BMS INSTITUTE OF TECHNOLOGY & MANAGEMENT,

YELAHANKA, BANGALORE.
Department of Artificial Intelligence and Machine
Learning

OPTIMIZING PLANT GROWTH THROUGH DYNAMIC

CLIMATE CONTROL USING ML ALGORITHHMS

RIYAZ UR REHMAN (1BY21AI043)

ROHAN SHARMA (1BY21AI044)
VEDA SHREE S (1BY21AI060)

UNDER THE GUIDANCE OF:

PROF MAYURI KP
ASSISTANT PROFESSOR ,
DEPT. OF AIML
BMSIT&M

2024-25
CONTENT
› Abstract
› Introduction
› Problem Statement
› Objectives
› Literature Survey
› Research Gap and Research Challenges
› Proposed Methodology and Architecture
› Highlights from Phase 1 Review 1 suggestions
› Expected outcomes
› Conclusion
› References
Cisco Confidential
ABSTRACT

This project integrates a network of sensors and sophisticated deep

learning algorithms to continuously monitor and adjust critical
greenhouse parameters, including temperature, humidity, CO2 levels, and
light intensity. By analyzing real-time data, the system optimizes
growing conditions, resulting in improved crop yields and reduced
resource consumption. Utilizing deep learning models that adapt based
on historical and current environmental data, the system automates
climate adjustments tailored to various plant species

Cisco Confidential
INTRODUCTION

This project aims to revolutionize greenhouse agriculture by integrating

advanced technologies for effective plant health monitoring.

It focuses on early disease detection using a hybrid machine learning model

that ensures optimized climate control.

The project also emphasizes data integrity during network outages and energy
efficiency in greenhouse operations to promote sustainable agricultural
practices.

Cisco Confidential
 PROBLEM STATEMENT
• The absence of an integrated model for plant disease detection and climate
control leads to suboptimal growth and increased crop loss. This project will
develop a hybrid model using CNNs and DNNs to enhance growth conditions.
• Network outages cause significant data loss and monitoring interruptions in
greenhouses. This project aims to create a backup system that stores sensor data
locally and synchronizes with the central server when connectivity is restored .
• Limited training data for rare environmental conditions hampers the effectiveness of
predictive models in agriculture. This project will use GANs to generate synthetic
data that simulates these conditions, improving model performance .
• Energy inefficiency in greenhouse operations results in high resource consumption
and costs. This project will integrate optimized algorithms to enhance energy
efficiency while maintaining optimal conditions for plant health.

Cisco Confidential
 OBJECTIVES
1. Develop a hybrid model that integrates a GAN-based convolutional neural
network (CNN) with a deep neural network (DNN) to optimize greenhouse
climate conditions using live sensor data.
2. Develop a reinforcement learning framework to minimize co2 emissions in
greenhouse environments effectively.
3. Implement LSTM networks to forecast and optimize irrigation schedules
based on soil moisture levels.
4. Implement a robust backup system for locally storing sensor data during
network disruptions, ensuring uninterrupted monitoring and preventing data
loss.
5. Develop a gan based cnn model to generate synthetic data for simulating
rare environmental conditions.
Cisco Confidential
 LITERATURE SURVEY
SL Paper Details Techniques / Observations/ Outcomes Limitations/
NO Methodology Used Gaps
1. Climate Modelling Fuzzy logic simplifies Plants grown in the FNN-controlled The Fuzzy Neural Network (FNN) model
Using Fuzzy Neural greenhouse climate control environment showed improved may overfit the training data, leading to
Network Machine by interpreting sensor data growth rates and yields compared to poor performance when handling new
Learning Technique using everyday terms, like traditional rule-based climate control conditions.
(2022) "low humidity" or "high systems . Despite integrating fuzzy logic, the
Khaled Mansouri, temperature," based on FNN enabled real-time autonomous neural network's "black box" nature
Charaf Lachouri expert-defined rules. climate adjustments, reducing the makes it hard to interpret its decision-
need for human intervention and making process. This can reduce user
improving overall system efficiency. control and trust.

2. Technical Paper on Based on real-time The automated control system The accuracy of the system's predictions
Environmental Environmental data, an optimized the use of water, energy, relied heavily on the quality and quantity
Monitor and Control, automated system was and pesticides, reducing resource of the data collected. Incomplete or
Pests and Diseases in developed to control climate consumption while maintaining inaccurate data could lead to suboptimal
the Greenhouse variables (ventilation, plant health . environmental control or incorrect
(2022) irrigation, heating, lighting) The system reduced the reliance on pest/disease detection .
Ediger Mutevani, to maintain optimal chemical pesticides by utilizing
Monica Gondo, conditions for plant growth biological controls and targeted
Edmore Tarambiwa and reduce pest/disease risk. interventions.

3. Real-Time Climate A network of sensors was Users were able to monitor and Setting up the LabVIEW system required
Control and Remote deployed across the control the polyhouse environment technical expertise in both hardware and
Monitoring of polyhouse to continuously remotely, providing flexibility in software, which may not be readily
Polyhouse based on monitor environmental management and reducing the need available in all agricultural settings.
LabVIEW. conditions . Sensors were for physical presence on-site . This could result in additional costs for
Janet Joseph, connected to data The automated control of irrigation skilled labor.
Nitha M Biju2 acquisition hardware, and ventilation led to more efficient The system’s reliance on continuous
interfaced with LabVIEW use of water and energy, reducing network connectivity and electrical
for real-time data wastage and operational costs. power meant that any disruptions could
processing. lead to lapses in monitoring and control,
potentially damaging crops.
Cisco Confidential
SL Paper Details Techniques / Observations/ Outcomes Limitations/
NO Methodology Used Gaps
4 Greenhouse Climate Fuzzy logic simplifies Plants grown in the FNN-controlled The Fuzzy Neural Network (FNN) model
Modelling Using greenhouse climate control environment showed improved may overfit the training data, leading to
Fuzzy Neural by interpreting sensor data growth rates and yields compared to poor performance when handling new
Network Machine using everyday terms, like traditional rule-based climate control conditions.
Learning Technique "low humidity" or "high systems . Despite integrating fuzzy logic, the
(2022) temperature," based on FNN enabled real-time autonomous neural network's "black box" nature
Mohamed M lafifi, expert-defined rules. climate adjustments, reducing the makes it hard to interpret its decision-
Khaled Mansouri, need for human intervention and making process. This can reduce user
Charaf Lachouri improving overall system efficiency. control and trust.

5 Technical Paper on Based on real-time The automated control system The accuracy of the system's predictions
Environmental Environmental data, an optimized the use of water, energy, relied heavily on the quality and quantity
Monitor and Control, automated system was and pesticides, reducing resource of the data collected. Incomplete or
Pests and Diseases in developed to control climate consumption while maintaining inaccurate data could lead to suboptimal
the Greenhouse variables (ventilation, plant health . environmental control or incorrect
(2022) irrigation, heating, lighting) The system reduced the reliance on pest/disease detection .
Ediger Mutevani, to maintain optimal chemical pesticides by utilizing
Monica Gondo, conditions for plant growth biological controls and targeted
Edmore Tarambiwa and reduce pest/disease risk. interventions.

6 Real-Time Climate A network of sensors was Users were able to monitor and Setting up the LabVIEW system required
Control and Remote deployed across the control the polyhouse environment technical expertise in both hardware and
Monitoring of polyhouse to continuously remotely, providing flexibility in software, which may not be readily
Polyhouse based on monitor environmental management and reducing the need available in all agricultural settings.
LabVIEW. conditions . Sensors were for physical presence on-site . This could result in additional costs for
Janet Joseph, connected to data The automated control of irrigation skilled labor.
Nitha M Biju2 acquisition hardware, and ventilation led to more efficient The system’s reliance on continuous
interfaced with LabVIEW use of water and energy, reducing network connectivity and electrical
for real-time data wastage and operational costs. power meant that any disruptions could
processing. lead to lapses in monitoring and control,
potentially damaging crops.

Cisco Confidential
RESEARCH GAP AND RESEARCH
CHALLENGES
Research Gaps :
• Lack of Integrated Systems: No unified model combines plant disease
detection and climate control.
• Inconsistent Data Collection: Network outages cause data loss, disrupting
monitoring and analysis.

Research Challenges :
• Combining Heterogeneous Models: Integrating CNNs and DNNs into a
cohesive system is complex.
• Synthetic Data Generation: Creating diverse, realistic data with GANs
requires advanced techniques.
• Robust Backup Systems: Building reliable local data backup and
synchronization systems is essential.

Cisco Confidential
PROPOSED METHODOLOGY
Sensor Setup:
Install multiple sensors (temperature, humidity, CO2, light).

Data Collection:
Preprocess sensor data and use synthetic data to
simulate rare events.

Model Development
CNN: Train on plant images for disease detection.
DNN: Train on sensor data for climate predictions.

Hybrid Model Integration:

Combine CNN and DNN to create a unified system for disease detection and climate
optimization.

Energy Optimization:
Minimize water and energy usage.
Cost-Benefit Analysis: Compare optimized resource usage with traditional systems.

Synthetic Data for Rare Events GANs for Data Augmentation:

Simulate rare environmental events and pest outbreaks.
Cisco Confidential
Cloud-Based Scalability
Cloud Integration: Host DNN and RL on cloud for scalability.
Remote Monitoring: Enable real-time, remote access and control.

Robustness and Backup Sensor Redundancy: Ensure continuous data with predictive
algorithms.
Local Backup : Offline data processing and power backup systems.

Pest and Disease Detection Real-Time Monitoring:

Integrate image recognition for pest detection.
Continuous Retraining: Retrain DNN with new pest/disease data.

Reinforcement Learning Process:

Dynamic adjustments based on real-time feedback.
Optimize plant growth and resource efficiency.

Cisco Confidential
PROPOSED ARCHITECTURE
GAN DATA
STORAGE
Sensor data

DNN Climate control systems

(Energy Optimization
HVAC)

CNN

REAL TIME MONITORING

ALERT SYSTEMS

Cisco Confidential
 How CNN model works
CNN Architecture:
• Convolution Layer: Extracts features (edges, textures) using filters to create
feature maps.
• Activation Function (ReLU): Adds non-linearity for learning complex
patterns.
• Pooling Layer: Reduces dimensionality and computational complexity (e.g.,
Max Pooling).
• Fully Connected Layer: Maps extracted features to class probabilities.
• Output Layer: Provides classification results (Softmax for multi-class,
Sigmoid for binary).

Inputs and Outputs:

• Inputs: Images (e.g., RGB images of plant leaves, size 128×128×3).
• Outputs: Predicted class or label (e.g., "Healthy" or "Diseased")

Cisco Confidential
 How DNN model works

The DNN processes real-time sensor data and predicts optimal settings for:
• HVAC controls to maintain ideal temperature.
• Lighting and water management for energy-efficient plant growth.

CNN input : plants image

CNN Output: Disease status and severity
like-{'Disease': 'Fungal Infection', 'Severity': 'High'}
DNN Input: sensors data and output of CNN
like-{'Temperature': 30, 'Humidity': 75, 'Disease_Severity':
'High'}
DNN Output: {'Optimal_Temperature': 28, 'Optimal_Humidity': 60}
Final Action: Adjust environmental conditions and notify operators.

Cisco Confidential
HIGHLIGHTS FROM PHASE 1
REVIEW 1 SUGGESTIONS
Review and suggestion:
• Suggested to include additional papers for literature survey.
• Suggested to work on real time data collection.

Solution to review and suggestion:

• Referred 30 more papers for literature survey.
• Working on real time data collection.

Cisco Confidential
EXPECTED OUTCOMES
Improved accuracy in plant disease detection and enhanced climate
control, leading to healthier crops and higher yields .
A reliable backup system that maintains continuous data monitoring and
ensures data integrity during network outages .
Increased energy efficiency in greenhouse operations, promoting
sustainable agricultural practices and reducing operational costs.

Cisco Confidential
CONCLUSION

This project highlights the role of advanced technologies in improving

greenhouse agriculture through effective disease detection and climate
optimization . the reliable data backup system ensures continuous
monitoring and data integrity, even during network outages .By
promoting energy efficiency and sustainable practices, the project aims
to enhance food security and support environmentally friendly
agriculture.

Cisco Confidential
 REFERENCES
[1].Mohamed M lafifi, Khaled Mansouri, Charaf Lachouri“Greenhouse Climate
Modelling Using Fuzzy Neural Network Machine Learning
Technique”,2022,ISSN :2319-7064,P ID:ART20161295.
[2]Ediger Mutevani, Monica Gondo, Edmore Tarambiwa,“Technical Paper on
Environmental Monitor and Control, Pests and Diseases in the
Greenhouse”,2022, IJISRT22MAR204 ,ISSN NO:-2456-2165.
[3]Janet Joseph, Nitha M Biju2, “Real-Time Climate Control and Remote
Monitoring of Polyhouse based on LabVIEW”,2023,IJSR,ISSN:2319-
7064,ART20161295.
[4]V. Suresh, D. Gopinath, M. Hemavarthini, K.Jayanthan, and M. Krishnan,
“Plant Disease Detection using Image Processing,” vol. 9, no. 03, pp. 78–82,
2023.
[5] Aria Dolatabadian, Ting Xiang Neik, “Image-Based Crop Disease Detection
Using Machine Learning,” 2024, Plant Pathology, ISSN: 0032-0862.
[6] Weiying Zhao, Natalia Efremova, “Grapevine Disease Prediction Using
Climate Variables from Multi-Sensor Remote Sensing Imagery via a
Transformer Model,” 2024, arXiv, ID: arXiv:2406.07094.
Cisco Confidential
[7] Shashwat Jha, Vishvaditya Luhach, Gauri Shanker Gupta,
Beependra Singh, “Crop Disease Classification Using Support Vector
Machines with Green Chromatic Coordinate (GCC) and Attention-Based
Feature Extraction for IoT-Based Smart Agricultural Applications,” 2023,
arXiv, ID: arXiv:2311.00429.

[8] Jianping Yao, Son N. Tran, Samantha Sawyer, Saurabh Garg,

“Machine Learning for Leaf Disease Classification: Data, Techniques,
and Applications,” 2023, arXiv, ID: arXiv:2310.12509.

[9] Novel Edge Device System for Plant Disease Detection with Deep
Learning," 2024, International Journal of Intelligent Systems and
Applications in Engineering, ISSN: 2147-6799.

[10] Trends in Machine and Deep Learning Techniques for Plant

Disease and Pest Detection,” 2024, MDPI Agriculture, ISSN: 2077-
0472.

Cisco Confidential