Digital Innovations Revolutionizing Plant Pathology

SUBMITTED BY,
Chandana Y Kundargi

Department of Plant Pathology

College of Agriculture,
Dharwad-580005

INTRODUCTION
1
 The search for effective diagnostic tools for plant pathogen detection and management
has to face new challenges in an era characterized by climate change and intensified global
trades, and recent epidemic events underline its urgency. A large number of harmful “alien”
species, such as viruses, phytoplasmas, bacteria, fungi, insects, nematodes, and weeds, travel
undisturbed along with people and goods (including plant materials), spreading on a large
scale all over the world and causing serious problems to agriculture. Thus, an early detection
of plant pathogens is more and more necessary in plant health monitoring in order to manage
disease infections in different stages of development, minimizing the risk of disease
spreading and avoiding the introduction of new ones. FAO estimates that between 20% and
40% of world crop production is lost annually due to pests, affecting also the main food crops
(rice, wheat, corn, potatoes, soy, and cotton) at national and regional level in the different
continents. Crop losses are correlated with production conditions, having higher losses in
food insecure hotspots (with emerging and re-emerging pests/diseases) and lower in those
with food surpluses.

As a result, “early detection” in combination with “fast, accurate, and cheap” diagnostics
have also become the new mantra in plant pathology, especially for emerging diseases or
challenging pathogens that spread thanks to asymptomatic individuals with subtle initial
symptoms but are then difficult to face. Traditional disease detection, monitoring, and
management methods are often time-consuming, labour-intensive, and prone to errors. With
advancements in technologies like AI, remote sensing, machine learning, and IoT, plant
pathologists can now accurately detect diseases in real-time, predict outbreaks, and
implement targeted interventions. These innovations enable early warning systems, reducing
crop losses and minimizing the excessive use of chemical pesticides, which are harmful to the
environment. By integrating digital tools, we can ensure food security and sustainable
farming amidst evolving threats.

Tools adopted for surveillance and monitoring, according to the traditional approach,
and possible improvements provided by a more advanced, digital-based approach.

Traditional
Advanced Approach Improvements
Approach
Wider range of information with
Weather stations Wireless Sensor Network (WSN)
increased efficiency.
Wider range of information with
Soil sensors Advanced sensors (IoTs)
increased efficiency.

2
 Soil and plant Wider range of information with
Advanced sensors (IoTs)
analysis increased efficiency.
Proximal and remote sensing: Increased data precision, geo-
Geolocalization (GPS, GIS) positioning.
Field
Drones, UAVs, Satellites Large-scale monitoring.
monitoring
Reduced time and cost of the
Monitoring platform
monitoring process.
Pheromone
Image-based traps Real-time data availability.
traps

Tools adopted for fruit and foliar disease diagnosis, according to the traditional
approach, and possible improvements provided by a more advanced, digital-based
approach.

Advanced
Traditional Approach Improvements
Approach
Image analysis
More precise disease quantification.
software
Artificial
intelligence, machine Shortened diagnostic time.
learning
Traditional diagnostic methods
based on symptoms or PCR, RT-PCR
functional changes
LAMP
More precise diagnosis, In-field
RT-RPA analysis, Simple and short sample
preparation methods
In-field analysis
ELISA

Tools adopted for decision making in crop protection, according to the traditional
approach, and possible improvements provided by a more advanced, digital-based
approach.

Traditional Approach Advanced Approach Improvements

Experience Consultants Decision Support Holistic approach.
System (DSS)
Thresholds, Simple rules or Mechanistic, integrated Expert knowledge, advanced model
empirical models models
Good practices, guidelines, Informed decisions Sustainability plays a part in the
protocols decision-making process.

3
 HISTORY

• 1927 - Neblette - Aerial photography (RGB) enables disease survey in agricultural

crops.

• 1933 - Bawden - Black-and-white representation of an infrared photography resulted

in high contrasts between necrotic leaf spots caused by potato viruses.

• 1956 - Colwell - Remotely determined wheat rust and other diseases of grains by
using military helicopters and infrared-filter cameras, as well as a spectrometer at
oblique and nadir observation angles. He proposed a new view on the interaction of
light with plants and the assessment and interpretation of crop photos for plant
diseases.

• Use of spectroscopy combined with chemistry and taxonomy has been coined as
“spectranomics” (Asner and Martin 2009).

• Spectranomics allows for accurate and non-destructive detection of direct and indirect
changes to plant physiology, morphology, and biochemistry which induces the
disease, both pre- and post-symptomatically (Arens et al., 2016).

• Changes to narrowband wavelengths, have proven valuable for plant disease sensing
due to their sensitivity to a range of foliar properties (Curran 1989).

4
 Spectral Range Wavelength (nm) Influencing Factors
Ultraviolet (UV) 100–380 Secondary plant metabolites
Visible (VIS) 400–700 Primary metabolites(e.g., pigments)
Near-Infrared (NIR) 700–1000 Internal scattering processes and leaf structure
Short-Wave Infrared (SWIR) 1000–2500 Chemicals and water

RECENT DIGITAL INNOVATIONS

1. Forecasting models and Expert system

2. Machine learning and AI
3. Remote sensing and Imaging techniques
4. Genomic tools and Bioinformatics
5. Digital platforms for disease surveillance
6. Virtual Reality (VR) and Augmented Reality (AR) in Plant Pathology

Plant disease Forecasting models:

Plant disease forecasting predicts the occurrence of diseases in specific areas ahead of
time to enable timely control measures and minimize economic losses. Predictions rely on the
disease triangle. It’s main objective is to determine when the three factors align to cause
disease outbreaks with potential economic impacts. Forecasting systems differ for
monocyclic and polycyclic diseases, based on factors like initial inoculum levels and inter-
seasonal weather conditions.

History of Disease Forecasting:

 1911 - First attempt at predicting potato diseases - Lutmen - concluded that epidemics
were favored in wet and cold conditions.
 1926 - Van Everdingen in Holland proposed the first model based on four climatic
conditions necessary for late blight development.
 1933 - Beaumont and Stanilund of England emphasized the importance of humidity
for late blight occurrence.
 1978 - Computerized forecasting system called FAST was developed for Alternaria
solani in tomato. FAST system requires fewer fungicide applications compared with
weekly spray schedules to obtain the same level of disease control.
 1985 - Modified FAST program called Tom-Cast was developed to aid in the
management of anthracnose, Septoria leaf spot and early blight on tomatoes. This
establishes a timed fungicide spraying program for processing tomatoes. This

5
 particular model is based on weather data, specifically hourly leaf wetness and
temperature. Number of fungicide sprays reduced by as much as 50% without
compromising fruit quality or yield by using Tom-Cast.

Importance and factors required for disease forecasting:

For timely plant protection measures. Information on whether the disease status is
expected to be below or above the threshold level is enough, models based on qualitative data
can be used – qualitative models. Loss assessment forewarning actual intensity is required -
quantitative model. Host, pathogen and environmental factors are required for plant disease
forecasting.

Pre-requisites for disease forecasting or developing a forecasting system:

1. The disease must have potential to cause losses (yield losses) which is economically
significant.
2. The crop must be a cash crop or economically valuable.
3. Control measures must be available at an economically acceptable cost.
4. The disease must vary each season in the timing of the first infections and its
subsequent rate of progress i.e. the disease should not be regular.
5. The criteria or model used in making a prediction must be based on sound
investigational work carried out in the laboratory and in the field and tested over a
number of years to establish its accuracy and applicability in all the locations where
its use is envisaged.
6. Growers must have sufficient man power and equipments to follow the management
measures when disease warning is given.
7. Long-term warnings or predictions are more useful than short-term warning or
predictions.

Methods of disease forecasting:

Disease forecasting requires field observations on the pathogen characters, collection of
weather data, variety of the crop and certain investigations and their correlations. Usually the
following methods are employed in disease forecasting.
1. Forecasting based on primary inoculums: Presence of primary inoculum, its
density and viability are determined in the air, soil or planting material. Occurrence of
viable spores or propagules in the air can be assessed by using different air trapping
devices (spore traps). In the case of soil-borne diseases the primary inoculum in the

6
 soil can be determined by monoculture method. Presence of loose smut of wheat,
ergot of pearl millet and viral diseases of potato can be detected in the seed lots at
random by different seed testing methods. The extent of many virus diseases is
dependent on the severity of the preceding winter which affects the size of vector
population in the growing season. e.g., Sugar beet yellows virus.
2. Forecasting based on weather conditions: Weather conditions viz., temperature,
relative humidity, rainfall, light, wind velocity etc., during the crop season and during
the intercrop season are measured. Weather conditions above the crop and at the soil
surface are also recorded.
3. Forecasting based on correlative information: Weather data of several years are
collected and correlated with the intensity of the diseases. The data are compared and
then the forecasting of the disease is done. Forecasting criteria developed from
comparisons of disease observation with standard meteorological data have been
provided for diseases like Septoria leaf blotch of wheat, fire blight of apple and barley
powdery mildew.
4. Use of computer for disease forecasting: In some advanced countries forecasting of
disease is made by the use of computers. This system gives the results quickly.
Programme Name Disease Pathogen
EPIDEM (1969) Early blight of potato Alternaria solani
EPIMAY Southern leaf blight of maize Helminthosporium maydis
EPICORN Southern corn blight Helminthosporium maydis
CERCOS Blight of celery Cercospora spp.
MYCOS Blight of Chrysanthemum Mycosphaerella spp.
EPIVEN Apple Scab Venturia inequalis
EPIDEMIC Stripe rust Puccinia striformis
TOM-CAST (Pitbaldo,
Early blight of tomato Alternaria solani
1992)
PLASMO (Rosa et al.,
Downy mildew of grapes Plasmopara viticola
1993)
Contact and aphid transmitted
EPIVET Viral diseases of Potato
viruses
EPIPIRE Cereal rusts and aphids Puccinia graminis tritici
BLIGHT CAST (Karuse et
Late blight of potato Phytophthora infestans
al., 1975)
SIMCAST (Grunwald,
Late blight of potato Phytophthora infestans
2002)
NEGFRY (Runno-and
Late blight of potato Phytophthora infestans
Kopple, 2002)
BioMA Various plant diseases Multiple pathogens
Machine Learning Models Rice blast disease Magnaporthe oryzae
Diseases caused by Xylella
Compartmental Model Xylella fastidiosa
fastidiosa

7
Forecasting models in India:

• Black Stem Rust of Wheat - Nagarajan and Singh (1975) formulated “India Stem Rust
Rules” for Puccinia graminis tritici.

• Brown Rust of Wheat - Nagarajan et al. (1978) proposed an empirical model.

• Potato late blight - Singh et al. (2000) developed JHULSACAST, a computerized

forecast of potato late blight in Western Uttar Pradesh.

• Rice blast - Kapoor et al. (2012) studied overall rice blast severity during 1997, 1998
and 1999 in Kangra district of Himachal Pradesh

Expert System in Plant Pathology:

According to Turban (1992) expert system is a system that uses human knowledge in
a computer to solve a problem that requires human expertise. The uniqueness of the system is
its ability to provide expert advice or recommendations and even justify why an action should
not be recommended.

Expert System PLANT/Disease Description

First Expert System (1983) Developed for the
PLANT/ds Soybean diseases
diagnosis of 17 soybean diseases (USA).
First expert system to incorporate decision-making
Mango and Apple
POMME processes for managing diseases and insects in
diseases and insects
apples.
Developed for 8 diseases and 17 insects, taking
Apple diseases and
PSACO biological, chemical, and control options into
insects
consideration.
Late and early blight of Expert system for managing late and early blight
POTATOES
potato in potatoes.
Uses 78 questions and 87 photographs for
TOMEX Tomato diseases
diagnosing 37 diseases of tomatoes.
Greenhouse disease Expert system used for greenhouse disease
Grenmen
management management.
Expert system offering managerial options for
Cereal, potato, sugar
PRO-PLANT reasonable, economical control of rusts and other
beet, and vegetables
plant diseases.
Managerial options for reasonable economical
More crop Rusts and other problems
control.
Wheat (spring and Epidemic Prediction and Prevention system for
EPIPRE
winter) spring and winter wheat (Zadoks et al., 1984).

8
 Expert system for flower-bulb diseases, developed
EXSYS Flower-bulb diseases
in 1997.
Expert system for the diagnosis of tomato diseases
TOMEx-UFV Tomato diseases
(1998).
Phytophthora infestans Expert system developed for managing
SIMPHYT-III
(Late blight of potato) Phytophthora infestans in 1998.

Advantages and disadvantages of Expert System:

Easy identification of disease, provide consistent answers for repetitive decisions, hold
and maintain significant levels of information, combine multiple human expert intelligences
and user friendly software are the advantages.

Cannot make creative responses, errors may occur in the knowledge base, and lead to
wrong decisions and cannot adapt to changing environments are the major disadvantages.

Remote sensing and Imaging techniques:

Remote sensing is the acquisition of information about an object or area by a

recording device which is no in direct contact with the object under study. Basically, it is a
process of observing and detecting the different characteristics of an area/object by measuring
its reflected and emitted radiation at a distance using satellite or aircraft. Sensor is simply a
device which detects/measures a physical property of an object, records it and in some cases
responds to it. We can find a variety of sensors in our day-to-day life from home to office,
cars, school, stations, hospitals etc., making our lives easier and more comfortable. In
agriculture, sensors can be used for detection of diseased plants, to measure intensity of
damage, to decide whether to control a pathogen or not, and for evaluation of spatial patterns
of plant diseases. This information can help in decision-making for chemical disease control
that can potentially reduce the number of fungicides to be used in future.

Remote Sensing Techniques:

1. Thermography – This method uses imaging of temperature patterns of plant tissues and
crops over a short period of time. It is highly sensitive to early changes in transpiration of
plants, so it may be suitable for pre-symptomatic detection of pathogen activities within plant
tissues.

9
2. Chlorophyll fluorescence of Plants – This method observes the photosystem of plants
which is highly sensitive to biotic and abiotic stresses. The pathogens which affect plant
photosynthetic apparatus e.g., pigments, electron transport chain and enzymes, can be easily
detected using this method.

3. Hyperspectral imaging – It collects and processes information from across the

electromagnetic spectrum. Hyperspectral imaging obtains the spectrum for each pixel in the
image of an object under study to identify it, its properties and changes occurring in it and
their causes.

Areas of Remote Sensing Application:

1. Quarantine and quality control of produce – Physical sensors like thermal sensors uses
autonomous systems for quarantine inspections, which can be applied to bulk material to test
the quality of the produce to detect presence of any quarantine plant disease. It includes
checking the ripeness of produce, color, storage suitability, detecting diseases and assessing
mycotoxin contamination from fungi.

2. Phenotyping of plants – Plant germplasm can be phenotyped for disease susceptibility

under controlled conditions and in field.

3. Monitoring plant health – While growing high value crops in green house if it is
provided with good infrastructure and sensors, it can provide images from ground-based
system close to crop which will be helpful in monitoring plant health in relation to diseases.

4. Study of Host-Pathogen interaction – Remote sensing can be used to monitor the

metabolic changes brought about by diseases development, early stages of diseases and effect
of pathogen colonization on distant non-colonized host tissues depending on host genotype.

5. Controlled application of chemicals- By detecting the incidence and intensity of disease

it will be possible to schedule chemical spray only at specific locations where needed and in
required quantity to control the pathogen.

Challenges in Remote sensing of Plant Diseases:

1. Improved detection of disease at an early stage – To achieve this goal it is necessary to

further exploit the feasibility of fluorescence, thermal and hyperspectral imaging and fuse
them with the Unmanned air vehicle (UAV) i.e. drones and satellite imagery to increase the
area coverage at high accuracy.

10
2. Accurate detection of specific disease under realistic field conditions – Currently most
of the monitoring experiments are conducted under controlled laboratory conditions with
prior information about the diseases and other stress. Without this prior information it is
challenging to get reliable and accurate monitoring results. Deep learning algorithms can be
used in overcoming this problem to establish a data base about the background and nature of
that specific disease under various conditions.

3. Continuous track of the dynamics of the disease at close range – Bad weather and
uneven topography are the major obstacles in collecting information at every stage of the
disease development. To overcome this, it is necessary to combine the remote sensing system
with high resolution satellite images collected with the help of UAV.

4. Data and information sharing – Plant disease epidemics is a global phenomenon, so

multinational collaboration is important for research and application to share experimental
data and ideas for research at global scale.

Drone – based Imaging:

Agricultural drones are also known as Unmanned Aerial Vehicles (UAVs) offer support in
surveillance activity, agronomic activities (planting crops), fighting pests and crop
monitoring. Drones are equipped with digital, multispectral, hyperspectral, thermal, and

fluorescence sensors. Data acquired by drones can be simultaneously sampled by their

autonomous systems at various heights in the atmosphere. Finally, information can be

delivered to farmers.

11
 Dr
one system to detect plant disease and Plant disease detection and classification model
architecture.

Drone system to detect plant diseases comprising four sections: data acquisition, data
preparation, training, and prediction. The system interacts with the target area (fields, farms,
or forests) directly or indirectly to obtain information. The information is preprocessed and
arranged into features, then sent to a supervised machine learning classifier that processes the
data and provides prediction reports via segmented images. The data acquisition process
includes drones which indirectly acquire airborne data and an expert who acquires direct
ground data through visual assessment of the fields. Drones consist of high-performance
brushless rotors, specific load capacity, and dimensions. They are controlled automatically
using ground station software which controls the route, speed, and altitude at a distance.

The plant disease detection and classification model architecture consist of image
acquisition (A), image preprocessing (B), image segmentation (C), feature extraction (D), and
classification (E). Sensors and cameras are mounted on drones that acquire plant disease
images. The images are then processed and segmented to remove background noise. Then,
features are extracted and classified using traditional machine learning models such as K-
nearest neighbor (KNN) or support vector machine (SVM) and deep learning models such as
Convolutional Neural Network (CNN). Note: PCA, principal component analysis; LESC,
local embedding based on spatial coherence algorithm; DWT, discrete wavelet transform.

Sensors mounted on drones for plant disease detection and monitoring:

Remote Sensor Advantages Disadvantages Diseases

Digital Camera Captures grayscale or color Limited spectral bands; Cotton bacterial

12
 images; visible spectrum angular, Ascochyta
enhances disease blight, grapefruit
less light visibility;
identification at the leaf citrus canker, sugar
(RGB) vulnerable to
level; lightweight, beet Cercospora leaf
environmental factors.
inexpensive, easy to use, spot, rust, blights,
simple data processing. smuts, spots.

Low cost, fast frame

imaging, higher efficiency
than RGB cameras; covers Few bands;
Multispectral Blights, blasts,
visible to Near-Infrared discontinuous spectrum;
Camera viruses.
(NIR) spectrum; enables low spectral resolution.
vegetation index
calculations.

Captures narrow bands and

Blasts, blights,
Hyperspectral continuous spectra; provides
Expensive. nematodes, viruses,
Sensing detailed spectral properties
rots, scabs, rusts.
for disease detection.

Issues with temporal and

Sensitive to infrared geographic resolution,
Thermal Recently used for
spectrum; works day and weather dependency,
Infrared Cercospora leaf spot,
night; provides deeper crop species variation,
Cameras (IR) scab, and mildews.
insights into plant health. and height of image
capture.

Limited disease
Fluorescence Assesses how plant stress detection; difficult to Mildews, rust, and
Imaging affects photosynthesis. use in field and cankers.
greenhouse conditions.

CASE STUDY – 1

13
Objectives:

1. Compare the performances of disease detection in the two spectral regions: visible
and near-infrared range (400–1000 nm) and short-wave infrared range (1000–2500
nm).

2. Test the effects of side-view and top-view.

3. Determine optimal input data for machine learning algorithms.

4. Identify the key wavelengths for CR disease detection.

Material and method:

14
 Experimental setup and Flowchart of methodology

Four Australian wheat varieties (Aurora - very low resistance to CR , Yitpi - low
resistance , Emu Rock and Trojan - moderately resistant ) selected for varying resistance to
crown rot (CR). Hyperspectral Imaging system is WIWAM integrated with LemnaTec plant
phenotyping platform. The Cameras used are FX10 for visible/near-infrared (VNIR: 400–
1000 nm) and SWIR for short-wave infrared (1000–2500 nm). Weekly captures (side-view
from 11 DAI, top-view from 29 DAI) and halogen light is used to avoid external light
interference. Manual Disease Scoring is done with the Scale: 0 (no symptoms) to 6 (complete

15
necrosis of the entire leaf sheath), Scored on 14, 21, 28, and 43 DAI and Random sampling
for pathogen isolation at 43 DAI.

Results:

Disease severity levels in four different wheat varieties.

The manual disease scoring results show the disease infection status on 14, 21, and 28
DAI (Fig. 3). On 14 DAI, infected plants only presented mild or no symptoms, so the severity
levels were around 0 to 2. However, by 21 DAI, the manifestation of symptoms in the
infected groups became noticeable, particularly in the case of the Aurora variety where the
median severity value reached 5. Notably, the severity levels of Aurora covered a range from
3 to 6. In contrast, the median severity level of the Trojan and Emu Rock on 21 DAI was
around 3 and the scores were more consistent. On 28 DAI, the infection level had increased
to approximately 4 for the three bread wheat varieties and 5 for Aurora. There was no score
difference in Yitpi on 21 and 28 DAI, with the median values remaining consistent on both
dates and the distribution of infection levels exhibiting similar patterns.

For the top-view, the F1 scores were already over 0.75 under hyper-hue and the SNV
methods measured on 29 DAI with the VNIR camera. Specifically, the SNV had the highest
value over 0.80 from 29 to 36 DAI. After 43 DAI, the F1 score from the four transformation
methods remained similar and stable. In terms of the SWIR camera, only the F1 scores for
hyper-hue and SNV methods were over 0.75 on 29 DAI. Notably, the hyper-hue and the SNV
methods exhibited better performance than the other two methods, as the F1 scores of these
two methods were close to or over 0.80 from 36 to 50 DAI. For the FULL camera, only the
hyper-hue method had a good performance, as the F1 score was around 0.80 from 29 and 43
DAI. For the remaining three transformation methods, their corresponding values were lower
than 0.75 and peaked on 43 DAI.

16
 2C-classification results from the three cameras and two view angles

Key wavelength results for (a) side-view and (b) top-view

The application of feature selection can determine the significant wavelengths and aid
in minimizing data redundancy. The feature selection results of the side-view and top-view
databases showed in graph. They used the reflectance data on key dates (32 and 39 DAI in
side-view as well as 29 and 36 DAI in top-view) to determine the key wavelengths, given the
increase of the F1 score to satisfactory levels on these key dates in the context of 2C-
classification. In the side-view dataset, the results of the two key dates using F-value and chi-
square statistical methods had similar trends. A total of 12 key wavelengths were identified,
including 515, 640, 695, 760, 913, 998, 1196, 1430, 1719, 1935, 2178, and 2284 nm. In the

17
top-view dataset, although the results of the two key dates were similar, however, the results
of the F-value and chi-square statistical methods had different performances on the VNIR
wavelength. The scores using F-value had clear peak points through the whole VNIR
wavelength, while the scores of chi-squares were generally low at the VNIR range.
Therefore, the selection of key wavelengths of VNIR data in top-view was based on the
results of F-value. In the top-view, 12 key wavelengths were selected, including 451, 530,
680, 730, 750, 1002, 1222, 1436, 1700, 1905, 2005, and 2390 nm.

Plant traits related to the key wavelengths.

View Key Literature/ Physiological and Biochemical Traits References

Wavelength Previous Related to Key Wavelengths
(nm) Research
Range (nm)
Topview 451 450–499 Related to the absorption of plant Hennessy et
carotenoids; carotenoids act as al., 2020
photoprotective agents to prevent
harmful photodynamic reactions.
Sideview 515 510 Related to plant carotenoid content. Gitelson et
al., 2002
Topview 530 531 Related to the inhibitory effect of plant Hennessy et
xanthophyll pigments on photo- al., 2020
oxidation damage during
photosynthesis.
Sideview 640 640 Related to plant chlorophyll b content. Curran,
1989
Topview 680 680 Related to the biomass and leaf area Singh et al.,
index of the plant, representing the 2020
green leaf area.
Sideview 695 700 Related to anthocyanin content in plant Gitelson et
leaves; anthocyanins filter yellow and al., 2001
green light to reduce oxidative load.
Topview 730, 750 700–750 Related to red edge regions, Hennessy et
chlorophyll abundance, nitrogen al., 2020
concentration, and water content in
plants.
Sideview 760 760 Related to plant biomass, chlorophyll Feng et al.,
content, and leaf area index. 2017
Sideview 913 910 Related to the C–H stretch (protein Curran,
synthesis). 1989
Sideview 998, 1002, 980–1200 990 nm: O–H stretch (starch synthesis); Curran,
1200 1002 nm: N–H stretch (protein 1989;
synthesis); 1200 nm: O–H stretch
(water, cellulose, starch synthesis);

18
 980–1200 nm: related to plant water Hennessy et
absorption features. al., 2020

Topview 1002 See above for 1002 nm.

Sideview 1196
Topview 1222
Sideview 1430 1410– 1420 nm: Related to C–H Curran,
1430 stretch and deformation 1989;
(lignin synthesis). Lignin is
vital for water transport, Hennessy
mechanical support, and et al.,
resistance to stresses. 1410– 2020
1430 nm: Related to water
absorption features.
Topview 1436
Topview 1700 1700– 1690 nm: C–H stretch Curran,
1750 (protein, starch, and lignin 1989;
synthesis). 1700–1750 nm:
Related to water absorption, Hennessy
lignin, and cellulose et al.,
synthesis. 2020

Sideview 1719
Topview 1905 1900 Related to O–H and C–O Curran,
stretch (starch synthesis); 1989;
also related to leaf liquid
water content. Hennessy
et al.,
2020
Sideview 1935 1940 Related to O–H stretch and Curran,
deformation (water content, 1989
starch, protein, lignin, and
cellulose synthesis).
Topview 2005 2000 Related to O–H and C-O Curran,
deformation (starch 1989
synthesis).
Sideview 2178 2180 Related to N–H bond, C–H, Curran,
C-O, and C-N stretch 1989
(protein synthesis and
nitrogen content).
Sideview 2284 2280 Related to C–H stretch and Curran,
CH2 deformation (starch 1989
synthesis).

19
 Topview 2390 Not previously reported

2C-classification results of using key wavelengths only and the original data.

The study compared the effectiveness of key wavelengths against the original dataset
using SNV and hyper-hue data transformation methods. Results showed that before 32 DAI,
F1 scores were below 0.75 for both key wavelengths and the original dataset. In the VNIR
camera, F1 scores of key wavelengths and original data were similar between 32 and 46 DAI,
but after 53 DAI, SNV on the original data performed better. In the SWIR camera, key
wavelength F1 scores were mostly lower but remained near the acceptable line after 39 DAI.
The FULL camera showed higher F1 scores for key wavelengths than the original dataset
after 32 DAI, especially with SNV, reaching 0.88 at 60 DAI. The top-view VNIR camera had
lower F1 scores for key wavelengths except for hyper-hue on 50 DAI, which was close to the
acceptable line. While SWIR camera F1 scores for key wavelengths were lower, they still
exceeded the acceptable line on 36 and 43 DAI. In contrast, the FULL camera showed

20
consistently higher F1 scores for key wavelengths, especially with SNV, which exceeded the
acceptable line from 29 DAI onwards.

Comparison of Hyper-hue and reflectance on 32 DAI side view. Shows the ‘none’ (left) and
Hyper-hue (right) transformation process results of the hyperspectral image of wheat.

2C-classification results from the three cameras using three different machine learning
algorithms by using key wavelengths.

Conclusion:

Crown rot significantly reduces wheat yield, requiring an efficient screening method
for resistance breeding. Traditional screening is labor-intensive and impractical in the field
due to the lack of early visible symptoms. This study supports hyperspectral imaging (HSI) as
a viable method to differentiate infected and healthy plants by 30 DAI, with top-view

21
imaging suited for field applications. The VNIR camera performed best at 30 DAI, detecting
photosynthetic changes, while the SWIR camera was more effective from 40–60 DAI,
capturing water and nutrient distribution effects. The FULL camera did not improve detection
beyond VNIR or SWIR but showed better performance using key wavelengths. A total of 26
key wavelengths (450–2,400 nm) were identified, suggesting potential for developing cost-
effective multispectral cameras for field use. SVM-based classification distinguished
infection severity but was limited by subjective manual scoring. SNV and hyper-hue
transformations performed best in reducing noise and retaining useful data. However,
limitations include the study being conducted in a controlled greenhouse, the need for field
validation under variable conditions, potential interference from coexisting diseases, limited
sample size for severity estimation, and the absence of abiotic stress factors like drought.
Future research should focus on field trials, larger datasets, and the impact of abiotic stress on
spectral changes.

IoT technology:

In 1999, Kevin Ashton coined the term IoT in the context of supply chain management.
The Internet of Things (IoT) refers to a network of interconnected physical devices that
communicate and exchange data with each other over the internet. It Offers many innovative
solutions, with technologies such as Radio Frequency Identification (RFID), wireless sensor
networks, and cloud computing. The IoT technologies are mainly based on three building

blocks: (i) Sensing component, (ii) Data transfer (that functions as a network) and data
storage, and Manipulation component.

Case Study – 2:

22
Objective: Designing, validating and implementing an IoT-based monitoring system with a
custom agro-weather station (hardware and software) optimized for early warning issuance
for plant disease epidemics (especially those encountered by major crops in Egypt, where this
research is conducted).

Material and method:

The proposed early plant disease forecasting system consists of three main modules:
hardware, software, and agriculture. The hardware module collects environmental data using
sensors and transmits it via a communication system. The software module processes and
stores data, provides a user interface, integrates disease models, and suggests preventive
actions based on data analysis. The agriculture module develops and validates disease models
under various weather conditions through field experiments. The system follows a three-layer
architecture: the front-end layer (hardware and sensors), the middle layer (data
communication and storage), and the back-end layer (AI-based disease prediction and user
interface). This system is versatile and can also be applied to precision agriculture, including
pest control and greenhouse monitoring.

23
 Structure of the AI Module

Agro-weather station (with a closeup view of its controller box) deployed in an agriculture field
in the Agriculture Research Center in Giza, Egypt.

Agro-weather station hardware is implemented and Middle layer implemented with a

key component is a wireless communication mechanism, utilizing the SIM900 GSM
transceiver to enable communication between the station and the host. This transceiver
supports GSM/GPRS functionality (voice, SMS, data, and fax) across multiple frequencies
with low power consumption, controlled via UART using AT commands. The plant disease
forecast agent is a GLAIR agent which is responsible for collecting the sensors’ data from the
agro-weather station (through the data logger), fusing the data over time and space using
BaFes theorem to achieve a reasonable estimate of the actual weather conditions (Khaleghi et

24
al., 2013), and representing said information as graded statements of a formal language over
which logical reasoning can take place.

Results:

Sample of the recorded DSI for the month of July 2015 and Sample of the recorded
environmental data in Beheira governorate.

The field experiments were conducted during 2015, 2016 and 2017 growing seasons. Fig. 11
depicts a sample of the recorded DSI of both crops in the different fields undertest. Weekly weather
parameters such as the maximum/minimum temperature (oC), morning/evening relative humidity
(%), rainfall (mm), rainy days sunshine hours (hr) and wind speed (km/hr) were studied and recorded.
A sample of the collected data collected from Beheira governorate were correlated to the weekly
disease severity index by calculating the Karl person’s correlation coefficient (r).

Y (Temp. From) Z (Temp. To) X = 4 h of RH X = 6 h of RH X = 8 h of RH

27 °C 31 °C Event A detected
23 °C 26 °C Event A detected
18 °C 22 °C Event A detected
Event A in Potato and Tomato early blight forecasting model.

Y (Temp. From) Z (Temp. To) X = 2 h of LW X = 4 h of LW X = 6 h of LW

27 °C 31 °C Event B detected
23 °C 26 °C Event B detected
18 °C 22 °C Event B detected
Event B in Potato and Tomato early blight forecasting model.

25
Y (Temp. From) Z (Temp. To) X = 6 h of RH X = 8 h of RH X = 10 h of RH
20 °C 26 °C Event A detected
13 °C 20 °C Event A detected
10 °C 13 °C Event A detected
Event A in Potato and Tomato late blight forecasting model.

Y (Temp. From) Z (Temp. To) X = 4 h of RH X = 6 h of RH X = 8 h of RH

25 °C 29 °C Event A detected
20 °C 24 °C Event A detected
15 °C 19 °C Event A detected
Event A in Tomato powdery mildew forecasting model.

Sample of the recorded environmental data in during model validation.

26
Potato early blight disease model and Potato late blight and tomato powdery mildew model

Conclusion:

This case study presents an IoT-based monitoring system for epidemic disease control
in precision agriculture. The system is designed to be flexible and adaptable for multiple plant
diseases, with a software architecture that supports various disease models and hardware
sensors that work across different crops. An AI-based expert system mimics human decision-
making for disease prediction and management. The system has been tested on Egyptian
potato and tomato crops, developing disease models based on climatic conditions. Future
applications include precision irrigation and automated fertilizer application to enhance
precision agriculture practices.

Machine Learning and Artificial Intelligence (AI):

The term AI has a bit of history in which takes back to 1956, where Alan Turing
initiated the work of this particular field. He called it “Machine Learning” in a workshop
held at Dartmouth College, USA. Agriculture being one of the most important and ancient
sector in the world, provides basic things that is a necessity for humans. As the population is
rising rapidly the demand for food is also increasing day by day. Avoiding barriers in the path
of production is the best approach for fulfilling the rising demand. The barriers could be
biotic or abiotic. Among the biotic ones, the major constraint is the infection of the diseases
within the field. In order to combat this barrier Artificial Intelligence (AI) could be a great
measure in this field in order to fulfill the demand of the world. These include various types
of systems viz., deep neural networks (DNNs), convolution neural networks (CNNs) and

27
artificial neural networks (ANNs). cabbage, cassava, cantaloupe, celery cherry, corn,
cucumber, eggplant, grapes, onion, orange, pepper, soybean, tomato, peach, etc. thereafter
best performance of the system was recorded to be 99.53 (%).

Working of AI (Machine Learning):

Basically this a prediction model, for this a numerical data is needed to be

incorporated into the system about the severity of the disease or its incidence level. The
DNNs can interpret the values in input layer of the system and then to output layer, produces
the accurate nodal values in continuous chains. To produce accurate photographs/images
properdatasets are needed that has to standardize. This can be done under following headings;

1. Data collection: this includes capturing of photographs of plant parts via live cameras or
mobile phones.

2. Data preprocessing: it is the second and important stage. These data should be collected
from large number of sources like physical devices, software, tools, etc. It basically improves
the effectiveness of the machine learning, normally done by cleaning and normalizing
pictures dimension.

3. Data segmentation: plays a vital role in the recognition and characterization of various
diseased samples/crops. It detects the content of a picture and also enables us to do image
analysis.

4. Feature extraction: in a hierarchal manner the photographs are taken into the observations
and it also identifies the parts of an object. In disease identification, color, texture and shape
play an important role.

5. Classification: the software named as ‘softmax’ is basically used to classify the plant
diseases separately.

Basic uses of AI in Plant Pathology:

Diseases in the crops are a matter of great grave concerns, as this could cause a lot of
yield and economical losses. Systems aided with the computers and internet are gaining
extreme importance as detection and diagnosis is becoming easy day by day, saving the yield
and wealth of the farmers. Below are given some of the important uses where AI are used
extensively;

28
 1. Disease detection and diagnosis: AI integrated with the DNNs helps to identify
the diseased plants. Promising results are being produced by the computerized
imaging system but also needs an expertise for understanding the data.
2. Sensor based monitoring system: spectral data is collected using these sensors
which are fitted by the hyper spectral cameras. It minimizes the external
disturbances hence giving clear and accurate image/data. These sensors are fitted
into the field and give the data 24x7.
3. Target spraying through drones: also known as UAVs (Unmanned Aerial
View) are fitted with the GPS (Global Positioning System), real time image
transmission systems, multiple sensors, wireless communications, automatic flight
control and other accessories. Drones can capture high resolution images from
aerial view making it easier to collect spatial data from any place. They can also
conduct routine rounds/patrol for any abnormality entering into the field. Among
all, German company which manufactures drones has become very famous for the
pesticide spray.

Diagrammatic representation of AI-enabled diagnostics of plant diseases.

Advantages of AI in Plant Pathology:

 It will help farmers to increase the yield by assisting them to manage the diseases by
giving proper guidelines.
 Identification of diseases and its management within friction of seconds is possible.
 It also suggests the pesticide or botanicals to be used in managing that particular
disease.

29
  It may reduce human error.

Disadvantages of AI in Plant Pathology:

 AI models are dependent on the quality and quantity of datasets.

 It will be very much expensive, especially in the regions/countries which are still
developing and in countries where this technology is newly introduced.
 Data can be misused by anyone, when it is communicating using internet phishing can
be a major issue.
 New models may not be competent with new emerging threats.

Key challenges in the adoption of AI technology:

 Lack of understanding of AI technology. If the technology is introduced newly then

for sure there will be some problem in the adoption process.
 Large expense for the installation initially. Any technology with excellent working
efficiency and accuracy will be expensive. Hesitation of people in that particular place
to adopt new technologies.
 The process may take longer time to be installed or when used initially. In short, a
lengthy adoption process has to be followed with proper protocols.
 Technological limitations may also occur, as the technology is still developing round
the globe.
 Privacy and security issues may be faced by the people in the process of adoption.

CASE STUDY - 3

30
Objectives:

1. Determining the most important factors in MDM forecasting by examining the

interactions between host, pathogen, and weather.

2. Developing machine-learning-based models for predicting MDM occurrences,

3. Developing an intuitive web application for MDM prediction.

Material and method:

Web Application Development

31
Results:

Temporal patterns of MDM disease progression over 48 disease cycles

Monitoring of per cent disease incidence (PDI) and incubation period (IP) over four
years provided insights into the temporal dynamics of MDM disease. The IP varied from 12
to 30 days, with the shortest IP (12 days) observed in September 2019, November 2019, and
October 2020, while the longest IP (30 days) occurred in June 2019, indicating differences in
disease progression. PDI fluctuated, peaking in September (2019–2022) with values between
66.3% and 88.8%, highlighting high disease susceptibility during this period. The lowest PDI
(~19%) was recorded in May, suggesting reduced disease pressure. The combination of
shorter IP and high PDI in certain months signified rapid disease spread, whereas longer IP
with lower PDI indicated a window for proactive disease management. Weather parameters,
including temperature and humidity, played a key role in influencing disease occurrence,
emphasizing the importance of seasonal monitoring for effective MDM control.

32
 Temporal dynamic of grain yield over 48 disease cycles

Grain yields were recorded from October 2018 to September 2022, revealing
significant variations due to MDM disease. The highest yields were observed in May (18.3–
19.8 q/ha), followed by April (17.1–19.6 q/ha) and March (15.4–19.0 q/ha), making summer
the most favorable season for grain production. In winter, yields were moderate, with January
(10.3–15.7 q/ha) and February (14.2–17.3 q/ha) showing better performance than December
(9.6–14.3 q/ha) and November (8.8–10.8 q/ha). The lowest yields occurred in the late rainy
season, with August (8.3–12.3 q/ha) and September (4.0–7.4 q/ha) exhibiting the most severe
reductions, highlighting the impact of MDM disease. Effective disease management
strategies are essential to mitigate yield losses, particularly in high-risk months like August
and September.

Tripartite interaction

From 2018 to 2022, temperature (TMAX) and relative humidity (RHMAX) were key
factors influencing MDM disease incidence. Lower TMAX and higher RHMAX correlated
with increased PDI, with September showing the highest PDI (80.6%) at TMAX 28.4°C and
RHMAX 92.9%. May had the lowest PDI (19.6%) with the highest TMAX (35.3°C),
indicating an inverse relationship between TMAX and disease incidence. The findings
highlight the importance of weather parameters in MDM forecasting and management.

33
Implications of disease forecasting and plant protection measures on maize yield and visualizing
the impact of disease forecasting and plant protection measures on maize.

Values of feature importance in MDM forecasting and Comparitive analysis of ML algorithm on

MDM forecasting

34
 Timely application of plant protection measures significantly reduced MDM disease
incidence and improved yields across different seasons. In 2021, protected treatments led to
yield increases of 60.0% in the rainy season, 162.5% in the late rainy season, 43.8% in
winter, and 24.2% in summer compared to control treatments. Similar trends were observed
in 2022, with the highest yield increase (192.0%) in the late rainy season. On average, yield
gains were 61.6% in the rainy season, 47.1% in winter, and 20.5% in summer under
protection. Using disease forecasting models, farmers can implement proactive disease
management strategies, especially in high-risk months like September, to maximize yield.

Feature importance (FA) analysis identified TMax as the most critical factor (76%) in
MDM disease forecasting, followed by RHMIN (15%). Higher TMax was associated with
lower disease incidence, emphasizing the need for temperature-based management strategies.
Among machine learning models, Random Forest achieved the highest accuracy (R² = 0.97)
with the lowest error rates (MAE = 3.78, MSE = 11.83, RMSE = 3.44). Bagging Regressor
and XGBoost followed closely, while KNN, Decision Tree, and SVM performed poorly.
Random Forest also had the lowest MAPE (9.09%) and sMAPE (8.65%), making it the most
reliable forecasting model. Its learning curve showed rapid accuracy improvement with
increased training data. The findings underscore the effectiveness of advanced ML techniques
in precise disease prediction.

S. TMAX TMIN RHMAX RHMIN Rainfall Actual Predicted

No. (℃) (℃) (%) (%) (mm) PDI PDI
1 29.5 19.2 90.2 78.4 3.2 75.1 73.7
2 30.6 19.1 91.1 77.3 0.0 61.1 63.6
3 31.4 18.7 85.2 77.7 0.1 45.5 43.9
4 31.5 17.0 83.5 78.7 0.0 36.5 42.1
5 31.8 19.3 81.1 73.0 0.0 29.5 34.0
6 34.5 20.8 76.5 65.6 0.0 22.5 24.7
7 34.9 22.2 77.4 59.0 1.0 22.5 25.4
8 35.5 23.5 74.2 53.1 4.3 20.6 20.2
9 33.6 22.8 74.6 55.7 1.3 19.9 23.8
10 33.0 21.5 83.0 61.3 0.5 29.9 30.7
11 28.9 18.7 87.2 79.4 3.9 59.5 60.3
12 27.9 19.0 93.7 84.5 4.3 88.8 87.1
13 28.0 18.8 91.0 77.1 10.7 72.5 73.2

35
 14 30.9 18.2 90.6 75.2 0.6 69.5 65.6
15 30.1 17.3 86.0 77.0 0.1 52.1 56.0
16 30.2 21.2 82.5 62.1 0.3 53.2 51.9
17 31.3 22.0 80.5 56.7 0.0 42.5 44.9
18 33.2 21.3 80.2 35.8 0.0 30.2 33.3
19 34.9 24.2 78.0 53.0 1.6 25.1 26.5
20 35.1 23.5 75.2 62.5 6.6 19.5 20.3
21 29.6 23.9 82.6 66.5 2.2 59.1 55.7
22 28.7 24.5 82.6 72.3 2.6 62.9 59.7
23 28.8 23.9 85.7 68.9 0.9 69.5 67.3
24 27.8 23.7 92.6 68.9 5.6 75.2 79.8
25 29.7 23.6 92.5 68.5 6.8 63.2 62.9
26 28.3 24.3 91.7 85.8 1.3 62.5 69.8
27 28.6 22.5 87.9 63.4 0.0 59.5 66.9
28 29.4 21.1 84.3 61.4 0.4 52.1 54.5
29 31.2 19.5 84.0 42.4 1.3 42.6 47.9
30 33.2 22.0 79.3 42.5 0.0 35.2 35.3
31 34.2 24.7 78.2 42.5 2.7 29.5 27.7
32 35.2 23.0 76.5 48.5 6.2 19.2 20.4
33 29.2 24.1 85.5 52.6 1.9 59.5 59.4
34 28.0 24.7 86.9 62.5 2.0 62.5 69.5
35 30.1 23.9 91.2 78.5 5.4 52.4 56.9
36 29.4 23.4 93.5 79.5 4.7 78.9 70.3
37 28.6 23.6 91.6 68.0 5.1 71.7 69.4
38 28.2 24.8 90.5 88.5 1.3 66.3 75.1
39 28.3 22.2 87.9 61.0 0.0 65.2 73.6
40 29.8 16.2 84.2 56.7 0.0 59.2 57.6
TMAX = Temperature Maximum; TMIN = Temperature Minimum; RHMAX = Relative Humidity
Maximum; RHMIN = Relative Humidity Minimum; PDI = Percent Disease Incidence.

Weather parameters and MDM Incidence : Insights from JASS predictive analysis

36
 Relationship between actual and predicted values of the PDI

Conclusion:

This study highlights the effectiveness of machine learning in forecasting maize

downy mildew (MDM). Using the Random Forest algorithm, the model achieved a high
predictive accuracy (R² = 0.97), identifying maximum temperature (TMAX) as the most
critical factor in disease prediction. The JASS web application, developed for real-time MDM
risk assessment, empowers farmers to take early preventive measures, reducing crop losses
and improving resource efficiency. By integrating ML into an accessible tool, this study
demonstrates its potential to enhance maize yields and support global food security. The
JASS application is available at [Link]

Mobile Applications:

List of mobile applications used in Plant Pathology

App Name Country of Free/Paid Download Platform

Origin s

Pestoz Identify Plant Diseases India Free 10K+ Android

AgroAI—Plant Diseases Diagnosis (Early Africa Free 10+ Android

Access)

Cropalyser Netherlands Free 10K+ Android

PDDApp: Plant Disease Detection Russia Free 1K+ Android

Leafy India Free 100+ Android

PlantDoctor India Free 1K+ Android

Plantix–Your Crop Doctor Germany Free 100K+ Android

Plant Disease Detector Unknown Free 10+ Android

Riceye Unknown Free 5+ Android

Cassava Plant Disease Identify Unknown Paid – iOS

Plants Disease Identification Unknown Paid – iOS

37
 Garden Plants Diseases Detector Unknown Paid – iOS

Plant Disease Identifier Unknown Paid – iOS

Plant Diseases and Pests Unknown Paid – iOS

PlantifyDr United States Free 10+ Android and iOS

Leaf Doctor United States Free 10K+ Android and iOS

Agrio United States Free 100K+ Android and iOS

MOBILE APPLICATIONS DEVELOPED BY FARM UNIVERSITIES OF KARNATAKA:

UAS, BANGLORE

Mobile application [Link] downloads

Sugarcane - IPM 10K+

Arecanut - IPM 5K+

Banana - IPM 5K+

Grapes - IPM 5K+

Beej Aadhar 5K+

Pomegranate - IPM 1K+

Tomatoes - IPM 1K+

Mango - IPM 1K+

Papaya - IPM 1K+

Peanut - IPM 1K+

Eggplant - IPM 1K+

Krishimela 2019 1K+

Chillies - IPM 1K+

Corn - IPM 1K+

Potatoes - IPM 500+

Ekrishi - IPM 500+

Toor Dal - IPM 500+

Pulse - IPM 100+

Sunflower - IPM 100+

Gram - IPM 100+

Finger Millet – IPM 100+

UAS, DHARWAD

Havaamaana Krishi 10K+

UAS, RAICHUR

e-SAP 10K+

38
 Popular apps in India (Android) and IOS

Advantages of Mobile Applications :

• AI algorithms analyze the photographs and provide instant results.

• Critical information on symptoms, triggers, chemical, and biological treatments is

shared.

• Geo-tagging of images for real-time monitoring of pests and diseases.

• Creation of high-resolution maps showing the spatial distribution of crops and

significant plant diseases.

• Weather information tailored to farmers' locations.

• Community feature for interaction among stakeholders in plant protection.

• Provides free, real-time diagnosis through a user-friendly dashboard in local

languages.

39
 • Offers easy-to-understand diagnostic suggestions and monitoring across regions.

Plantix Application:

Agriculture is a key sector of the Indian economy, contributing 16% to GDP and
employing 60% of the workforce. However, challenges such as pest infestations,
unpredictable monsoons, exploitative moneylenders, and low prices from intermediaries
hinder productivity. Globally, plant diseases and pests destroy 20–40% of crops annually.
Digital agriculture, supported by widespread smartphone usage, offers solutions by providing
real-time data to farmers.

Collaboration among researchers, technology firms, and governments has improved

disease detection, achieving 85% accuracy. The Plantix app is an B2C approach powered by
the Strey‟s Berlin-based startup developed by PEAT GmbH in partnership with ICRISAT,
uses machine learning and image recognition to diagnose crop diseases, pests, and nutrient
deficiencies. With a database of 500 diseases across major crops, the app has been
downloaded 12 million times, mainly in India, and is available in eight local languages. This
digital innovation empowers farmers with timely insights to mitigate crop losses and improve
yields.

CASE STUDY - 4

Objective: To assess the detection of BLB disease using a remote sensing method, aiming to
enhance agricultural productivity by developing an assessment method through textural
analysis using conventional UAVs RGB sensors to improve the precision of BLB assessment.

40
Material and method:

The workflow commences with field data collection, encompassing aerial data
obtained from drones and terrestrial data gathered by pest observers. Drone-acquired data are
processed for image stitching and rectification, whereas terrestrial data are organized into a
sample dataset with unique identifiers (IDs). These IDs facilitate the integration of data with
digitized sample points derived from the visually digitized orthomosaic drone data. Outputs
from image stitching include orthomosaic images, vegetation indices, and thermal images.
Orthomosaic images were further processed to extract Haralick textural features and NDTI.
Subsequently, zonal statistics were applied to integrate the vegetation indices and thermal
data, yielding a dataset suitable for machine learning applications. The dataset underwent
synthetic data generation and the synthetic minority oversampling technique (SMOTE) to
achieve a balanced dataset. Machine learning models were trained and tested using the
exhaustive feature selection method to identify an optimal subset of features based on various
accuracy metrics.

Sensors used in the present study and a) Quantum- Systems Trinity F90 b) MicaSense Altum-PT

Features prepared for machine learning

Feature Feature Source Number Formula Denoted

Type of As
Features
Haralick Haralick 1 R_B8
Textural correlation from
Features red band
Cluster shade 1 G_B6
from green band
NDTI 1 NDTI=(T1−T2)/(T1+T2) NDTI
Vegetation NDVI 1 NDVI=(NIR−Red)/(NIR+Red) NDVI
Index
NDRE 1 NDRE=(NIR−Red NDRE

41
 Edge)/(NIR+RedEdge)
GNDVI 1 GNDVI=(NIR−Green)/(NIR+Green) GNDVI
Thermal Temperature 1 temp_1
from morning
shoot
TVDI from 1 TVDI=(T−Tmin)/(Tmax−Tmin) tvdi_1
morning shoot
Temperature 1 temp_2
from afternoon
shoot
TVDI from 1 tvdi_2
afternoon shoot
Total 10

Results:

Maps of classification models applied to two testing locations and utilized all textural, thermal,
VIs, and all features.

Obtained favorable results when testing our model using textural features with our
existing UAV imagery. Both the RF and NN models yielded outcomes that closely resembled

42
the observations made by the POs, although at Location 1, overestimation was observed for
both RF. When applying the models using thermal features, unsatisfactory results were
obtained. While RF produced outcomes nearly identical to those of the current method at
Location 1, NN showed overestimation at the same location. At Location 2, both the RF and
NN resulted in an underestimation compared with the current method. The application of the
VI-based model yielded promising results, as shown in Fig. 25. At Location 1, both the RF
and NN models produced classification results that were closely aligned with the benchmark,
demonstrating an absolute error of approximately 1 %. However, at Location 2,
underestimation was observed, with RF and NN exhibiting absolute errors of approximately 7
% and 3 %, respectively. By integrating all the features, they obtained results that closely
resembled those obtained from the textural feature-based model. Notably, the NN algorithm
classifies all pixels into the same class, resulting in a uniform classification outcome. In
contrast, RF exhibited a broader range of classes, with absolute errors of approximately 7 %
and 2 % at Location 1 and Location 2, respectively.

Cost comparison between conventional RGB, multispectral, and thermal UAV based on
two use-case scenarios.

Used Conventiona Enterprise- Enterprise Percentage Percentage Source

Case l UAV with Level UAV -Level Streamline Streamline
Scenari RGB Sensor with UAV with d (a vs b) d (a vs c)
o (a) (USD)* Multispectra Thermal (%) (%)
l Sensor (b) Sensor (c)
(USD) (USD)

Mappin 300 900 1800 66.7 83.3 Indonesia

g local drone
Service mapping
(for 10 service
ha area) provider

Purchas 1599 9100 25,900 82.4 93.8 [Link]

e m

Conclusion:

43
 This study explores the use of textural features from drone data to assess bacterial leaf
blight (BLB) damage in paddy fields. The integration of the Normalized Difference Texture
Index (NDTI) significantly improved classification accuracy, with the Random Forest (RF)
algorithm outperforming Neural Networks (NN). Despite limitations, this approach offers a
cost-effective solution for resource-constrained farmers in precision agriculture. The study
demonstrates that textural features from standard RGB drone data can classify BLB more
accurately than traditional vegetation indices (VIs). This method eliminates the need for
costly multispectral or thermal sensors, making it accessible to farmers with limited
resources. The use of remote sensing enables large-scale, non-destructive monitoring,
improving disease detection and management.

Augmented Reality (AR) and Virtual Reality (VR) in Plant Pathology:

Augmented Reality (AR) and Virtual Reality (VR) are innovative tools increasingly
being used in plant pathology for research, education, disease diagnosis, and training.
Augmented reality, in computer programming, is a process of combining or “augmenting”
video or photographic displays by overlaying the images with useful computer-generated
data. Specifically, AR feeds the input data from the live environment to the application,
which automatically presents them to the eyeglass display. The end user automatically
receives the supervised information without any other need to search for information. AR
wearable glass makes the farmers to view the virtual portion of the field with their naked eye
and also provides flexibility to operate by their hands. It serves as a piece of distinct hardware
equipment for the farmers. Combining IoT with AR deploys live environment makes
superimposed IoT data into physical objects and delineation of such information. A
supporting computer graphics system is developed to visualize direction for farming
operations and overlay them on a field.

44
 Block diagram of augmented reality cloud processing system

Smart Glass: Real-Time Leaf Disease Detection using YOLO Transfer Learning:

Prototype of the proposed Smart Glass and Test results from real-time footage
(Blue=Healthy; Red=Unhealthy)

Methodologies:

A. Software implantation:

1. Dataset Creation and Data Collection: A custom dataset of 304 images was collected by
visiting agricultural lands. The images of tomato plants were classified as 'Healthy' and
'Unhealthy' based on parameters such as camera quality, zoom level, exposure, and the
number of leaves in the frame.

2. Image Labeling and Preprocessing: The images were resized to 416x416 resolution, and
custom labeling was done using Python's Labeling software. Annotations were stored in
YOLO format, including coordinates, width, height, and class for each image.

3. Model Training and Implementation: The YOLO algorithm, using pre-trained YOLOv3
weights, was implemented in the Keras library. The trained model was used for real-time leaf
detection and classification, with the system capable of localizing and classifying images in
real-time.

B. Hardware implementation:

45
4. Camera Module and Live Streaming: The Smart Glass uses a camera module to monitor
the physical environment, capturing video at 1080p. The live stream is transferred to the
Raspberry Pi Zero W (RPiZW) for processing.

5. Real-Time Data Processing: The live stream is sent from the microcontroller to a local host
using Motion software. At the local host, OpenCV is used to capture the stream, and a trained
model performs real-time classification of the data.

6. Display Output: The processed data is transferred to an LCD through the RPiZW
microcontroller for visual output.

Robotics:

Robots could act as very experienced pathologists that recall thousands of images to
deliver a diagnosis. Machine learning, machine vision, and AI have been utilized and applied
in agriculture for automated and robotic farming. The vision-based systems can be mounted
on mobile vehicles (e.g., autonomous vehicle) to optimize disease detection procedures and
disease management. Agricultural robotic systems and tools may utilize hazardous materials
and chemicals, which could increase costs and environmental impact for their disposal.

Many recognition methods and technologies for identifying plant disease symptoms have
been successfully developed; still, the majority of them require a controlled environment for
data acquisition to avoid false positives. In recent years, research of on-site detection assay is
growing fast and methods such as Loop-Mediated Isothermal Amplification (LAMP) or the
lab-on-chip based on Electrical Impedance Spectroscopy (EIS) may permit fast molecular
diagnosis in the field. Because of their speed, robustness, and simplicity, these tests should be
integrated in traditionally vision-based robots, enhancing their diagnostic specificity when
required.

46
Lab-on-a-drone components: (a)The convective thermocycling with a single heater, for PCR
Reactions. (b) Instrument assembly with available components (c) A smartphone camera is used for
fluorescence detection of reaction products. (d) the entire assembly, characterized by lightweight (e)
Successful in-flight replication of two different DNA targets

Genomic Tools:

Genomic tools have revolutionized plant pathology by enabling rapid and precise
identification of plant pathogens. Next-generation sequencing (NGS) technologies allow
researchers to sequence entire genomes of pathogens, providing insights into their genetic
diversity, evolution, and virulence factors. Whole-genome sequencing (WGS) and
metagenomics approaches help detect and characterize pathogens directly from infected plant
tissues without the need for prior culturing. This advancement significantly enhances disease
diagnosis and surveillance, allowing early detection and targeted management strategies.

In addition to sequencing, digital innovations in genomic tools include high-throughput

functional genomics techniques such as CRISPR-Cas9 gene editing. This tool helps in
understanding host-pathogen interactions by enabling precise modifications in plant defense
genes or pathogen virulence genes. Comparative genomics and transcriptomics further
provide a deeper understanding of how pathogens overcome plant immune responses, guiding
the development of disease-resistant crop varieties. Coupled with bioinformatics and artificial
intelligence (AI), these genomic datasets can predict potential pathogen outbreaks and
resistance breakdown, enhancing disease forecasting models.

Furthermore, digital platforms integrating genomic data with remote sensing and
environmental parameters improve disease monitoring at a larger scale. Cloud-based
databases and online repositories facilitate global collaboration by enabling researchers to

47
share genomic data and identify emerging threats in real-time. Portable sequencing devices,
such as Oxford Nanopore MinION, allow field-based pathogen detection, reducing the time
lag between sampling and diagnosis. The integration of genomic tools with digital
technologies is transforming plant pathology, leading to more efficient disease management
and safeguarding global food security.

Digital platforms used for disease surveillance in plants:

1. PlantVillage Nuru - Uses AI to detect plant diseases through smartphone images.

Provides real-time advice for disease management.
2. Global Biodiversity Information Facility (GBIF) - Tracks the spread of plant pests and
pathogens using biodiversity data.
3. CIMMYT’s Pest and Disease Surveillance Tool - Helps monitor and predict the
spread of pests and diseases in crops like wheat and maize.
4. Agroclimate - A decision-support system integrating weather data and disease models
to predict outbreaks.
5. FAO’s FAMEWS (Fall Armyworm Monitoring and Early Warning System) - Tracks
the spread of Fall Armyworm globally through community data reporting.
6. Crop Disease Monitoring Networks - Tracker for wheat rust diseases, which uses GIS
mapping for disease spread predictions.

In India:

1. e-National Pest Reporting and Alert System (e-NPRAS) - Developed by the Ministry
of Agriculture, this platform helps monitor and alert farmers about pest and disease
outbreaks.

2. ICAR-FUSICONT - A mobile app developed by ICAR for real-time monitoring and

management of Fusarium wilt in crops like bananas.

3. Pest Surveillance System of Tamil Nadu - Provides early warnings and advisories to
farmers based on pest and disease monitoring in the state.

4. Kisan Suvidha App

5. Crop Pest Surveillance and Advisory Project (CROPSAP) - Maharashtra government

initiative for monitoring and managing crop pests and diseases like Pink Bollworm in
cotton.

48
 6. Plantix

In Karnataka:

1. Yuktix GidaBits - Early warning system that utilizes IoT technology and data analysis
to predict weather-related plant diseases. It provides farmers with advance alerts and
management advice through a mobile application.
2. e-SAP (Electronic Solutions against Agricultural Pests)
3. Karnataka’s Digital Agriculture Initiatives - The state has embraced digital agriculture
by integrating technologies like GPS systems, sensors, and drones. These tools collect
data on soil moisture, temperature, and nutrient levels, enabling farmers to monitor
crop growth and apply inputs more efficiently.

Future Prospects:

The future of digital innovations in plant pathology is expected to be driven by

advancements in artificial intelligence (AI), machine learning (ML), and big data analytics.
AI-powered diagnostic tools will enhance early disease detection by analyzing images from
drones, smartphones, and remote sensing platforms in real time. ML algorithms will improve
predictive disease models by integrating genomic, environmental, and climatic data, enabling
proactive disease management. Automated robotic systems and smart sensors will further
enhance precision agriculture by continuously monitoring plant health and applying targeted
treatments, reducing pesticide use and improving sustainability.

Genomic advancements, including CRISPR-based diagnostics and portable

sequencing technologies, will revolutionize pathogen detection and disease resistance
breeding. Rapid, field-deployable sequencing tools will enable real-time identification of
emerging plant pathogens, facilitating faster responses to outbreaks. AI-assisted genomic
analysis will aid in identifying novel resistance genes, accelerating the development of
disease-resistant crop varieties. Blockchain and cloud-based platforms will enhance global
collaboration by securely storing and sharing genomic and epidemiological data, improving
disease surveillance and outbreak management.

In the long term, integrating digital innovations with climate-smart agriculture will
play a crucial role in addressing plant health challenges under changing environmental
conditions. Internet of Things (IoT)-enabled smart farming systems will provide real-time
plant health monitoring, optimizing resource use and reducing losses due to plant diseases.

49
The use of digital twins—virtual models of plants and ecosystems—will enable researchers to
simulate disease spread and evaluate control strategies before field implementation. As digital
technologies continue to evolve, they will drive sustainable plant disease management,
ensuring global food security and resilience against future agricultural challenges.

Conclusion:

Digital innovations in plant pathology have revolutionized disease detection,

monitoring, and management by integrating advanced technologies such as artificial
intelligence, remote sensing, genomic tools, and big data analytics. These innovations enable
rapid and accurate pathogen identification, real-time disease surveillance, and predictive
modeling, allowing for proactive decision-making in crop protection. By leveraging drones,
sensors, and cloud-based platforms, researchers and farmers can monitor plant health on a
large scale, reducing crop losses and improving agricultural sustainability.

The incorporation of genomic tools, machine learning algorithms, and digital disease
databases has significantly enhanced our understanding of host-pathogen interactions and
resistance mechanisms. Technologies like CRISPR-based gene editing and portable
sequencing devices facilitate faster and more targeted disease control strategies. Additionally,
AI-driven diagnostics and mobile applications empower farmers with accessible tools to
detect and manage plant diseases efficiently, even in resource-limited settings.

Overall, digital innovations are transforming plant pathology into a more precise, data-driven
discipline, ensuring sustainable crop production and global food security. Continued
advancements in these technologies, coupled with interdisciplinary collaboration, will further
enhance disease management strategies, making agriculture more resilient to emerging plant
health threats.

REFERENCES:

Abbas A, Zhang Z, Zheng H, Alami M M, Alrefaei A F, Abbas Q, Naqvi S A H , Rao M J,

Mosa W F, Abbas Q and Hussain A, 2023, Drones in plant disease assessment,
efficient monitoring, and detection: A way forward to smart
agriculture. Agronomy, 13(6), pp.1524.

Ampatzidis Y, De Bellis L and Luvisi A, 2017, iPathology: robotic applications and

management of plants and plant diseases. Sustainability, 9(6), pp.1010.

50
Brenchley G H, 1968, Aerial photography for the study of plant diseases.

Buja I, Sabella E, Monteduro A G, Chiriacò M S, De Bellis L, Luvisi A and Maruccio G,

2021, Advances in plant disease detection and monitoring: From traditional assays to
in-field diagnostics. Sensors, 21(6), pp.2129.

Charaya M U, Upadhyay A, Bhati H P and Kumar A, 2021, Plant disease forecasting: Past
practices to emerging technologies. Plant Disease: Management Strategies, pp.1-30.

Dewanto S and Lukas J, 2014, Expert system for diagnosis pest and disease in fruit plants.
In EPJ Web of Conferences (Vol. 68, p. 00024). EDP Sciences.

Hasanaliyeva G, Si Ammour M, Yaseen T, Rossi V and Caffi T, 2022, Innovations in disease

detection and forecasting: a digital roadmap for sustainable management of fruit and
foliar disease. Agronomy, 12(7), pp.1707.

Jadesha G, Castelino E, Mahadevu P, Kitturmath M S, Lohithaswa H C, Karjagi C G and

Deepak D, 2025, Smart solutions for maize farmers: Machine learning-enabled web
applications for downy mildew management and enhanced crop yield in
India. European Journal of Agronomy, 164, pp.127441.

Khalkho A, Jain A K, Bhatt J, Nema S, Shyam M, Bisht K and Sahu A, 2024, Use of artificial
intelligence in plant pathology. International Journal of Plant Pathology and
Microbiology, 4(1), pp. 23-27.

Khattab A, Habib S E, Ismail H, Zayan S, Fahmy Y and Khairy M M, 2019, An IoT-based

cognitive monitoring system for early plant disease forecast. Computers and
Electronics in Agriculture, 166, pp.105028.

Kuska M T, Heim R H, Geedicke I, Gold K M, Brugger A and Paulus S, 2022, Digital plant
pathology: A foundation and guide to modern agriculture. Journal of Plant Diseases
and Protection, 129(3), pp.457-468.

Nath R K, Begum K H and Choudhury M R, 2020, Plant disease forecasting models. Indian
Journal of Pure Application of Biosciences, 8(4), pp.454-61.

Ponnusamy V, Coumaran A, Shunmugam A S, Rajaram K and Senthilvelavan S, 2020, Smart

glass: real-time leaf disease detection using YOLO transfer learning. In 2020
International Conference on Communication and Signal Processing (ICCSP), pp.
1150-1154.

51
Samal I, Bhoi T K, Pradhan A K and Mahanta D K, 2023, Plantix app: A success story of
artificial intelligence in plant protection. Van Sangyan, 10, pp.24-44.

Wijayanto A K, Prasetyo L B, Hudjimartsu S A, Sigit G and Hongo C, 2024, Textural features

for BLB disease damage assessment in paddy fields using drone data and machine
learning: Enhancing disease detection accuracy. Smart Agricultural Technology, 8,
pp.100498.

Xie Y, Plett D, Evans, M, Garrard T, Butt M, Clarke K and Liu H, 2024, Hyperspectral
imaging detects biological stress of wheat for early diagnosis of crown rot
disease. Computers and Electronics in Agriculture, 217, pp.108571.

52
53