AGRI-SENSE AGRICULTURE PROJECT PROGRESS

25-26J-001

05/01/2026 1

Project supervisor: Prof. Anuradha Jayakody

Co-Project supervisor: Ms. Narmada Gamage

IT22091048 | YOMAL.M.S

IoT and Data Propagation, Liquid Distribution and Robot Navigation

Specialization: CSNE

IT22091048 | YOMAL.M.S | 25-26J-001

07/01/2026 2

RELEVANCE AND ABUSE FILTERATION

​

Problem:

• Manual irrigation in Tomato farming leads to water wastage and low yield.
• Existing IoT solutions depend on continuous cloud connectivity, which is unreliable in rural Sri Lanka.

​

Our Solution:

Current systems → unstable and energy inefficient.

Agri-Sense: LoRa-based wireless sensor network + Raspberry Pi edge gateway.

Ensures real-time monitoring & automated irrigation.

Operates reliably even in low-connectivity farms.

​

IT22091048 | YOMAL M.S | 25-26J-001

07/01/2026 3

Initial Implementation:

• Environmental sensors were successfully integrated

with the ESP32 platform.

• A rule-based irrigation control mechanism driven by

soil moisture, temperature, and humidity measurements

was implemented.

• Established reliable UART-based communication between

the ESP32 and the LoRa transceiver module

• Point-to-point LoRa data transmission was tested and

confirmed for reliability

• verified stable serial data transfer between the ESP32

receiver node

​

​

​

​

IT22091048 | YOMAL M.S | 25-26J-001

07/01/2026 4

PROGRESS UNTIL NOW...

IT22167996 | Bandara K M D I

SDN and Robot Navigation

Specialization: CSNE

IT22167996 | Bandara K M D I | 25-26J-001

07/01/2026 5

Problem:

• Smart farming adoption is limited by unstable connectivity, low bandwidth, and delayed sensor data.
• Current systems lack edge processing, SDN-based traffic management, VPN security, and failover capabilities, limiting scalability and reliability.
• Farmers need a semi-autonomous system to follow predefined paths for spraying, irrigation, harvesting, or monitoring.

​

Our Solution:

• Edge Computing: Raspberry Pi for local processing, encoding, and decision-making.
• Software-Defined Networking (SDN): ONOS-based controller for dynamic traffic prioritization and QoS with secure networking.
• Computer Vision / Sensor Processing: IR or camera sensors for path following (white strip detection).

​

SDN AND ROBOT NAVIGATION

IT22167996 | Bandara K M D I | 25-26J-001

07/01/2026 6

As of now:

• Successful SDN Controller setup in Azure.
• Successful OpenV Switch implementation.
• StrongSwan IPsec tunnel setup.

​

Challenges:

• Control and Data plane abstraction in test environment.

​

Upcoming tasks:

• Implement the tested environment in Raspberry Pi.
• Setup and optimize time sensitive traffic flow to cloud.
• Finalize navigation with robot construction.

PROGRESS UNTIL NOW...

IT22167996 | Bandara K M D I | 25-26J-001

07/01/2026 7

IT22032874 | Weerasekara W M G V

​

Autonomous tomato harvesting,

Data pipeline & Web dashboard

Specialization: CSNE

IT22032874 | Weerasekara W M G V | 25-26J-001

​

07/01/2026 8

Problem:

• Manual tomato harvesting is labor-intensive, slow, and prone to fruit damage.
• High labor costs and inconsistent harvest quality.
• Lack of real-time soil and environmental data (moisture, pH, temperature).
• Decisions on irrigation, fertilization, and crop management are often based on estimation.

​

Our Solution:

• Robotic Arm for Automated Harvesting: Picks tomatoes carefully and efficiently, reducing labor and fruit damage.
• Data Pipeline: Securely transmits sensor data to the cloud for storage and processing.
• Web-Based Visualization Dashboard: Displays real-time data to help farmers make data-driven decisions.
• Outcome: Improved yield, better crop quality, reduced labor dependency, and precision agriculture adoption.

​

AUTONOMOUS TOMATO HARVESTING, DATA PIPELINE & WEB DASHBOARD

​

IT22032874 | Weerasekara W M G V | 25-26J-001

07/01/2026 9

Robotic Arm:

• A 3D-printed gripper is developed to handle tomato peppers delicately, minimizing damage to the fruit and plant.
• The arm structure uses lightweight but sturdy materials for stability and low energy consumption.
• Actuation is achieved using servo and stepper motors for precise joint movements.

​

Ripeness Detection System:

• Dataset of ripe and unripe tomatoes was collected and organized into training, validation, and test.
• A Convolutional Neural Network (CNN) with MobileNetV2 as the backbone was trained on this dataset, achieving 98% training accuracy, 96% validation accuracy, and 92.8% test accuracy.
• For real-time testing, OpenCV was used to capture webcam frames, preprocess them, and pass them through the CNN to classify each tomato as ripe or unripe.
• To improve detection, we integrated YOLOv8n object detection, which identifies tomatoes in each frame, draws bounding boxes, and then passes the detected regions to the CNN for ripeness classification.
• This system supports multiple tomatoes, provides real-time visual feedback, and is ready for deployment on lightweight platforms such as the Raspberry Pi.

​

PROGRESS UNTIL NOW...

IT22032874 | Weerasekara W M G V | 25-26J-001

07/01/2026 10

IT22093264 | Nilessh P.

​

Leaf Disease Detection, Automated Chemical Spraying & Cloud Infra Provisioning

Specialization: CSNE

IT22093264 | Nilessh P. | 25-26J-001

07/01/2026 11

LEAF DISEASE DETECTION, AUTOMATED CHEMICAL SPRAYING & CLOUD INFRA PROVISIONING

IT22093264 | Nilessh P. | 25-26J-001

07/01/2026 12

Problems

• Manual detection of tomato leaf diseases is time-consuming and inaccurate, leading to delayed treatment.
• Cloud-based systems cause latency, high bandwidth usage, and connectivity issues, limiting real-time response.
• Limited integration between disease detection and automated chemical spraying reduces efficiency and increases chemical usage.

​

Our Solution

• Edge AI: Raspberry Pi–based detection of 7+ tomato leaf diseases with low latency.
• Automation: Robotic arm–assisted targeted chemical spraying using X, Y, Z coordinates.
• Secure Deployment: Containerized deployment with secure, reliable, and cost-efficient infrastructure provisioning (IaC).

​

PROGRESS UNTIL NOW...

IT22093264 | Nilessh P. | 25-26J-001

07/01/2026 13

As of Now .....

• Successfully trained the model to detect 7+ tomato leaf diseases
• Evaluated the model for accuracy and latency, with results meeting expected benchmarks
• Successfully deployed the model on an edge device (Raspberry Pi) using Docker
• Tested data communication between the controller (Raspberry Pi) and actuator controller (ESP32) for:

Robotic arm movement angles

X, Y, Z coordinate positioning(Pinhole camera

model)

​

​

Upcoming Tasks .....

• Test the end-to-end workflow with real-time robotic arm movements & Chemical Spraying
• Perform infrastructure provisioning for production deployment (edge + supporting services)

​

07/01/2026 14

25-26J-001