Artificial Intelligence and Machine Learning Based Plant Monitoring

Computer Science & Engineering: An International Journal (CSEIJ), Vol 15, No 1, February 2025
DOI:10.5121/cseij.2025.15115 129
ARTIFICIAL INTELLIGENCE AND MACHINE
LEARNING BASED PLANT MONITORING
Shrutika C Rampure, Bhavyashree S P, Rakshitha B T
Department of Computer Science and Engineering, Acharya Institute of
Technology, Bangalore, India
ABSTRACT
One of the main kinds of life in the world is plant. By giving food to individuals and, what's
more, untamed life, plants benefit the environment and human existence in the world in
different ways. Plants lead to a country's beneficial farming creation. A convolutional
neural network (CNN) can be utilized to future plant development and wellbeing through
leaf assessment. AIML had played a fundamental impact on checking plants' wellbeing.
AIML grants objects to be detected or controlled from a distance across the organization's
foundation. The outcome further develops exactness, financial advantages and productivity.
AIML framework is intended to gather information and give continuous input on the
condition of the plant, soil, and climate factors. Acknowledgment of plant diseases utilizing
Convolutional neural Networks (CNN) is an emerging field of exploration that expects to
recognize and analyse plant infections naturally. This method utilizes picture based
calculations in light of profound realizing, which take into consideration the extraction of
complex highlights from pictures of plant leaves, natural products, or stems impacted by
different diseases. via preparing the CNN on an enormous dataset of solid and sick plants,
it can figure out how to recognize designs and recognize various kinds of diseases. The
acknowledgment of plant infections involving CNN has huge consequences for plant
diseases across the board, as it takes into consideration early discovery and exact analysis,
which can prompt ideal interventions and decreased crop misfortunes. In this paper, we
will give an outline of the idea of acknowledgment of plant diseases utilizing CNN, its
expected applications, and some of the difficulties that should be addressed to work on the
precision and versatility of this technology.
KEYWORDS
Plant leaf disease detection, Visualization, Internet of Things (IoT), Convolutional neural
networks, Image classification.
1. INTRODUCTION
Rural efficiency, a foundation of worldwide food security and financial strength, faces huge
difficulties because of biotic burdens, prompting significant yield misfortunes. The relationship
between food handling, sustenance, and rural economy frames a complicated transaction,
especially influencing creating an immature nation, compounding wellbeing and monetary
emergencies. Remarkably, in locales like Africa, where roughly 80% of harvests are developed
by smallholder farmers, assets requirement represents a huge danger to horticultural creation.
Plant infections address an imposing hindrance to horticultural supportability. Convenient
recognition and the board of these diseases are basic to moderate food weakness and monetary
misfortunes. Generally, diseases-distinguishing proof has depended intensely on emotional
evaluations by horticultural specialists or farmers, a strategy burden with shortcomings,

130
mistakes, and ecological dangers. To address these difficulties, there is a developing interest in
utilizing mechanical progressions, especially in picture handling and the Web of Things (IoT), to
upset plant diseases acknowledgment and observing. Picture handling methods offer a promising
road for computerizing diseases as recognizable proof by breaking down obvious signs, for
example, sores or stains on plant leaves. All the while, the expansion of IoT gadgets open new
doors for continuous observing and information collection in farming settings. By using sensors
to quantify boundaries like humidity, temperature, and moisture content, combined with cutting
edge man-made reasoning and AI calculations, like Convolutional Neural Networks (CNNs), it
becomes practical to precisely predict and analyze plant diseases. The union of these
advancements in the proposed project, named "AIML Based Plant Monitoring," means to foster
an exhaustive answer for checking and overseeing plant wellbeing. Through the coordination of
sensor information procurement, picture handling, and man-made intelligence driven diseases
detection, the undertaking tries to engage farmers with ideal and significant experiences,
consequently improving horticultural efficiency, and adding to worldwide food security. In this
paper, we outline the reasoning, system, and expected results of the undertaking, highlighting its
importance in tending to basic rural difficulties and encouraging maintainable cultivating
rehearsals.
2. LITERATURE REVIEW
In recent years, the utilization of artificial intelligence and machine learning practices in
agriculture, notably in plant diseases detection and monitoring, has drawn a considerable amount
of focus point. This review consolidates discoveries from four unique research documents, each
adding valuable understandings into the realm of AIML - based plant monitoring. The first paper
[1] accentuates using machine learning algorithms, predominantly Convolutional Neural
Networking (CNN), for recognizing and sorting crop diseases. In addition, the success of
hyperspectral sight and deep learning methods in plant disease detection is emphasized in this
paper. Nevertheless, crucial shortcomings are spotted,
Table 1: Gap Analysis

131
compromising the insufficient managing of limited sample datasets and the lack of early diseases
detection techniques. Upon building over that base, the paper [2] which introduces an
incremental transfer education method targeted at reducing deficiencies in tiny farms and
enhancing crop output. Not with standing its Nevertheless, difficulties in implementing IoT in
the farming sector, particularly in under developing countries because of Internet of Things (IoT)
technology and machine learning for immediate monitoring of plant growth, health, and disease
detection. The paper reports a gap in sufficiently recognizing diseases with different symptoms,
highlighting a necessity for more exploration in this space. Expanding on the scope, the paper [3]
examines the merging of Internet of financial limitations and the necessity for particularized
knowledge, are highlighted as issues needing focus. Finally [4] present a complete crop health
monitoring system using sensors for collecting data and CNN for detecting and classifying
diseases. The paper recommends to integrate Google weather forecast to improve predictive
capacities and to include characteristics for recommend solutions and get expert feedback, deal
with gaps identified in prior researches. The following papers present a varied outlook on AIML-
based plant monitoring, highlighting changes and gaps for further research. Upcoming ventures
in this area could revolutionize agricultural methods and improve harvest oversight by tackling
acknowledged weaknesses and utilizing new innovations.
3. SYSTEM DESIGN
3.1. Hardware Requirements
The sensor module includes three sensors which are temperature and humidity sensor (DHT11),
Soil moisture sensor, PIR motion sensor and Rain detection module. A microcontroller to
analyze the data from these sensors and interface sensors with a GSM module.
Fig 1. Overview of System
1. Soil moisture sensor: The dirt humidity sensor includes two tests used to gauge the
volumetric substance of soil water. The current flows through these two tests, involving

132
soil as a channel, to get the obstruction worth of soil. The humidity content of soil is
estimated utilizing this obstruction esteem. Contingent upon how much measure of
humidity is expected for a specific sort of plant, we choose the opposition edge. As the
humidity content of the dirt is contrarily relative to the opposition noticed, when the
obstruction increments over a set edge, it implies that the dirt humidity has dipped under
the expected levels and an alarm will be shipped off the farmer. The water system
framework will be activated by the sensor esteem.
2. Temperature and Humidity Sensor: To identify temperature and humidity, a DHT11
sensor is utilized. Additionally, to gauge stickiness, the sensor utilizes two cathodes,
which have humidity holding in the middle between them. As the humidity changes,
substrate conductivity changes, and the obstruction between two terminals changes. This
opposition change is estimated and handled by the IC.
3. Rain sensor: A board with lines of nickel coated on it is the raindrop sensor. It relies on
resistance to function. Analog output pins in the Rain Sensor module measure moisture
levels. The LM393 op amp is used in the module. The electronic module and a printed
circuit board used to collect raindrops make up the module. Raindrops tend to form paths
that serve as parallel resistance when they fall and saturate the circuit board. The op amp is
used to measure this resistance. The sensor is essentially a resistive dipole with lower
resistance when wet and higher resistance when dry. Because water is a good conductor of
electricity and its presence connects parallel nickel lines to reduce resistance, raindrops on
board reduce resistance.
4. PIR sensor: The Passive Infrared (PIR) sensing module seems out for shifts in the infrared
emissions emitted by things within its seen range. This offers it the capacity to detect
motion. It splits its sight area into recognition places and creates an end signal when
recognizing movement. The reality that PIR sensors do not productively send signs makes
them energy-effective and distinguishes them from another sorts of sensors. They are fit
for a wide variety of purposes due to this aspect, like automated lighting structures and
safety mechanisms.
5. Node MCU (ESP8266) Wifi Module: NodeMCU is a highlevel programming interface for
equipment input/output gadgets, which emphatically decreases the work for designing
manipulative equipment. It utilizes a code like Arduino yet rather is an intelligent content
called Lua. It is an open source IoT stage. NodeMCU has 16 info/yield pins, and
consequently 16 hubs can be associated with a solitary hub. "NodeMCU" imply in default
to the firmware as opposed to the advancement units. ESP8266 is an inbuilt WiFi module
that can likewise be utilized as a singular module as a WiFi module.
3.2. Software Requirements
To make a web application for the discovery of plant infections, a wide range of profound
learning libraries have been utilized. The stage used to execute the code is a Jupyter notebook,
utilized for the more profound preparation of the model and to acquire high exactness.
1. Arduino IDE: The Atmega 328P is coded by utilizing the Arduino IDE. This product
likewise upholds numerous other electronic modules and, furthermore, has an easy-to-use
interface. The fundamental programming dialects utilized incorporate C/C++. The IC has
an in-constructed boot loader to transfer and execute codes from the IDE.
2. Jupyter Notebook:Jupyter Notebook is an open-source web application that is utilized to
run codes consecutively, and the result is created all the while. It very well may be utilized
for different applications, which incorporate AI calculations, successive demonstrating,
information recreation, and some more. It permits simple sharing of reports.
3. Crop Disease Dataset: The dataset being utilized for crop infection discovery comprises
just about 15,000 pictures, which incorporate 14 different harvest species. There are

133
pictures of leaf diseases brought about by contagious, microscopic organisms, forms,
viruses, and parasites. Likewise, pictures of 12 solid yields are available there. This dataset
is taken from Plant Town, which is an internet-based stage to enable little ranchers by
furnishing them with information about the yield diseaseses and different innovations
accessible to fix them so they can develop more food.
4. METHODOLOGY
4.1. Sensor Module
Temperature and humidity sensors (DHT11), soil moisture sensors, and PIR motion sensors will
collect data. Moisture values will be compared with its threshold, which is different according to
the plant. According to the reading of a soil moisture sensor, the irrigation pump will be actuated
to turn on or off the water supply. All the peripherals come together to water the plants to an
optimum level.
Fig. 2. Block diagram of sensor module
4.2. Software
4.2.1. Creating Plant Disease Detection Model
First and foremost, a dataset was imported, out of which 80% of the dataset is being utilized as
training and the rest for testing. After the dataset, every one of the libraries was imported, and
the path for the dataset was defined. To get the labels of all the folder names, a Label Binarizer
function was utilized. After this, the information was standardized according to imagenet
boundaries. A model function was used to create a transfer model, and metrics were printed.
After training the model for 15 epochs, we could achieve an accuracy of up to 99%. Ultimately,
the model was saved and the trained data was interpreted by plotting graphs of loss vs. learning
rate.

134
Fig. 3. Overview of Model
4.2.2. Using Trained model for real-time prediction
Subsequent stage was to stack the model, which was prepared prior, and afterward utilize a web
cam to catch the constant picture of the leaf and foresee whether it shows some symptoms of
disease with the assistance of the trained model.
5. RESULT
5.1. Hardware
As you can observe from Fig. 4. temperature, humidity and moisture content values can be seen
on the mobile phone, where an alert is generated by the pir motion sensor if any motion is
detected around the plant.
Fig. 4. Serial Monitor

135
Fig. 5. User’s Phone
5.2. Software
Learning rate is a significant element during the training of the model. Whenever kept less, it
could result in dialing back and can set aside some margin to arrive at exact outcomes. Assuming
it is kept high, it could not obtain an exact outcome by any stretch of the imagination. It very
well may be seen from Fig. 6. that with every iteration of the model training, the loss is getting
continuously diminished and the precision is expanded.
Fig. 6. Analyzing the training model

136
You need to determine the acceptable error tolerance for your model first, and then iterate as
much as needed until you either reach that threshold (i.e. until convergence), or your solver fails
to make additional progress (keeps returning the same number for error over and over again). It
can be observed from figure 7.1 and 7.2 that with every iteration of the model training, the loss is
getting gradually decreased and the accuracy is getting increased.
Fig. 7.1. Training and Validation Accuracy
Fig. 7.2. Training and Validation Loss
The dataset used to train and test the CNN model comprises two indexes: healthy and disease. In
the event of real-time testing, a Webcam catches the picture of the leaf that we have given and
will be stored inside another directory, and the model uses this image to predict whether it is
healthy or diseased. The leaf in Fig. 8.1. has bacterial spot disease, and the model predicts the
named disease by recognizing the obvious signs with the given likelihood. The leaf in Fig. 9.1. is
healthy and the model predicts by recognizing the obvious signs with the given likelihood.

137
Fig. 8.1. Real-time image capturing
Fig. 8.2. Prediction Probability
Fig. 9.1. Real-time image capturing
Fig. 9.2. Prediction Probability
6. CONCLUSION
The discovery of plant leaf diseases by utilizing machine learning is done effectively. The
machine learning algorithm CNN is utilized for crop disease recognition and classification via
training the datasets. The framework is carried out for early detection of crop diseases and
precautionary measures. The temperature and humidity sensor DHT11 and the soil moisture
sensor are effectively aligned. The unit for temperature value is degree Celsius, humidity and
soil moisture content are given in rate. Alert messages for rain and motion detection were sent
effectively, utilizing the GSM module. Analysis and detection of various crop diseases through
the photo are carried out successfully, along with monitoring soil parameters.
7. FUTURE ENHANCEMENT
For future scope, our mobile application can be converted into a web application, making it
vaster and more reachable to audiences for monitoring and detection. An additional feature can

138
be added to this mobile application, which will suggest remedies (pesticides) to cure the
diseases, and they can also take feedback from the experts. Also, the Google Weather Forecast
can be used to alert the farmers in advance about the mishaps, and he can take the necessary
steps. Detection of plant diseases can also be extended from leaves to stem, root and other parts,
assuring complete recognition of diseases. Our model can be powered up using either solar
power or hydro power saving energy.
REFERENCES
[1] “Plant Disease Detection and Classification by Deep learning”, Lili Li, Shujuan Zhang, Bin Wang
(2021)
[2] “Plant Disease Detection in Imbalanced Datasets Using Efficient Convolutional Neural Networks
with Stepwise Transfer Learning”, Mobeen Ahmad, Muhammad Abdullah, Hyeonjoon Moon,
Dongil Han (2021)
[3] “A Comparative study of IoT Technology in Precision Agriculture”, Immanuel Zion Ramdinthara,
Dr. P Shanthi Bala (2019)
[4] “Crop Health Monitoring System”, Kirti Tyagi, Aabha Karmarkar, Simran Kaur, Dr. Sukanya
Kulkarni, Dr. Rita Das (2020)
[5] “Plant Disease Detection and Classification by Deep learning”, Lili Li, Shujuan Zhang, Bin Wang
(2021)
[6] “An Intelligent System for Plant Disease Identification Using Machine Learning Techniques”, G.
Saleem, M. M. Javaid, A. Zahra (2020)
[7] “AI-Based Smart Agriculture Monitoring System Using IoT and Deep Learning”, M. A. Qureshi,
T. Khan, F. Khan (2022)
[8] “Deep Learning-Based Plant Disease Detection and Monitoring Using Convolutional Neural
Networks”, R. J. Poonia, S. Rawal, P. Kumar (2021)
[9] “Plant Growth Monitoring Using IoT and AI Techniques”, K. G. Bansal, A. K. Goyal (2020)
[10] “Machine Learning for Precision Agriculture: Plant Growth Prediction Using Satellite and Sensor
Data”, M. S. Abbas, A. S. Malik (2021)
[11] “Early Disease Detection in Plants Using Deep Learning Algorithms”, A. K. Srivastava, N. Gupta
(2021)
[12] “Automated Plant Disease Monitoring Using Machine Learning Techniques and IoT”, D. R. Gupta,
S. K. Patel (2020)
[13] “Smart Agriculture with Machine Learning-Based Plant Stress Detection”, H. Sharma, K. Sinha
(2021)
[14] “AI-Based Real-Time Plant Health Monitoring System Using Edge Computing”, S. Verma, P. B.
Das (2022)

Artificial Intelligence and Machine Learning Based Plant Monitoring

More Related Content

Similar to Artificial Intelligence and Machine Learning Based Plant Monitoring (20)

More from CSEIJJournal (20)

Recently uploaded (20)

Artificial Intelligence and Machine Learning Based Plant Monitoring