Page 167 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

156 Chapter 6 Plant leaf disease classification based on feature selection

data set, a widely known data set that is available online, but this
data set is limited in terms of training samples in each category.
Also, their approach is convolutional neural network (ConvN/
CNN), which consumes a large amount of time and memory to
train.

2. Literature review

2.1 Plant diseases recognition using convolutional
neural networks
CNNs are a class of hierarchical model where object’sfea-
tures are learned by training through many examples. CNNs
consist of multiple layers with later ones built on top of previ-
ously learned features [1]. Saleem et al. [2]conducted areview
on plant disease detection and classification by DL techniques.
Konstantinos et al. [3] implemented a VGG model for plant dis-
ease detection. Rangarajan et al. [4] used AlexNet and VGG16
to classify tomato leaf diseases. Previous works used the “Plant
Village” [5] data set. This data set has a simple or plain back-
ground, and the sample size in each category is limited, resulting
in high chance of overfitting. A much wider variety of training
data should be collected, from several sources of different
geographic areas, cultivation conditions, and image capturing
modes.

2.2 Plant diseases recognition with artificial
neural network
Khirade et al. [6] discussed various techniques to segment the
disease part of the plant. This chapter also discussed some
feature extraction and classification techniques to extract the
features of infected leaf and the classification of plant diseases.
The use of ANN methods for classification of disease in plants
such as self-organizing feature map, backpropagation algorithm,
support vector machines (SVMs), and so on. Singh et al. [7] used
ANN together with image segmentation to detect diseases on
various types of plants, namely banana, beans, jackfruit, lemon,
mango, potato, tomato, and sabota. The classification is first
done using the minimum distance criterion with k-mean
clustering. In the second phase, classification is done using
an SVM.