Page 246 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

Chapter 8 A review on plant diseases recognition through deep learning 237

backpropagation and convolution neural networks by Yann
LeCun at Bell Labs. Support vector machine (SVM) was developed
by Dana Cortes and Vladimir Vapnik, which maps and recognizes
similar data. Long short-term memory (LSTM) for recurrent
neural networks was developed in 1977 by Sepp Hochreiter and
Juergen Schmidhuber. ImageNet was assembled by an AI expert
Fei-Fei Li and can process more than 14 million labeled images.
AlexNet, the champ of deep learning, won many competitions
during 2012. The Cat Experiment, developed by Google in 2012,
was a boon for unsupervised learning.

6.2 Without visualization technique
Deep neural networks (DNNs), proposed in 2012, particularly
convolutional neural networks (CNNs), are providing the most
booming applications such as object recognition, detection,
biometry, and classification. CNNs produce a hierarchy of visual
representations and thus act as a matching filter. The main
feature of CNNs is the generalization concept that is the capacity
to handle the data that are never observed before. Sibiya et al.
proposed a CNN model to classify the maize plant disease along
with histogram techniques [42]. For identifying tomato leaf
disease, Zhang et al., implemented ResNet, which is considered
as the superior algorithm among all the CNN, GoogLeNet, and
AlexNet [43]. LeNet architecture was implemented to identify
banana leaf disease, and F1 and CA scores were used to evaluate
the model in both grayscale and color model [44]. In 2018, Feren-
tinos proposed five CNN architectures such as VGG, AlexNet,
Overfeat, and GoogLeNet. Among these models, VGG architecture
is considered as the outclassed model [45]. Extreme learning
machine (ELM), SVMs, and K-nearest neighbor (KNN) are com-
bined with deep learning models such as ResNet-50, Inception-
v3, SqueezeNet, InceptionResNetv2, ResNet-101, and GoogLeNet
recognized eight different types of plant diseases [46]. These
models are analyzed and concluded that ResNet-50 with SVM
classifier has produced accurate results in terms of specificity
and sensitivity F1 score. Superresolution convolutional neural
network (SRCNN) replaced the traditional plant disease diagnosis
method [47]. AlexNet and SqueezeNet v1.1 were used to classify
diseases of the tomato plant; in this, AlexNet is identified as a
more accurate deep learning model [48].