Page 244 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

Chapter 8 A review on plant diseases recognition through deep learning 235

the water content in soil or plant predominantly in determining
habitat features of pests and plant diseases [38,39].

5.4 Fluorescence and thermal sensors
Fluorescence and thermal infrared remote sensing systems
monitor photosynthetic and respiratory processes of plants to
track pests and plant diseases. Laser-induced fluorescence (LIF)
is a promising technique to obtain the response of plants.
Some chlorophyll fluorometers such as PAM-2100, PAM-2500,
and IMAGINE-PAM comprise LED light sources in the wave-
length of 655 and 735 nm to produce fluorescent indicators and
records some parameters and kinetic curves. By analyzing
thermal images, biotic and abiotic stresses can be identified
[40]. Some limitations exist in exploiting remote sensor systems
to monitor pests and plant diseases. To measure fluorescent
parameters, dark adaption is required, which limits the feasibility
and efficiency of this remote sensing technique. Sometimes, fluo-
rescence signals are weak that may get converged with natural
light that restricts the number of applications in many areas.

6. Plant disease detection by well-known
deep learning architectures

Several state-of-the-art deep learning architectures are evolving,
following the introduction of AlexNet [41] for classification,
segmentation, and image detection. In this section, various deep
learning algorithms and architectures to detect and classify plant
diseases are discussed elaborately. New visualization techniques
were discussed along with their enhanced version to obtain better
result. Determining the plant disease through image is a tedious
process as the leaves, flowers, and fruits are often changing
throughout the season. The shape and size also change based
on the angle of incident solar radiation. Many techniques have
been emerging to identify the plant diseases based on VIS-NIR
reflectance. The issues related to these techniques are weather
constraints, image acquisition, availability, deployment costs,
real-time diagnostic capabilities, and processing speed. Analyzing
images undergo problems such as nonuniform background or
foliage. Phytosanitary is another complex problem, which is asso-
ciated with symptom variation over time and between varieties or
many disorders at the same time. To detect the disease effectively,
an efficient method is required that should overcome all these
limitations.