑖𝑓 𝑇𝑜𝑡𝑎𝑙𝑌𝑥 > 𝜃 (6)
D. Filtering of the Image
𝑆𝑦 = 𝑦 𝑖𝑓 𝑇𝑜𝑡𝑎𝑙𝑋𝑦 ≤ 𝜃 (7)
Filtering is a process of cleaning up the appearance of the
image from noise caused by different lighting conditions. 𝐸𝑦 = 𝑦 𝑖𝑓 𝑇𝑜𝑡𝑎𝑙𝑋𝑦 > 𝜃 (8)
Digital image processing required filtering to yield a usable
and attractive end result. There are different techniques where 𝜃 is the threshold, 𝑆𝑥 and 𝐸𝑥 are respectively the
available and the best options depend on the image and how it starting and ending x coordinates of the object image and 𝑆𝑦
will be used. In this study the authors use median filter. and 𝐸𝑦 are respectively the starting and ending y coordinates
Median filter looks at its nearby neighbor’s pixel values to of the object image.
decide whether or not it is representative of its surrounding The size of the extracted object image is (𝐸𝑥 -𝑆𝑥 ).(𝐸𝑦 -𝑆𝑦 ).
pixels and replaces with the median of those values. Fig. 3 The application of this algorithm continues until the row at
illustrates an example calculation. the last y-coordinate pixel is reached.

191
 International Journal of Computer and Communication Engineering, Vol. 3, No. 3, May 2014

IV. EXPERIMENT RESULTS [4] A. Phinyomark, C. Limsakul, and P. Phukpattaranont, “A novel feature
extraction for robust EMG pattern recognition,” Journal of Computing,
This section presents the experiment results after testing vol. 1, pp. 71-80, Dec. 2009.
the proposed system. It explains first the experimental setup [5] M. K. Alsmadin, K. B. Omar, S. A. Noah, and I. Almarashdeh, “Fish
recognition based on robust features extraction from color texture
then shows the results on different test. measurements using back-propagation classifier,” Journal of
The proposed pest detection and extraction system was Theoretical And Applied Information Technology, 2010.
tested in the practice farm of Pampanga State Agricultural [6] R. K. Samanta and I. Ghosh, “Tea insect pests classification base on
artificial neural networks,” International Journal of Computer
University, Magalang, Pampanga, Philippines. The area has
Engineering Science, vol. 2, June 2012.
50x50 meters wide. The specimens were collected on the [7] M. T. Do et al., “A test of a pattern recognition system for
sticky traps that were captured by the 4 wireless cameras identification of spiders,” Bulletin of Entomological Research, 1999.
installed in the paddy field. The cameras were used to [8] S. M. Al-Saqer, “A robust recognition system for pecan weevil using
artificial neural networks,” American Journal of App. Sci., 2012.
observe the sticky traps which capture a still image every 1 [9] L. Zhu and Z. Zhang, “Insect recognition based on integrated region
minute. These images were used both in the development and matching and dual tree complex wavelet transform,” Journal of
system testing. Zhejiang University-SCIENCE C (Computers & Electronics), 2011.
The proposed pest detection system based on image
processing techniques was tested in five consecutive days in Johnny L. Miranda received BS in information
the paddy field and was found efficient. The result is shown technology and masters in information technology on
2002 and 2007. Currently, he is pursuing his doctoral
on Table I. studies in information technology at Technological
Institute of the Philippines, Quezon City. His research
TABLE I: THE DETECTED IMAGES OF DIFFERENT INSECT PESTS IN FIVE interest includes data mining, image processing,
CONSECUTIVE DAYS artificial intelligence and database.
Camera Camera Camera Camera
Total
1 2 3 4
Day 1 15 12 13 10 50 Bobby D. Gerardo is currently the vice president of
Day 2 Administration and Finance of West Visayas State
12 10 12 10 44
University, Iloilo City, Philippines. His dissertation is
Day 3 10 13 12 11 46 “Discovering driving patterns using rule-based
Day 4 14 15 10 13 52 intelligent data mining agent (RiDAMA) in distributed
Day 5 12 11 10 5 38 insurance telematic systems.” He has published 54
research papers in national and international journals
and conferences. He is a referee of international
conferences and journal publications such as IEEE
V. CONCLUSION Transactions on Pattern Analysis and Machine Intelligence and IEEE
Transactions on Knowledge and Data Engineering. He is interested in the
In this paper, an automatic detection and extraction system following research fields: distributed systems, telematics systems, CORBA,
was presented, different image processing techniques were data mining, web services, ubiquitos computing and mobile
communications.
used to detect and extract the pests in the captured image. The
Dr. Gerardo is a recipient of CHED Republica Award in Natural Science
presented system is simple and yet efficient. The authors used Category (ICT field) in 2010. His paper entitled “SMS-based automatic
background modeling to detect the presence of insect pests in billing system of household power consumption based on active experts
the captured image, and a median filter was used to remove messaging” was awarded best paper on December 2011 in Jeju, Korea.
Another best paper award for his paper was “Intelligent decision support
the noise produced by different lighting conditions. The using rule-based agent for distributed telematics systems,” presented at the
mechanism used to extract the detected objects from the Asia Pacific International Conference on Information Science and
image is simple, the image was scanned both horizontally and Technology, on December 18, 2008. An excellent paper award was given for
his paper “Principal component analysis mechanism for association rule
vertically to determine each coordinates and save the object mining,” on Korean Society of Internet Information’s (KSII) 2004 Autumn
image. Conference on November 5, 2004. He was given a university researcher
The result presented in this paper is promising but several award by West Visayas State University in 2005.
improvements on both materials and methods will be carried
Bartolome T. Tanguilig III was born on February 24,
out to reach the requirements of fully automated pest 1970 in Baguio City, Philippines. He took his bachelor
detection, extraction and identification system. In the future, of science in computer engineering in Pamantasan ng
other image processing techniques may be used to enable the Lungsod ng Maynila, Philippines in 1991. He finished
his master degree in computer science from De la Salle
detection and extraction more efficient and accurate. Other University, Manila, Philippines in 1999. His doctor of
future work may include the identification system of the philosophy in technology management was awarded
extracted objects. This time a neural network mechanism will by the Technological University of the Philippines,
Manila in 2003.
be used to identify and classify the detected pests. He is currently the assistant vice president of Academic Affairs and
concurrent dean of the College of Information Technology Education and
REFERENCES Graduate Programs of the Technological Institute of the Philippines, Quezon
City. His research entitled “J-master: an interactive game-based tool for
[1] Resource Manual on Integrated Production and Pest Management teaching and learning basic java programming” was awarded the best
(IPPM) in Rice, World Education (INGO) Philippines, Inc., 2005, ch. 6, research in the 10th National Convention for IT Education held in Ilocos
pp. 113-115. Norte, Philippines in 2012. He published a research entitled “Predicting
[2] Sampling insect pests, their damage and beneficial organisms faculty development trainings and performance using rule-based
Philippine rice self-sufficiency plan (PRSSP). [Online]. Available: classification algorithm” in Asian Journal for Computer Science and
http://pinoyrkb.com/main/resources/publications-and-figures/rice-scie Information Technology.
nce-for-decisionmakers/doc_download/517-sampling-insect-pests Dr. Tanguilig is a member of Commission on Higher Education Technical
[3] F. A. Carino, P. E. Kenmore, and V. A. Dyck, “A FARMCOP suction Panel for IT Education, Board Chairman of Junior Philippine IT
sampler for hoppers and predators in flooded rice fields,” The Researchers, member of Computing Society of the Philippines and
International Rice Research Newsletter, vol. 4, ch. 5, pp. 21-22, 1979. Philippine Society of IT Educators-NCR.

192

View publication stats