The estimation of rust disease of daylily leaf images with GLCM based different classification methods
Abstract
Crop diseases can affect yield and/or quality of the
harvested commodity. This can influence
profitability and raise the risks of farming. When the
diseases are estimated early, the yield will increase
by taking measures thanks to farmers. The rust
disease is one of the most major crop diseases that
affect crop yield.
Rust disease can be defined as a fungus; it makes the
crops weak by blocking food to the roots and leaves.
It is named “rust” disease, since the spots on the
leaves look like grain of rust which is coloured in
the range of yellow to bright orange, to brown or
red. Some spots have a planar surface, while others
are raised. This disease is infectious amongst
vegetations but not between flowers and vegetables.
The rust firstly seems bright orange. Then, it turns to
dark brown as it proceeds. The infected leaves drop
off and the main stems will show diseased spots as it
spreads. Finally, the crops will die (Dauber 2008).
In general, rust disease can be found in three types
of planting areas. These are yellow rust, brown rust
and black rust. The most common type, called a leaf
or brown rust. This disease is usually seen in the wet
type long leaves. Another common type of rust
disease in plants is called stripe or yellow rust. It is
seen most frequently in the leaves. The last common
type of rust is called black rust and which is the most
destructive kind of rust disease and it causes about
50 % losses per month of crop production efficiency
(Çoklu2011).
In this paper, daylily leaf images are used as crop
sample and derived from different agricultural sites
under expert control and daylily rust disease is
estimated by using GLCM based different classifier
techniques.
Before classification process, the features are
extracted from images with using Gray Level CoOccurrence
Matrix (GLCM) method and 7
parameters are derived by this method for each
digital camera image. These parameters are
contrast, correlation, energy, homogeneity, entropy,
standard deviation and mean for first texture feature
vector.
Then, the extracted feature vectors are applied to
different type of classifiers and these vectors are
used as inputs in classification systems. The
Multilayer Perceptron neural network (MLP) , kNearest
Neighbor (k -NN) and Least Squares
Support Vector Machine (LS-SVM) classifiers have
been chosen for learning and testing of 53 image
data where 32 of them belongs to class I (normal),
21 of them belongs to class II (rust diseased).
Different structures of networks are tested and the
results are compared in terms of testing
performance for each network model.
Artificial Neural Network (ANN) techniques are
non-linear statistical data modeling or decision
making tools. They can be used to model complex
relationships between inputs and outputs or to find
patterns in data. In pattern recognition, the knearest
neighbor algorithm (k-NN) is a method
for classifying objects based on closest training
examples in the feature space. A Least Squares
Support Vector Machine (LS-SVM) is a concept
in computer science for a set of related supervised
learning methods that analyze data and recognize
patterns, used for classification and regression
analysis .These methods were used for classification
system in this paper.
Finally, the best performance was observed as
88.90% in the k-NN and MLP network with 7-5-1
structure. Our results suggest this method is an
accurate and efficient means of estimating daylily
rust disease.
Keywords
