The 5tn Student Conference on Research and Development -SCOReD 2007
11-12 December 2007, Malaysia
Classification of Rubber Tree Leaf Diseases Using
Multilayer Perceptron Neural Network
Noor Ezan Abdullah, Athirah A. Rahim, Hadzli Hashim and Mahanijah Md Kamal
Abstract-This paper presents about classification of rubber tree color of a pixel is given as three values corresponding to the
leaf diseases through automation and utilizing primary RGB well known band R (Red), G (Green) and B (Blue) [4]. Other
color model. Several rubber tree leaf diseases are been studied of color spaces are also used in color features extraction.
for digital RGB color extraction where three sets of rubber tree Artificial neural network (ANN) is basically a different
leaf diseases image are digitally captured under standard and
control environment. The identified regions of interest (ROI) of
paraig formptin
wh.ic
a
base
on
the
tprocessof
these diseases images are then processed to quantify the
human informatinesy Itnis
a
fom ofeuliprocesso
normalized indices from the RGB color distribution. This system computer system generally consists of simple processing data,
involved the process of image classification by using artificial high degree interconnection, simple scalar massages and
neural network where 600 samples were used as training while adaptive interaction between elements [5]. ANN has been
another 200 samples were for testing. The optimized ANN model used for evaluating various pattern recognition or classifying
in this work has two method which based only on the dominant purposes with various algorithms such as multi-layer back
pixel RGB (mean) and applying principle component analysis propagation, unimodal Gaussian, K- nearest neighbor, and
(PCA) on the pixel gradation values of each image. The nearest cluster algorithms [6].
optimized model was evaluated and validated through analysis of
With the advancement of
computer technology,
the performance indicators. Findings in this work have shown .
that both models have produced about 70% in diagnostic processlng
and analysis
of
any rubber leaves images
can be
accuracy with more than 80% achievement for sensitivity. visualized and can be very cost effective. Since leaves
However, model with the applied PCA has lower network size. presentation can also be presented in terms of digital images,
therefore they can be processed and measured to produce
Index Terms- RGB, Rubber Leaf Diseases, ANN, PCA important quantitative features information. These features
can be used then in designing an automated model for
I. INTRODUCTION discriminating type of leaf disease. Therefore in this work,
Rubber is one of the most important products of two models are proposed for automated disease identification
Malaysia. Every year, large amount of latex are produced and using ANN Levenberg Marquardt. Model 1 (after this will be
due to the increasing of rubber-based products in the market, it known as Ezanl) uses the three dominant pixel (mean) in
is a necessity to maintain the quality and quantity. Thus, good RGB color space as the inputs, while Model 2 (after this will
care of the rubber tree must be taken so that they are free from be known as Ezan2) utilizes only the input size of pixel
diseases. The diseases of rubber can be divided into four gradation values for each image that has being reduced after
categories; root, panel, stem and branch and leaf diseases [1]. applying principle component analysis (PCA). The optimized
Conventionally, prevention from diseases is through assessing model is later evaluated and validated through analysis of the
via visual inspection regularly and tries to match its performance indicators.
appearance to the closest appearance photo from a library text.
However, this evaluation process are time consuming, has low II. METHODOLOGY
percentage in accuracy and as well as costly [2]. A. Data Collection
Color perception plays important role in pattern The leaf samples used in this research were collected from
recognition. It interacts with the surface and the interior of an the nursery in Lembaga Getah Malaysia, Sg. Buloh. Three
object through absorption and scatterings. Thus, causes different classes of rubber trees were selected for this study
alteration in the spectral composition. Since color conveys which is the corynespora, bird's eye spot and collectotrichum.
significant information, color information seems to be suited The leaf samples were collected from a number of different
as the first step in analyzing leaf disease [3]. Color space is rubber trees that exhibited the various disease conditions. The
commonly used for object recognition, color segmentation, samples were brought to a laboratory and then sealed in plastic
image retrieval and image understanding. In color extraction, bags with appropriate labels to maintain the moisture level of
ASP Research Group Faculty Electrical Engineering Universiti Teknologi
MARA 40450 Shah Alam, Selangor, Malaysia (e-mail:
1-4244- 1470-9/07/$25.OO ©C20071 IEEE.