EDGE AI SYSTEMA+

Farmer Eye Robotic Car

Edge AI system for real-time plant disease detection and autonomous field monitoring using computer vision and IoT.

King Salman International University · B.Sc. Computer Science AI · 2025

View GitHub Watch Demo

TensorFlowOpenCVRaspberry PiEdge AIIoT

Overview

The Problem

• Manual plant health monitoring is time-consuming and prone to human error.
• Inefficient use of resources (water, fertilizers) due to lack of real-time data.
• Critical factors like soil temperature and wind often remain undetected.
• Delayed identification of diseases leads to significant crop yield losses.
• Absence of data-driven tools limits farmers' ability to make timely decisions.

The Solution

Farmer Eye is an AI-driven autonomous vehicle ecosystem. It navigates farm fields collecting real-time data through high-resolution cameras and sensors. It analyzes plant health metrics using Machine Learning to detect diseases, assess pest infestations, and provide actionable insights via a user-friendly mobile app, ensuring sustainable and precise agriculture.

System Architecture

Subsystem engineering index: Hardware robotics, edge cognitive inference, mobile control, and data pipeline.

Smart Vehicle (Robotics)

Chassis & Mobility: Custom-built RC car chassis powered by high-torque motors and coordinated via an Arduino microcontroller.

Sensors & Geolocation: Integrated DHT sensors for environmental ambient monitoring (temperature & humidity) paired with a high-precision GPS module for spatial geotagging of scanned crops.

Clean Power: Onboard solar panel array providing continuous battery replenishment during field sweeps.

AI & Cognitive Core

Edge Optimization: Neural network models quantized to TensorFlow Lite for low-latency real-time inference on the Raspberry Pi 4 edge device.

Vision Acquisition: High-definition 8MP camera module capturing real-time leaf foliage streams.

Deep Learning Architecture: Transfer learning pipeline fine-tuning pre-trained CNN architectures across the PlantVillage Dataset.

→ Accuracy: 97.95% · Macro F1-Score: 0.98 (25 Classes)

Mobile Command Center

Field Client: Cross-platform Flutter mobile application delivering real-time diagnostic reports, disease geolocation maps, and targeted treatment guidance.

Real-Time Telemetry: Asynchronous streaming connecting the rover via edge APIs and Firebase/Firestore for live field updates and historical logging.

Data Engineering

Image Preprocessing: Strict pixel rescaling (0–1 normalization) and fixed 224x224 dimension cropping for consistent batch tensor shapes.

Augmentation Pipeline: Synthetically expanded dataset via spatial operations (Rotation, Shear, Zoom, Shift) to prevent overfitting.

Dataset Scale: Trained on 54,000+ expert-annotated images covering healthy and diseased crop states.

My Role

I spearheaded the development of the system's cognitive core, acting as the AI Architect. My primary focus was constructing the 'brain' of the rover:

•Model Architecture Design: Designed and implemented specific Deep Learning architectures (CNNs) using Keras and TensorFlow to maximize detection accuracy on edge devices.
•Edge Optimization: Optimized heavy neural networks to run efficiently on the Raspberry Pi 4, ensuring low-latency inference without internet dependency.
•Data Engineering Pipeline: Engineered robust data pipelines for preprocessing real-time camera feeds and handling the massive PlantVillage dataset for training.

Impact & Results

97.95%

High Accuracy

Achieved classification accuracy on the PlantVillage dataset using fine-tuned CNNs.

~200ms

Real-Time Inference

Optimized performance on Raspberry Pi with inference times of per image.

Comprehensive Coverage

Supports detection of distinct plant disease classes across multiple crop species.

70%

Efficiency Boost

Reduced manual crop inspection time by approximately through automation.

Challenges & Learnings

• Edge Deployment ConstraintsRunning complex CNNs on a Raspberry Pi 4 was computationally expensive. Solution: We utilized TensorFlow Lite quantization to reduce model size by 75% without significant accuracy loss.

• Real-time Inference LagBalancing frame rate with detection accuracy was critical. Solution: Optimized the inference pipeline to process frames at ~5 FPS for a slow-moving agricultural rover.

• Handling Class ImbalanceThe dataset had unequal samples for some disease classes. Solution: Applied advanced data augmentation (Shear, Zoom) and class weighting during training.

• IoT + AI IntegrationSynchronizing the rover's movement with cloud data streaming was complex. Solution: Architected an asynchronous event loop using Python's asyncio to handle motor control and Firebase updates concurrently.

Recognition

OFFICIAL PRESS MENTION2025

KING SALMAN INTERNATIONAL UNIVERSITY — OFFICIAL NEWS

Outstanding Grant Awardees — ASRT Research Funding

The university officially recognized this project for reaching the finals of the ASRT (Academy of Scientific Research & Technology) competition, winning a research grant for the project "Smart Vehicle for Plant Disease Detection Using AI and IoT".

Read official announcement

LINKEDIN POSTDr. Saeed Mohsen · 2025

DR. SAEED MOHSEN — PROJECT SUPERVISOR

"As a graduation project supervisor at KSIU University —Smart Vehicle for Plant Disease Detection and Classification Using AI and IoT (Farmer Eye Robotic Car). Thanks to my best team and best wishes."

View on LinkedIn