Page 169 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

158 Chapter 6 Plant leaf disease classification based on feature selection

3. Our proposed framework

3.1 Data set
Our data set contains 450 images of mango leaves, which be-
longs to four different types (three diseases and one healthy):
Anthracnose, Gall Midge, Powerdery Mildew, and Healthy. These
are also four classes in our classification as in Fig. 6.1. The sam-
ples are collected from various places in An Giang province, which
is known as one of the places with the largest productions of
mango in Vietnam. The images are captured using a smartphone
camera in the resolution of 3096 3096 pixels with no back-
ground. Some images in the data set are shown in Fig. 6.2.
The proposed model for mango disease identification is
shown in Fig. 6.3, in which there are four main stages. To begin
with, by rescaling, the images in the data set are converted into a
lower resolution, compared with the original size. Then, center
alignment step is responsible for guaranteeing the region of a
leaf to be in center of image fitting exactly top and bottom of
the image. Since there are various contrasts in leaf images, we
apply the contrast enhancement method to adjust pixel inten-
sities which benefit in case of providing more information in
some areas of an image. The image data set will be divided into
two subsets: training set and test set. Finally, the convolution
neural network is applied to classify the given images.

3.2 Image preprocessing

Since the leaves have different sizes, it is necessary to perform
rescaling to ensure the training and testing image have the same

Figure 6.1 Four classes of leaf diseases in this study: Anthracnose, Gall Midge,
Healthy, and Powerdery Mildew.