Page 248 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

Chapter 8 A review on plant diseases recognition through deep learning 239

hyperspectral imaging (HSI), multispectral imaging, fluorescence,
and thermal imaging [63]. HSI is the most effective model when
compared with others in terms of efficiency, performance, and
accuracy. This chapter focuses on deep learning approaches
based on HIS technique.
Many applications are based on deep learning, which clas-
sifies the hyperspectral images. In the plant disease detection,
the combination of HSI and deep learning produces a clear vision
about the symptoms. Yue et al., proposed a hybrid methodology
in the classification of hyperspectral images along with principal
component analysis, logistic regression, and deep convolutional
neural network algorithms and produced the best results [64].
Signoroni et al. proposed a review of the deep learning technique
with HSI concept, concentrating on overfitting problems to pro-
duce greater accuracy [65]. Several state-of-the-art deep learning
algorithms such as 1D/2D-CNN, 2D-CNN-LSTM/GRU, and
LSTM/GRU are compared and observed. 2D-CNN-BidLSTM/
GRU, a novel hybrid approach of bidirectional gated recurrent
and convolutional network, for wheat disease detection resolves
the overfitting issues by achieving 0.75 F1 score and 0.73 accuracy
[66]. Generative adversarial networks (GANs), a deep learning
technique based on a hyperspectral proximal-sensing procedure,
were proposed to identify disease in tomato plants before the
appearance of clear symptoms [67]. To detect the charcoal rot
disease in soybeans, a 3D-CNN approach was proposed for hyper-
spectral images that produced F1-score (0.87) and accuracy
(95.76%) [68]. Saliency map visualization technique is used, which
produced 733 nm wavelength, approximately the wavelength of
NIR. Zhang et al. proposed a multiple Inception-ResNet model
to detect the yellow rust in wheat, which works on spectral and
spatial data on hyperspectral UAV images that produced 85%
accuracy when compared with RF classifier, which produced
77% of accuracy [69].

7. Conclusions
Plant disease is considered as the major issue for food safety;
hence, it is necessary to identify rapidly but remains difficult in
many parts of the world due to poor technology and infrastruc-
ture. The recent advancement in the computer vision and the
deep learning eliminates these shortcomings effectively. This
chapter reviewed the deep learning approaches to detect the
plant disease. The classification of plant diseases is discussed
with their symptoms and root cause. Many visualization and