Content uploaded by Chathura Suduwella
Author content
All content in this area was uploaded by Chathura Suduwella on Mar 20, 2018
Content may be subject to copyright.
Poster Abstract: A Machine Learning Approach for Identifying
Mosquito Breeding Sites via Drone Images
Akarshani Amarasinghe, Chathura Suduwella, Charith Elvitigala, Lasith Niroshan, Rangana
Jayashanka Amaraweera, Kasun Gunawardana, Prabash Kumarasinghe, Kasun De Zoysa,
Chamath Keppetiyagama
University of Colombo, School of Computing, Sri Lanka.
akarshani@scorelab.org,cps@ucsc.cmb.ac.lk,{charitha,lasith}@scorelab.org
{rja,kgg,jpk,kasun,chamath}@ucsc.cmb.ac.lk
ABSTRACT
Dengue is one of the deadly and fast spreading diseases in Sri
Lanka. The female Aedes mosquito is the dengue vector and these
mosquitoes breed in clear and non-owing water. The Public Health
Inspectors (PHIs) are tasked with detecting and eliminating such
water collection areas. However, they face the problem of detecting
potential breeding sites in hard-to-reach areas.
With the technological development, the drones come as one of the
most cost eective unmanned vehicles to access the places that a
man cannot access.
This paper presents a novel approach for identifying mosquito
breeding areas via drone images through the distinct coloration
of those areas by applying the Histogram of Oriented Gradients
(HOG) algorithm. Using the HOG algorithm, we detect potential
water retention areas using drone images.
CCS CONCEPTS
•Computing methodologies →Supervised learning by clas-
sication;•Computer systems organization →Robotics;
KEYWORDS
Dengue, Drone Systems, Mosquito Breeding Sites
ACM Reference Format:
Akarshani Amarasinghe, Chathura Suduwella, Charith Elvitigala, Lasith
Niroshan, Rangana Jayashanka Amaraweera, Kasun Gunawardana, Prabash
Kumarasinghe, Kasun De Zoysa, Chamath Keppetiyagama. 2017. Poster
Abstract: A Machine Learning Approach for Identifying Mosquito Breeding
Sites via Drone Images. In Proceedings of 15th ACM Conference on Embedded
Networked Sensor Systems (SenSys’17). ACM, New York, NY, USA, 2 pages.
https://doi.org/10.1145/3131672.3136986
1 INTRODUCTION
Good health is one of the expectations of all human beings. As well
as the life expectancy is an indicator of the development in a certain
country. The inuence of spreading deadly diseases such as Dengue
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for prot or commercial advantage and that copies bear this notice and the full citation
on the rst page. Copyrights for components of this work owned by others than ACM
must be honored. Abstracting with credit is permitted. To copy otherwise, or republish,
to post on servers or to redistribute to lists, requires prior specic permission and/or a
fee. Request permissions from permissions@acm.org.
SenSys’17, November 6–8, 2017, Delft, The Netherlands
©2017 Association for Computing Machinery.
ACM ISBN 978-1-4503-5459-2/17/11. . . $15.00
https://doi.org/10.1145/3131672.3136986
can victimize thousands of human lives per year. According to the
WHO (World Health Organization), 70% of dengue infected persons
are from South East Asia and Western Pacic [
5
]. As well as Latin
American, Caribbean, African and Eastern Mediterranean regions
are severely aected by Dengue in the last decade [
5
]. Dengue is a
viral infection and transmitting by the bite of an infected female
Aedes mosquito [
5
]. Nowadays, dengue becomes a global threat
and spreads its authority in urban and semi urban areas with the
tropical and the subtropical climates in many countries. Unfortu-
nately, there is not a specic treatment against dengue fever [2].
The female Aedes mosquitoes aim the places with stagnant wa-
ter for breeding. However, there can be some places with stagnant
water that a man cannot easily reach or identify (i.e. roof gutters,
water tanks, inaccessible rooftops and cement materials which are
capable of retaining water). Due to the impact of retaining water
for a long period time, those areas are covered with lichens and
algaes [
1
]. According to the observations, these lichens and algaes
of those places can be distinguished from the rest, due to their dark
color [
1
]. This color specication is utilized for the identication of
mosquito breeding sites in unreachable places.
A drone or an Unmanned Aerial Vehicle (UAV) bulks among a
larger audience, due to its smaller size and maneuverability com-
pared with other unmanned equivalents. Accordingly, a drone can
reach and perceive the places that a human cannot. Therefore, this
work presents a solution for identifying possible mosquito breeding
sites through a drone management system by analyzing the drone
images of the places that are dicult to reach in the urban and the
semi urban areas with the tropical and the subtropical climates as
a guidance for the PHIs to suppress dengue mosquito breeding.
2 HOG FEATURES FOR WATER RETENTION
AREA IDENTIFICATION
According to our reviews, there are only two approaches for water
retention area identication; Suduwella et al.
'
s [
7
] and Amaras-
inghe et al.
'
s [
6
]. The main problem of those two advances is that
the color of some roads really looks like the color of the water re-
tention areas. So some roads have also been marked as the possible
water retention areas in the output image. Furthermore, the nal
results depend on the drone camera tilt angle and the eects of
shadows in Suduwella et al.'s methodology [7].
SenSys’17, November 6–8, 2017, Del, The Netherlands Amarasinghe et al.
As a solution for aforementioned problems and as an improve-
ment of Amarasinghe et al.
'
s method, we decided to come up with
a solution for detecting certain water retention areas through a
machine learning approach. The HOG algorithm is used to extract
features from test dataset, since it is one of the major object detec-
tion methods [
3
]. We have generated 660 positive features and 140
negative features using the HOG feature extraction methodology.
After that, we generated and trained three classiers referring Sup-
port Vector Classication (SVC) and utilizing those positive and
negative features under dierent gamma values (kernel =
'
rbf
'
, cost
function (C) = 1). Then we analyzed those classiers by determining
the capability of detecting the possible water retention areas.
3 PRELIMINARY EVALUATION AND FUTURE
WORK
The approach is mainly focusing on the water retention area iden-
tication that a man cannot access. The evaluation was carried
out for 100 images which are captured from dierent areas in Sri
Lanka through several drone ights. After that, we checked each
image after applying the HOG feature detection algorithm using
three dierent gamma values and observe whether it identies the
possible water retention areas.
For the nal result we analyzed True Positive (TP), True Negative
(TN), False Positive (FP) and False Negative (FN) counts. In here,
TP is the HOG feature detection algorithm detects a possible water
retention area, and there is such an area in the image. Similarly,
remaining TN, FP and FN are dened accordingly in this context.
Based on those values, recall and precision were derived (Table 1,
Figure 1).
Gamma Value Recall Precision
0.01 90.56% 84.01%
1 94.33% 91.14%
100 72.64% 82.51%
Table 1: Recall and Precision Values for Three Dierent
Gamma Values for 100 Images
Figure 1: The Line Chart for Recall and Precision Values for
Three Dierent Gamma Values
According to the Table 1 and the Figure 1, when the gamma
value equals to 1, it shows the higher recall and precision values.
Furthermore, when the gamma value is decreasing and increasing
from 1 to 0.01 and from 1 to 100, the recall and precision values are
decreasing. So to get better results the more adequate gamma value
is 1 when the kernel is 'rbf 'and the cost function is 1.
We compared our suggested approach when gamma equals to 1
(since it shows the higher recall and precision) with Amarasinghe
et al.
'
s approach (Figure 2). The suggested approach shows a com-
paratively higher recall and precision values.
With this paper we are publishing a database which contains
possible water retention areas [
4
]. Also, we expect to train those
positive and negative features using other machine learning algo-
rithms to get more accurate results. Furthermore, we are planning to
create another database which consists of other mosquito breeding
sites such as coconut shells and tyres.
Figure 2: The Bar Chart for the Comparison Results with
Other Approaches
ACKNOWLEDGMENTS
Special thanks go to the University Grants Commission (UGC), Sri
Lanka.
REFERENCES
[1]
2017. Algae, Lichen, Moss can eat roof, concrete or siding. (2017). Retrieved
August 14, 2017 from http://wparc.net/algae-lichen-moss- eating-roof/
[2]
2017. Dengue fever - Treatment - Mayo Clinic. (2017). Retrieved Au-
gust 14, 2017 from http://www.mayoclinic.org/diseases-conditions/dengue-fever/
diagnosis-treatment/treatment/txc- 20345589
[3]
2017. Histogram of Oriented Gradients | Learn OpenCV. (2017). Retrieved August
14, 2017 from http://www.learnopencv.com/histogram-of-oriented- gradients/
[4]
2017. scorelab/D4D-Drone-4-Dengue, GitHub. (2017). Retrieved August 14, 2017
from https://github.com/scorelab/D4D---Drone-4- Dengue/tree/master/d4d-data/
detecting_water_retention_areas
[5]
2017. WHO | What is dengue and how is it treated? (2017). Retrieved August 02,
2017 from http://www.who.int/features/qa/54/en/
[6]
Akarshani Amarasinghe, Chathura Suduwella, Lasith Niroshan, Charith Elvitigala,
Kasun De Zoysa, and Chamath Keppetiyagama. 2017. Suppressing Dengue via
a Drone System. Advances in ICT for Emerging Regions (ICTer), 2017 Sixteenth
International Conference on. IEEE. (2017).
[7]
Chathura Suduwella, Akarshani Amarasinghe, Lasith Niroshan, Charith Elviti-
gala, Kasun De Zoysa, and Chamath Keppetiyagama. 2017. Identifying Mosquito
Breeding Sites via Drone Images. Proceedings of the 3rd Workshop on Micro Aerial
Vehicle Networks, Systems and Applications (2017), 27-30. https://doi.org/10.1145/
3086439.3086442