Page 245 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

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

6.1 Evolution of Deep learning
Deep learning started its journey in the early 1940s when a
human neural systemebased computer model was developed
by Warren McCulloch and Walter Pitts. This model imitated the
reasoning capacity of the human brain by implementing applied
mathematical concept called threshold logic. Deep learning, a
branch of machine learning, develops abstractions by processing
data. Many algorithms are available to process data and recog-
nize human speech and objects (Fig. 8.2).
Henry J. Kelley developed the backpropagation model in 1960,
but it was not so efficient. Cybernetics and forecasting techniques
were introduced by Valentin Grigor'evich Lapa in 1965. Polyno-
mial activation functions were implemented to handle data by
Alexey Grigoryevich Ivakhnenko. The initial version of convolu-
tional neural networks was developed by Kunihiko Fukushima
with many pooling and convolution layers. The extension of this
network is released in 1979; a multilayered and hierarchical
artificial neural network called the Neocognitron identifies visual
patterns. The learning method adopted by this model is a
top-down approach, which can identify individual pattern. The
later versions of Neocognitron used inference concepts that iden-
tified the missing and unknown values. The efficiency of the back-
propagation model was enhanced by using FORTRAN code
developed by Seppo Linnainmaa in the late 1970s. In 1989, hand-
written characters are identified with the hybrid approach of

AlexNet ZFNet NiN OverFeat
VGG
2012 2013 2013 2013
2014
SegNet Resnet FCN GoogleNet
RCNN
2015 2015 2014 2014
2014
YOLO
FractalNet SSD DenseNet
U-Net
2016
2016 2016 2017
2015
CapsuleNet
PSPNet RefineNet IRRCNN IRCNN
2017
2017 2017 2017 2017
DCRN R2U-Net DeepLab
Mask-RCNN Fast-RCNN
2018 2018 2018
2017 2017
Figure 8.2 Evolution of deep learning model from 2012.