Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Literature Survey:

1. Introduction:

Agriculture plays a huge role in people's lives and the economy. In India, it's the
main way many people make a living, around 58% of the population. India is
even one of the top countries globally when it comes to growing crops. Back in
2018, over half of the workers got their jobs from farming, and this helped make
up about 18-20% of the country's money. So, India is doing well in farming, but
there are problems to deal with. Some of these problems include not using good
farming methods, not using enough helpful things like compost and fertilizers,
not having enough water, and having plants get sick. When plants get sick, it's
really bad for how well they grow. In fact, about 20-30% of crops don't grow well
because of diseases. This causes a big loss in how much food we get from farms.
That's why it's super important to spot these diseases early and do something
about it.

Before, people used to look at plants and try to figure out if they were sick, but
this took a lot of time and sometimes they made mistakes. But now, thanks to
new technology, things have gotten better. There's a new way to figure out if
plants are sick using computers and smart systems. This system is all about
identifying diseases in the leaves of 14 different plants like apples, strawberries,
and more. It uses a special kind of computer program called a Convolutional
Neural Network (CNN) which looks at pictures of the leaves and figures out if
they're healthy or not. This helps farmers take care of sick plants quickly and
accurately, so they can keep growing lots of good food.

2. Research Objects and Scope

The survey focuses on leveraging Deep Learning, particularly Convolutional


Neural Networks, to develop a robust system for plant disease detection. A
comparative analysis of different CNN variants, namely Faster R-CNN, R-FCN,
and SSD, is conducted to ascertain their efficacy in pinpointing diseased plant
regions. The study places equal emphasis on the preprocessing of data to
optimize the accuracy of the models. The research methodology involves
training the chosen CNN variants on a dataset comprising images of plants with
various diseases.

The primary focus of this study is to harness the potential of Deep Learning,
specifically Convolutional Neural Networks (CNNs), for the purpose of detecting
plant diseases. The overarching goal is to evaluate the effectiveness of these
advanced technological tools in accurately identifying signs of illness in plants
based on visual imagery. The scope of this research encompasses a thorough
investigation into the applicability and performance of various CNN
architectures, including Faster R-CNN, R-FCN, and SSD.

3. Methodology

Plants are vulnerable to various diseases and problems that affect their growth
and health. These issues can be caused by factors like changes in the
environment, such as temperature and humidity, as well as insufficient
nutrients, light, and common diseases like bacterial, viral, and fungal infections.
To address this, we propose using a powerful tool called Convolutional Neural
Network (CNN) to detect diseases in plant leaves. CNN is known for its ability to
achieve high accuracy when the data is of good quality.

A. Dataset:

For our study, we utilized the Plant Village Dataset, which contains a collection
of 54,303 images showing both healthy and unhealthy plant leaves. These
images are categorized into 38 groups based on the plant species and the type
of disease. Our analysis involved examining over 50,000 of these images, each
with labels indicating its category. We resized the images to a standardized 256
× 256 pixels and conducted further optimization and model predictions using
these modified images.

B. Data Processing and Augmentation:

In order to build an effective image classifier, we employed image augmentation


techniques. These methods are crucial because even though a dataset may
contain hundreds or thousands of examples, it might still not capture enough
variety to create an accurate model. Augmentation involves making changes to
the images, such as flipping them vertically or horizontally, rotating them at
different angles, and adjusting their brightness. These augmentations expand
the range of data available for training. Our dataset's images were all set to a
size of 256 x 256 pixels. We executed data processing and image augmentation
using the Keras deep-learning framework. The specific augmentation techniques
included:

- Rotation: Randomly rotating training images at different angles.


- Brightness: Adapting to variations in lighting by using images with varying
brightness levels during training.
- Shear: Adjusting the shearing angle to account for potential irregularities
in the images.

C. System Overview and Methodology:

The entire process is divided into three distinct stages:

i. Input Image Capture: Users can take pictures of plant leaves using an
Android device or upload images to our web application.

ii. Segmentation Pre-processing: This stage involves several steps like image
segmentation, enhancement, and conversion to a different colour space.
First, the digital image is enhanced using a filter. Then, each image is
transformed into an array format. Using the scientific names for plant
diseases, we convert the image names into binary fields.

iii. CNN Classification: The final stage employs Convolutional Neural Network
(CNN) classifiers trained to detect diseases in different plant classes. Level
2 results are used to activate a specific classifier, which has been trained
to identify various diseases within that specific plant category. If no
disease is detected, the leaves are categorized as "healthy."

4. Overview of Plant Disease Detection

Plant disease detection has undergone a transformative evolution with the


integration of Convolutional Neural Networks (CNNs), a cutting-edge branch of
artificial intelligence. At its core, CNN technology employs layers of specialized
filters to meticulously analyse images of plant leaves, extracting intricate
features that reveal the presence of diseases. This process commences with the
accumulation of a comprehensive dataset comprising images of both healthy
and diseased leaves, meticulously labelled for training. During training, the CNN
learns to decipher and differentiate subtle patterns within these images,
associating distinct visual cues with specific disease categories. Fine-tuning
further refines the model's predictive accuracy, ensuring its adeptness in
identifying diseases beyond its training dataset. As a deployed system, the CNN
effortlessly processes new images of plant leaves, instantaneously recognizing
telltale signs of diseases that could otherwise jeopardize crop yields. By
seamlessly amalgamating the prowess of CNNs and their innate ability to discern
intricate visual nuances, plant disease detection has emerged as a
transformative force, empowering agricultural stakeholders with timely and
precise insights that augment disease management and crop protection
strategies.

5. Convolutional Neural Networks (CNN)

A. Deep Learning:

Deep learning is a subset of machine learning algorithms that involves multiple


layers to extract higher-level information from raw input data. It operates
similarly to how the human brain filters information. Many deep learning
techniques utilize neural network architectures. The term "deep" refers to the
presence of multiple hidden layers within the neural network. Unlike
conventional neural networks with only 2-3 hidden layers, deep neural networks
can consist of up to around 150 layers.

B. Convolutional Neural Network (CNN):

A specific type of deep neural network is the convolutional neural network


(CNN). CNNs combine learned features with input data, making them
particularly suitable for processing 2D data like images. CNNs eliminate the need
for manual feature extraction when classifying images. The model itself extracts
features directly from images during training. CNN architecture includes layers
such as Input layer, Output Layer, Convolutional Layer, Fully Connected Layer,
Softmax Layer, and Pooling Layer.

C. VGG 16 Model:

VGG16 is a CNN model used for large-scale image tasks. It excels in both object
localization and image classification. The model is depicted below:

The model comprises seven distinct layers, each processing specific information:

1. Input Layer: Holds image data with parameters like height, width, depth, and
colour information (RGB).

2. Convolutional Layer: Also known as the feature extraction layer, it extracts


significant features from input images using dot products.

3. Pooling Layer: Reduces computational power by decreasing the dimensions


of featured matrices obtained from dot products.

4. Fully Connected Layer: Connects neurons from one convolutional layer to


another, facilitating complex connections.

5. Softmax Layer/Logistic Layer: Performs multi-classification or binary


classification, determining the probability of an object's presence in an image.
6. Activation Function - ReLU: Rectified Linear Unit (ReLU) transforms weighted
input and activates nodes in convolutional operations, a common activation
function in neural networks.

Deep learning involves intricate layers to extract meaningful information, with


CNNs being a specialized architecture for processing images. The VGG16 model
exemplifies this by efficiently performing both object localization and image
classification tasks through its distinct layers, each handling specific aspects of
the data.

6. CNN Applications:
1. Leaf Disease Identification: CNNs can classify and identify various types
of leaf diseases based on patterns and visual cues present in leaf images,
helping farmers detect diseases early.
2. Disease Severity Assessment: CNNs can quantify the severity of a
disease by analysing the extent of damage visible in plant images,
allowing for targeted interventions.
3. Multi-Class Classification: CNNs can distinguish between different
diseases affecting the same plant species, aiding in accurate diagnosis
and treatment recommendations.
4. Automated Diagnosis: CNNs can automate the process of diagnosing
diseases, reducing the need for human experts and saving time in critical
situations.
5. Data-driven Insights: CNN-based systems can analyse large datasets of
plant images, helping researchers identify trends, disease patterns, and
factors contributing to outbreaks.

CNN architecture used:


VGG-16: VGG-16, short for the Visual Geometry Group 16-layer network, is a
deep convolutional neural network architecture that gained significant
popularity for its simplicity and effectiveness in image classification tasks. It
was developed by the Visual Geometry Group at the University of Oxford and
was one of the first architectures to demonstrate the potential of deep
learning. It has been successfully utilized in plant disease detection systems,
showcasing its effectiveness in analysing and classifying plant images to
identify various diseases.

7. Performance Evaluation:
A. Data Analysis:
The dataset that is used in this proposed system project is the Rice Leaf
Diseases dataset which was downloaded from Kaggle website. This dataset
contains 120 jpg images of disease infected rice leaves. The images are
grouped into 3 classes based on the type of disease. There are 40 images in
each class.

B. Result Analysis:
Result analysis in the context of a plant disease detection system using CNN
involves a comprehensive examination of the outcomes and metrics obtained
during the performance evaluation. The result about whether the plant is
healthy or diseased will be provided to the farmer via SMS. By implementing
this SMS-based communication approach, you can empower farmers with
timely and actionable information about their plants' health, enabling them to
make informed decisions and take appropriate measures to manage diseases
effectively.

8. Transfer Learning and Pre-Trained Models:


Transfer learning is a machine learning technique that involves utilizing
knowledge gained from one task to improve the performance of a related but
different task. In the context of Convolutional Neural Networks (CNNs) and
plant disease detection, transfer learning leverages pre-trained CNN models
that have been trained on large datasets for general image recognition tasks.
These pre-trained models capture a wide range of features useful for various
image-related tasks, and this knowledge is transferred to the specific task of
plant disease detection.
How Transfer Learning is Applied to CNNs in Plant Disease Detection:
1. Feature Extraction: In transfer learning, the lower layers of a pre-trained
CNN, which capture basic features like edges and textures, are retained. These
layers act as a feature extractor for the new task. The higher layers,
responsible for more complex features and task-specific details, are often
replaced or fine-tuned.
2. Fine-Tuning: The higher layers of the pre-trained CNN are modified or fine-
tuned to adapt to the specific characteristics of the plant disease detection
task. This involves training the model on a smaller dataset of labelled plant
images, updating the model's weights to better match the patterns relevant to
disease detection.

9. Challenges and Future Directions:


While Convolutional Neural Networks (CNNs) have proven to be effective for
plant disease detection, there are several challenges and limitations associated
with their use in this context:
1. Limited Data and Diversity: Collecting large and diverse labelled datasets for
various plant species and diseases can be challenging. The performance of
CNNs heavily depends on the availability and quality of training data.
2. Generalization to New Diseases: CNNs might struggle to generalize to new
diseases with subtle visual differences that were not present in the training
data.
3. Limited Robustness: CNNs might struggle with variations in lighting,
environmental conditions, and plant growth stages, affecting their robustness
and accuracy.
4. Seasonal Variability: Changes in lighting, weather conditions, and plant
growth during different seasons can impact the accuracy of disease detection.
5. Data Privacy and Security: Collecting and sharing plant images for disease
detection might raise privacy and security concerns, especially if sensitive
information is involved.
Despite these challenges, CNNs have shown significant promise in plant
disease detection. Researchers and developers continue to address these
limitations through techniques like data augmentation, transfer learning,
active learning, and the development of specialized architectures that can
enhance the accuracy, robustness, and real-world applicability of CNN-based
systems for plant disease detection.

10. Conclusion:
Protecting crops in organic farming is not an easy task. This depends on a
thorough knowledge of the crop being grown and possible pests, pathogens
and weeds. In our system, a special deep learning model has been developed
based on a special architectural convolution network to detect plant diseases
through images of healthy and diseased plant leaves. Studies show that
managing plant disease can help increase the yield by about 50%.
11. References:
1. Turkish Journal Of Computer and Mathematics Education. (PDF)
https://turcomat.org

2. Plant Disease Detection and Classification using CNN. (PDF)


https://www.ijrte.org

3. Plant Disease Detection Using Image Processing Techniques. (PDF)


www.ijirset.com

You might also like