PERFORMANCE ANALYSIS OF CROP DISEASE

DETECTION AND PREDICTION USING AI

 KISHORE KUMAR J - 113221051303

 RAGUL M - 113221051305
 SURYA D - 113221051306
 VISHVA S - 113221051308

Project Guide
Mrs.V. SELVE
Assistant Professor III – EEE

Department of Electrical and Electronics Engineering

Velammal Engineering College
Chennai – 600 066
12 July 2024 Project Review 1
• Location of Project : In-House

12 July 2024 Project Review 2

 Objectives
 Enhance disease management with AI and IoT.
 Improve crop yields through data-driven insights.
 Promote sustainable farming and targeted interventions.
 Enable timely action to reduce crop loss and ensure food
security.
 Provide farmers with accessible disease diagnosis tools.
 Increase farmer awareness for early disease detection.

12 July 2024 Project Review 3

 Problem Statement
Precision in Disease Detection: AI algorithms offer higher accuracy
in identifying and diagnosing crop diseases from visual and sensor data,
surpassing traditional methods.
Timely Intervention: Rapid data processing enables early disease
detection, allowing farmers to implement timely interventions and
minimize crop losses.
Scalable Solutions: AI-based systems can be scaled across large
agricultural areas, providing comprehensive coverage and support for
diverse crops.
Promotion of Sustainability: By optimizing resource use and
reducing reliance on chemical treatments, AI supports sustainable
farming practices, enhancing long-term agricultural productivity and
environmental health.

12 July 2024 Project Review 4

 Existing Solution
Visual Inspection and Manual Sampling: Farmers
predominantly rely on visual inspection and manual sampling to
detect crop diseases, which are labor-intensive and subjective
methods.
Expert Consultation: Agricultural experts provide guidance
based on their experience and knowledge, but access to experts
can be limited and consultations may not always be timely.
Chemical Treatments: Farmers often use pesticides and
fungicides preventively or reactively, which can lead to overuse,
environmental impact, and resistance issues among pathogens

12 July 2024 Project Review 5

 Proposed Solution
AI-Powered Image Recognition: Implementing machine
learning algorithms that analyze crop images to accurately
identify disease symptoms, improving diagnostic precision and
efficiency.
Sensor-Based Monitoring: Utilizing IoT sensors to
continuously monitor crop health metrics and environmental
conditions, enabling early detection of diseases before symptoms
are visible.
Decision Support Systems: Developing AI-driven platforms
that integrate data from various sources to provide timely
recommendations for disease management, optimizing resource
use and reducing reliance on chemical treatments.

12 July 2024 Project Review 6

 Methodology Overview

 Data Collected from Google public datasets.

 Machine learning models like Random forest, XGBoost,
Support Vector Classifier.
 DRL, LSTM models are trained using outputs of above
mention machine learning models.
 The evaluation metrics used are cross-entropy evaluation.
 Data Collection

 Data Collected then for variety data from Google public

datasets.
 The datasets include types of leaf and insect data with
their type of disease.
 The diverse version of our dataset ensure the robust
nature of the model.
 Data Preprocessing

Libraries used:
 pytouch (Python)
 Pandas (python)
 Pillow (python)
 CNN Algorithm (machine learning)
 Data Visualization

 Understand dataset structure and identify class imbalances.

 Visualized the number of images per class to detect imbalances.
 Showcased leaf images from different disease categories to highlight
visual differences.
 Demonstrated techniques like flipping, rotation, and noise injection to
increase data variety.
Key Insights:
 Augmentation helps address class imbalances.
 Ensures the model learns from diverse visual representations,
improving generalization.
Sample Input Images
Sample working of CNN Algorithm
Data collecting using CNN Algorithm
Disease Prediction using CNN
Algorithm
Image Processing
Training and validation of Mobile Net
and densenet
Loss and accuracy of proposed CNN
model
Real time working this application
Sample Input Image
Output for given sample input
 Model Architecture (CNN)

 Convolutional Layers: Extract features like edges,

textures.
 Pooling Layers: Reduce dimensionality, focusing on the
most important features.
 Fully Connected Layers: Flatten the data and classify
into 39 categories.
 Activation Functions: ReLU for hidden layers, Softmax
for the final layer.
CNN algorithm for accuracy

 Leverage CNN for accurate crop disease detection

through image analysis, enabling timely interventions.

 Improve crop yields and reduce losses using AI-driven

insights and real-time data.

 Promote sustainable farming by minimizing chemical use

through precise, targeted treatments.

 Provide farmers with accessible tools for automated

disease diagnosis and early detection, ensuring food
security.
 Data Augmentation Technique

 Image Flipping: Mirror image horizontally/vertically.

 Gamma Correction: Adjust brightness levels.
 Noise Injection: Add random noise to images.
 PCA Color Augmentation: Change color distribution.
 Rotation and Scaling: Rotate or resize the images.
Data Visualization in Deep Learning

 Visualized the dataset to identify class imbalances across

39 plant disease categories.
 Displayed sample images to observe visual differences
between healthy and diseased leaves.
 Applied augmentation techniques (flipping, rotation,
noise injection) to increase data variety.
 Augmentation improved model generalization and
prevented overfitting.
 Enhanced the CNN’s ability to accurately classify plant
diseases by learning from diverse features..
 Long Short Term Memory (LSTM)

 Memory Retention: LSTM networks store and retain

long-term dependencies, helping remember past decisions
made by DRL agents.

 Sequential Data Handling: LSTM processes sequential

data to capture time-based patterns, ensuring DRL
decisions consider historical contexts.

 Enhanced Decision-Making: By retaining relevant past

information, LSTM improves the accuracy and reliability
of DRL's real-time decision-making processes.
 Hybrid Network of the Model

 Integrated Architecture: The hybrid model combines DRL

and LSTM to monitor power system states and make
adaptive decisions based on historical and real-time data.

 Sequential Learning and Memory: LSTM networks

capture and retain temporal dependencies from past
machine learning outputs, providing context for DRL-
based decision-making.

 Dynamic Optimization: DRL uses insights from LSTM

memory to dynamically adjust parameters and control
strategies, optimizing fault management and system
reliability in real-time.
 Block Diagram

Data Data Feature

Collection Preparation Extraction

Model Model
Fine tuning
Evaluation Training

Model Real-time
Deployment Predictions

12 July 2024 Project Review 2

7
 Components
1. SOFTWARE REQUIRED:

• PYTOUCH
• CONVENTIONAL NEURAL
NET WORK(CNN)
• BLOCK CHAIN
• PILLOW
• PANDAS

Project Review 2
8
 Advantages

(1) Real time decision making capability.

(2) Reduces Time consumption for fault detection.
(3) Detects potential security breaches.

12 July 2024 Project Review 2

9
 Road Map
REVIEW WORK TO BE COMPLETED STATUS

FIRST Identification of Soft Wares required completed

REVIEW

SECOND Software Implementation completed

REVIEW

THIRD Completion of full project &

REVIEW
submission of project
report

12 July 2024 Project Review 30

 References
 Y. Kaya and E. Gürsoy, "A novel multi-head CNN design to identify plant
diseases using the fusion of RGB images", Ecol. Inform., vol. 75, pp.
101998, 2023.

 E. Moupojou et al., "FieldPlant: A Dataset of Field Plant Images for Plant

Disease Detection and Classification With Deep Learning", IEEE Access,
vol. 11, pp. 35398-35410, 2023.

12 July 2024 Project Review 31

 References
 T. Tamilvizhi, R. Surendran, K. Anbazhagan, and K. Rajkumar, "Quantum
Behaved Particle Swarm Optimization-Based Deep Transfer Learning
Model for Sugarcane Leaf Disease Detection and Classification", Math.
Probl. Eng., pp. 1-12, Jul. 2022.

 L. Falaschetti, L. Manoni, D. Di Leo, D. Pau, V. Tomaselli, and C.

Turchetti, "A CNN-based image detector for plant leaf diseases
classification", HardwareX, vol. 12, pp. e00363, Oct. 2022.

12 July 2024 Project Review 32

 References
 T. A. Salih, A. J. Ali, and M. N. Ahmed, "Deep Learning Convolution
Neural Network to Detect and Classify Tomato Plant Leaf Diseases",
OALib, vol. 07, no. 05, pp. 1-12, 2020.

 A. Ghosh, A. Sufian, F. Sultana, A. Chakrabarti, and D. De, "Fundamental

Concepts of Convolutional Neural Network" in Recent Trends and
Advances in Artificial Intelligence and Internet of Things, Cham: Springer
International Publishing, vol. 172, pp. 519-567, 2020.

12 July 2024 Project Review 33

 References
 S. Ashok, G. Kishore, V. Rajesh, S. Suchitra, S.G.G. Sophia, B. Pavithra,
Tomato leaf disease detection using deep learning techniques. 2020 5th
International Conference on Communication and Electronics Systems
(ICCES),pp. 979–983, 2020.

 T. Bezdan and N. Bacanin Dzakula, "Convolutional Neural Network Layers

and Architectures", Proceedings of the International Scientific Conference -
Sinteza 2019, pp. 445-451, 2019 .

12 July 2024 Project Review 34

 References
 KuN. Jmour, S. Zayen, A. Abdelkrim, Convolutional neural networks for
image classification. 2018 International Conference on Advanced Systems
and Electric Technologies (IC_ASET), Hammamet, Tunisia, pp. 397–402,
2018

 S. Ramesh, et al., Plant disease detection using machine learning, in:

International Conference on Design Innovations for 3Cs Compute
Communicate Control (ICDI3C), pp. 41–45, 2018.

12 July 2024 Project Review 35