International Journal of Scientific Research in Engineering and Management (IJSREM)

Volume: 08 Issue: 01 | January - 2024 SJIF Rating: 8.176 ISSN: 2582-3930

Manuscript IJSREM template, Bold, Times New Roman ,Size 16, 1.5 Line Spacing
First Author1, Second Author2, Third Author3 Font Size 12
1FirstAuthor Department & College
2Second Author Department & College
3Third Author Department & College

---------------------------------------------------------------------***---------------------------------------------------------------------
ABSTRACT lower part of the leaf, which affects whole plant
Agriculture impacts life and economic status of the people. gradually. For the prevention of huge agricultural loss in
Improper management of diseases results in annual loss of the future, detection of foliar diseases in the initial stage
agricultural yield which will have serious effects on the is necessarily. It helps farmers to improve crop yield
quality, quantity and productivity if no proper care is taken. which in turnover improves GDP. Identifying the
By using some automatic technique such as image processing,
condition of the plant is essential. There are leaves basic
detection of leaf disease is quite significant and beneficial.
The use of the most utilized deep learning classification parts for disease identification. Irregular shaped black
mechanism, Convolutional Neural Network, helps in this spots or spots appear above the surface of the leaf and
regard. This paper proposes an innovative machine learning mold can occur on surfaces in damp conditions. These
approach for automated leaf disease detection. By utilizing patches are small in size but will start later covering the
image processing and deep learning algorithms, the system entire leaf area, resulting in rotten leaves. It is necessary
analyzes leaf images taken with digital cameras or smart to accurately detect foliar diseases in a particular one in
phones. Through training a convolutional neural network the initial state, before basic operations plants, such as
(CNN) on a comprehensive dataset containing healthy and pollination, fertilization, transpiration, photosynthesis,
diseased leaves, the system becomes adept at distinguishing germination, etc. are affected. Ineffective disease
between various disease types. Leveraging tagged images of
detection leads to inexperienced use pesticides that lead
healthy and diseased leaves, our system showcases robustness
and high accuracy. The automated image processing, to long-term pathogen resistance, thereby reducing the
particularly involving deep convolutional networks, ensures crop's ability to defend itself. A precise and accurate
rapid and accurate results. The system's effectiveness will be diagnosis of foliar disease was a a significant challenge .
gauged through extensive experimentation, comparing its Deep learning is widely used because it helps the
performance against existing methods. Ultimately, this project computer learning the most appropriate functions
contributes to the progress of precision agriculture and autonomously without human intervention. A neural
sustainable crop management practices. network is computational Model. CNN is mostly used for
image recognition on storage effort and time . Unlike
Key Words: Convolutional neural network (CNN), Pre- traditional methods which to use hand-crafted features,
Processing, Deep Learning, Image Processing, Classification,
CNN optimizes the hidden layers filter the parameters
Remedy Recommendation .
and weights and generate the required functions solve the
classification problem. CNN consists of different layers
1.INTRODUCTION
and uses different foci for identification, detect, classify
and predict diseases. Provides fast, accurate and reliable
In India many people are farmers and grow a different
results and faster treatment that reduces negative impact
crops. Agriculture is the main source of livelihood and
on harvest. Agricultural yield is higher and productivity
primary occupation. India is ranked second in the world
is increased with a minimum expenses.
agricultural production. Many people depend on
manufacturing agriculture directly or indirectly.
Producing agricultural high quality yield is essential for
sustainable development Earth. Factors such as climatic
2. Proposed Method
conditions, soil conditions and diseases affect plant Plants are susceptible to many different types of
production. Due to population growth, political diseases. There are variety of factors which can influence
instability and changing climate conditions, the plant growth, including disturbances caused by external
agricultural industry is looking for new and better ways factors such as temperature, moisture, food, light, and
to increase food production. Produce crops with better disease. Since we use the CNN algorithm in our model to
productivity and quality, we must monitor and control
identify disease, we employ it for good data. Predicting
requirements such as temperature, humidity and light.
Another major threat to food security is leaf blight. It those leaf diseases through advanced technology like
reduces the quality and reduces the yield of the crop. machine learning enables early detection, preventing
Pathogens such as insects, pests, fungi, bacteria and widespread crop damage. Automated systems analyze
viruses cause disease leaves. They eat the upper and plant health data, identifying potential issues promptly.

© 2024, IJSREM | www.ijsrem.com DOI: | Page 1

 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 08 Issue: 01 | January - 2024 SJIF Rating: 8.176 ISSN: 2582-3930

Moreover, these systems can recommend targeted

remedies, optimizing agricultural practices and
minimizing the need for broad-spectrum pesticides. This
not only enhances crop yield but also promotes
sustainable and eco-friendly farming methods.

2.1. ALGORITHM :
We are pre-processing the database by picture
reshaping, re-sizing, and by changing the image to an
array shape. The test picture is additionally prepared. A
database comprising of 78,000 different plants has been
compiled, from which any image can be selected as a
test picture for a computer program. You are going to
have to train your machine learning algorithm with a
huge amount of data. The training database is the source
from which the model learns to recognize the images of Table -1: Plant Village Dataset
the test set. You don't want to use the same training
dataset for both training and testing. Convolutional
neural networks have diverse layers that are
convolution, Dropout, Activation, Flatten, Convolution
2D,MaxPooling 2D. You can use them to detect objects
in an image. After the model has been trained, it will
then be able to determine the disease of the plant.

Fig-1:Process model for detection of leaf diseases 2.3. DATA PROCESSING AND IMAGE
AUGMENTATION :
2.2. DATASET STUDY :

Based on species and disease, the 54303 healthy and Image augmentation is essential for developing an
unhealthy leaf images from the Plant Village dataset are effective image classifier. Although datasets may
divided into 38 categories. We attempted to predict the contain hundreds to thousands of training examples,
class of diseases by looking at more than 50,000 images there may not be enough variety to produce a
of plant leaves with distributed labels from 38 classes. trustworthy model. The many image enhancement
Prior to optimisation and model predictions, the image is options include scaling, angular rotation, and vertical or
shrunk to 256 x 256 pixels. horizontal flipping of the image. The amount of relevant
data in a dataset is increased by these additions. One
finds that each picture in the Plant Village dataset is 256
x 256 pixels in size. Both data processing and image
© 2024, IJSREM | www.ijsrem.com DOI: | Page 2
 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 08 Issue: 01 | January - 2024 SJIF Rating: 8.176 ISSN: 2582-3930

enhancement are performed using the Keras

deeplearning framework. 2.5. FEATURE EXTRACTION:
The following are the augmentation choices used for
training: In leaf disease detection using Convolutional Neural
• Rotation - Randomly rotate a training image through a Networks (CNN), feature extraction is a crucial step that
range of angles. involves identifying relevant patterns and characteristics
• Brightness - When training, the model is fed images of of the leaf images that can be used to distinguish
varying brightness, which helps it adapt to changes in between healthy and diseased leaves.
lighting.
The process of feature extraction in CNN involves
We also use other techniques like rotation, shear, height several steps:
zoom, width zoom, height shift, width shift, saturation,
contrast, brightness, solarize, equalize, and dropout. Input Image: The input image is first preprocessed to
remove any noise or artifacts that could affect the
2.4. METHODOLOGY: accuracy of the detection model.

Convolutional Layers: The input image is then passed

through several convolutional layers, which apply a set
of filters to extract features at different levels of
abstraction.

Fig-2:System overview
Pooling Layers: After each convolutional layer, a
These Steps are for identifying plant diseases the entire pooling layer is added to reduce the
process is mainly divided into two stages: dimensionality of the output and capture the most
important features.
1.Training Phase
a)Pre-processing and segmentation of the dataset's input Flatten Layer: Once the convolutional and pooling
images will take place. layers have been applied, the resulting output is
b)Image augmentation, image segmentation, and colour flattened into a vector of features.
space conversion are all included in pre-processing.
Fully Connected Layers: Finally, the vector of features
First, a filter is used to enhance the image's digital is passed through a set of fully connected
representation. then make an array from each image. layers, which learn to classify the input image into the
Each image name is changed to a binary field using the appropriate category (i.e., healthy or diseased).
medical term for binarized diseases.
2.Testing Phase
2.6. OVERVIEW OF TECHNOLOGIES:
After the same processes like which have been done in
the training phase –
To identify each plant disease, CNN classifiers have
LIBRARIES-
been trained. The classifier, which has been trained to
NUMPY:
categorise various diseases in that plant, is called upon
Multidimensional array objects and functions to
using Level 2 results. If they are missing, the leaves are
manipulate them are found in the NumPy (Numerical
considered healthy.
Python) package. The Python library NumPy enables
FINALLY Our model will recommend fertilisers to treat
you to manipulate arrays mathematically and logically.
diseased leaves and safety precautions to prevent them.

© 2024, IJSREM | www.ijsrem.com DOI: | Page 3

 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 08 Issue: 01 | January - 2024 SJIF Rating: 8.176 ISSN: 2582-3930

CV2:
The import name for the OpenCV module is cv2. An 2.8.RESULTS:
open-source library called Opencv is used for computer
vision tasks like video and CCTV footage processing The dataset will be used to train the CNN model, which
and picture analysis. will then be used to determine whether a leaf is healthy
or diseased. The developed model can recognize
KERAS: different types of diseased leaves out of healthy leaves
Keras is a robust and user-friendly open-source Python and has the ability to distinguish the leaves from their
package. The Keras machine learning framework was surroundings in the image. After successfully detecting
created using TensorFlow, Theano, and Cognitive the leaf disease with a good level of confidence, the
Toolkit. (CNTK). Quick numerical calculations are remedy corresponding to it i.e., the pesticide which is to
possible with a Python programme called Theano. The be used as cure is displayed. This remedy acts as a
most popular symbolic math library used in TensorFlow defense mechanism against the disease. The remedy
neural network development. recommendation system was developed using saved
pickle model by integrating it with flask. It has been
MATPLOTLIB: observed that, compared to the existing traditional
Matplotlib is a well-known Python data visualisation approaches, proposed CNN approach provides better
library. From array data, this cross-platform library accuracy and is an efficient method for diagnosis of leaf
creates 2D charts. It includes an object-oriented API for diseases.
embedding plots in Python GUI toolkits like PyQt and
WxPython Tkinter. It is compatible with Jupyter
notebooks, web application servers, and other tools in 3. CONCLUSIONS
addition to Python and IPython shells.
The main goal is to identify and detect the leaf diseases
TENSOR FLOW: accurately keeping in mind the benefits of farmers.
The TensorFlow platform helps you implement best Neural Networks help in generating a model which
practices for data automation, model tracking, mimics human brain. Before neural networks, very few
performance monitoring, and model retraining. Using models were there that were actually trained in this way.
production-level tools to automate and track model A CNN model for automatic leaf disease detection using
training over the lifetime of a product, service, or python gives an optimum accuracy of 96%. By the use
business process is critical to success. of GPU for processing, accuracy and speed can be
increased. The proposed method solves the problem of
2.7.TEST CASES: expensive domain expert. After successful prediction of
the leaf disease, it also provides the remedy that must be
taken as a cure in order to improve the health of the
plant within a very less time. This model can be installed
on drones so that aerial surveillances and live coverage
of large agricultural fields can be done which will
reduce manual work to be done and the time consumed.
A camera with high resolution attached to the drone
captures images of leaves which act as input for the
model. For low scale farming purposes, the budget is
very high, but for large scale farming it is valuable.

ACKNOWLEDGEMENT

We would like to thank Department of Computer Science and

systems Engineering , Lendi Institute of Engineering and
Technology, Vizianagaram for helping us carry out the work
and supporting us all the time.

REFERENCES

Table -2: different testcases 1. Baldonado, M., Chang, C.-C.K., Gravano, L., Paepcke, A.: The
Stanford Digital Library Metadata Architecture. Int. J. Digit. Libr.
1 (1997) 108–121
2. Bruce, K.B., Cardelli, L., Pierce, B.C.: Comparing Object
Encodings. In: Abadi, M., Ito, T. (eds.): Theoretical Aspects of

© 2024, IJSREM | www.ijsrem.com DOI: | Page 4

 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 08 Issue: 01 | January - 2024 SJIF Rating: 8.176 ISSN: 2582-3930

Computer Software. Lecture Notes in Computer Science, Vol.

1281. Springer-Verlag, Berlin Heidelberg New York (1997) 415–
438
3. van Leeuwen, J. (ed.): Computer Science Today. Recent Trends
and Developments. Lecture Notes in Computer Science, Vol.
1000. Springer-Verlag, Berlin Heidelberg New York (1995)
4. Michalewicz, Z.: Genetic Algorithms + Data Structures =
Evolution Programs. 3rd edn. Springer-Verlag, Berlin Heidelberg
New York (1996)

© 2024, IJSREM | www.ijsrem.com DOI: | Page 5