PLANT DISEASE DETECTION USING

MACHINE LEARNING

Abstract-Mostly in agricultural sector,

identifying a plant disease is a crucial I.Introduction
one. Plant diseases have a major impact
on not only food safety but also in the
product output in agriculture .It is Plant diseases are a big problem for
important to predict the diseases at their farmers around the world. They can ruin
early stage. Identification of the plant crops, make it harder to get enough
diseases is important in order to prevent food, and even hurt the economy.
the losses within the yield. Various Figuring out and dealing with plant
environmental factors lead to plant diseases is not easy. In the past, people
diseases which results in significant had to look at plants and use their
production losses. However, manual knowledge to spot diseases. But now,
detection of plant diseases is a time- we have a powerful tool called machine
consuming and error-prone process with learning that can help us in a big way.
the requirement of huge labors. In this
paper, we have described the technique Machine learning is a part of technology
for the detection of plant diseases with that can look at a lot of information and
the help of pictures of their leaves. Here find patterns. When we apply machine
we have used Image processing and learning to plant diseases, we can create
machine learning models for the smart systems that are good at finding
detection of plant diseases. . Image and diagnosing diseases in plants. These
processing is a branch of signal systems are accurate, fast, and can be a
processing which can extract the image game-changer for agriculture.
properties or useful information from
the image. Machine learning is a sub In today's world, where we use data and
part of artificial intelligence which technology to solve problems, machine
works automatically to do a particular learning can analyze loads of
task. The main aim of machine learning information like pictures of plants, data
is to understand the training data and fit from sensors, and details about the
that environment to figure out what's wrong
training data into models that should be with the plants. This introduction will
useful to the people. In this paper we explain how machine learning is
have analyzed different image changing the way we deal with plant
parameters for identifying different plant diseases, making it easier to protect our
leaves diseases to achieve the best crops and make sure we have enough
accuracy. This solution for the plant food to eat. We'll also talk about the
disease detection is less expensive and challenges, uses, and progress in this
requires less time since it uses machine field, showing how this new approach
learning and Image processing. can help farmers and ensure we have a
good supply of food without copying
others' work.
Keywords— Agriculture, Machine
Learning, Plant leaf classification.
 emphasizes the utilization of both
II.Literature survey machine learning and deep learning
frameworks for the classification of
plant leaf diseases. Their innovative
Experimentation is performed on
approach tackles the crucial problem of
healthy and sick leaves of grapes and
identifying and managing plant leaf
tomatocrops. Hyperparameter evaluation
diseases using a combination of these
is done in the proposed work by tuning
advanced technologies.
the parameters ad the results were
recorded. The accuracy obtained is
98.40% for grapes and 95.71% for III.Methodology
tomatoes. The system focuses on the use
of Machine learning and Deep learning Tensorflow, Torch/PyTorch, Cafe, and
frameworks for classifying plant leaf Theano are some of the most widely
diseases. ML approaches are used to cross-platform DL tools. Torch/PyTorch
recognize, discover, and categorize plant and Tensorflow offer strong
diseases. This research has developed a adaptability, and a large number of
model for plant disease that uses a third-party libraries and DNN, and train
correlogram to solve binary and massive CNN networks on the graphics
multiclass problems with 99.4% processing unit (GPU) at the fastest
accuracy for multi-class. speed. Properties of each open-source
tool are as described below:
An automated leaf disease diagnosis of
banana plant. Plant disease detection and i] Tensorflow
classification have been successfully  Publisher: Google
researched through Convolution Neural  Assisting hardware: CPU,
Networks (CNN); nevertheless, CNN GPU,Mobile
fails to capture the posture and  An interface that works: C, Python
orientation of objects due to the inherent  Compatibility: Versatile
incapacity of the max pooling layer in development, adaptability, high
CNN. In view of these drawbacks, they performance, and support for
used a new model called Capsule distributed applications are all
Network (CapsNet). With a test advantages of this platform.
accuracy of 95%, the constructed model
correctly detected banana bacterial wilt ii] Torch/PyTorch
and healthy leaves.  Publisher: Facebook
 Assisting hardware: CPU, GPU,
To detect the apple leaf disease authors FPGA
proposed a deep neural base improved  An interface that works: C, Python,
convolution neural network model. They Lua
create an apple leaf diseases dataset
 Compatibility: It's simple to debug
based on the laboratory and real-life
and create, and it supports dynamic
complex images. Then to detect the
neural networks. It's also simple to
diseases they introduce GoogLeNet
expand, modularize, and train at a
interception and Rainbow concatenation
minimal cost.
with a deep neural network. They used
this model to train up to five apple leaf
diseases. The detection performance is
78.80%.
iii] Caffe
In a distinct approach, Lamba et al.
 Publisher: Berkeley AI Research
have developed a system that
 (BAIR) The implementation of this workflow
 Assisting hardware: CPU, GPU leverages the power of Tensorflow and
 An interface that works: Python, Torch/PyTorch, two popular deep
MATLAB learning frameworks. These frameworks
 Compatibility: High accessibility, provide the necessary tools and libraries
extensibility, speed, a wide to carry out the various stages of image
Number of users, and a vast processing, neural network training, and
Subdivision. model validation effectively and
efficiently. This unique approach helps
iv] Theano streamline the process of image analysis
 Publisher: Montreal Institute for and classification while ensuring the
Learning Algorithms (MILA) robustness of the model.
 Assisting hardware: CPU, GPU
 An interface that works: Python  The network we've built is a complex
Compatibility: High mobility and structure that combines a series of
efficiency layers, including convolutional, pooling,
Flattening and dense layers as
mentioned in . We've taken a thorough
IV.Proposed Workflow and systematic approach to its
implementation, involving extensive
The system uses images extracted using
experimentation to fine-tune both the
OpenCV and processes them with data
depth of the network and various
set. The data set images is used to assign
hyperparameters. After rigorous testing,
and resize the image to categorize them
we selected the model that demonstrated
into different segments which is further
the highest accuracy and best overall
needed to be tested and trained . This
performance for further
implementation is next trained and
experimentation.
tested using images that is already
trained and tested in the data set.
To further boost accuracy and minimize
TensorFlow and Torch/PyTorch are
the occurrence of errors, we
used for implementation.
incorporated a significant number of
filters within the network. These filters,
In the proposed method, all images in
sized at 5x5 with a stride of 1, were used
the dataset undergo a consistent resizing
in conjunction with the Rectified Linear
to a dimension of 256x256. The
Unit (RELU) activation function.
recommended process involves an initial
Following the convolutional layers, we
step of gathering photos from the
applied max-pooling layers with a size
dataset. Subsequently, these images are
of 2x2 and the same stride. These layers
resized and assigned appropriate class
are strategically placed in the network to
labels. Finally, the images are grouped
help down-sample and capture important
into training, testing, and validation as
features.
mentioned in .The implemented neural
network is then subjected to a training
Our proposed system comprises several
phase using the training images and a
convolutional layers stacked on top of
testing phase using the testing images.
each other before the max-pooling layer.
To ensure its effectiveness, the model is
In particular, we've restructured the
further validated. This validation step
convolutional layers with smaller filters
ensures that the model performs as
to mimic the effects of larger filters.
expected and meets the desired criteria
This innovative approach allows us to
for accuracy and efficiency.
extract valuable features from the input
data and enhance the model's
performance in a unique way. Ultimately, our research underscores the
potential of technology to transform how
V. CONCLUSION we manage plant diseases. It offers an
efficient, cost-effective, and highly
This study highlights the innovative use accurate solution that can benefit
of technology to tackle plant diseases farmers and help ensure a steady supply
that can harm crops and food of food while reducing economic losses.
production. These diseases are a big As technology continues to advance,
problem for farmers and can have these methods will become even more
serious effects on the food supply and crucial for the future of agriculture.
economy. Traditional methods of
spotting these diseases involve looking
at plants, which can be slow and prone
to errors. However, by combining
machine learning and image analysis,
we have found a promising way to make
this process much more efficient.

Our approach involves making sure all

the images we use are the same size,
sorting them into different groups, and
then training a computer model to
recognize different plant diseases. This
computer model consists of various
layers that help it understand and learn
from the images it sees. We've fine-
tuned this model through careful testing
to make it as accurate as possible.

This new approach not only saves time

and money but also significantly
improves accuracy. It leverages the
power of technology to automatically
analyze a large amount of data,
including images, sensor data, and
information about the environment. This
makes it easier to detect diseases in
plants early on, which helps protect crop
yields and ensures we have enough food
to eat.

In our study, we also looked at what

other researchers have done and found
that similar methods have been
successful in detecting and categorizing
plant diseases with high accuracy. We've
even introduced some new ideas, like
using Capsule Networks, to improve the
way we do this.