🌿 We built an AI that detects plant diseases — running entirely on a Raspberry Pi, with no internet required.
After months of research, development, and real-world testing, I'm proud to present HydroNova — an end-to-end intelligent hydroponic greenhouse system built as my final year engineering project at ISET Mahdia, in collaboration with MicroEdition.
━━━ The Problem ━━━
Hydroponic farmers lose 20–40% of their crops every year to plant diseases that go undetected until it's too late. Traditional monitoring is manual, slow, and inaccessible to small-scale farmers — especially in regions with poor internet connectivity.
━━━ What We Built ━━━
🤖 Edge AI Disease Detection
Instead of sending images to the cloud, our Raspberry Pi 4 runs a Multi-CNN ensemble model (EfficientNet-B4 + ConvNeXt-Tiny + InceptionV3) orchestrated by META-DES — a meta-learning algorithm that dynamically selects the most competent model for each individual image. Result: 98.96% F1-Score, 110ms inference time, zero cloud dependency.
🌍 Multilingual AI Assistant
We fine-tuned PaliGemma (Google DeepMind's Vision-Language Model) on hydroponic plant diseases. Combined with a RAG-based chatbot, farmers can ask questions about their crops in Arabic, French, or Tunisian dialect — and get contextual answers powered by their real greenhouse data.
📡 Complete IoT Pipeline
The system reads sensors (temperature, humidity, pH, EC), controls actuators (irrigation pump, LED lights, ventilation), and streams everything in real time from the physical serre → MQTT → Spring Boot 4 backend → Flutter mobile app — in under 5 seconds, even in offline mode.
━━━ Key Technical Concepts ━━━
→ Dynamic Ensemble Selection (META-DES) for plant pathology — a first for agricultural edge deployment
→ ONNX/FP16 quantized inference running natively on ARM (Raspberry Pi 4)
→ Serre → Zone → Appareil isolation architecture for multi-greenhouse multi-tenant scalability
→ VLM fine-tuned with QLoRA on PlantVillage VQA dataset
→ Clean Architecture (Spring Boot 4) with 19 isolated MQTT topics and WebSocket real-time bridge
━━━ The Result ━━━
An autonomous greenhouse that monitors itself, detects diseases before they spread, controls its own environment, and explains everything to the farmer in their own language — even when there's no internet.
This project taught me that the most powerful innovations don't always require the biggest infrastructure. Sometimes they just require putting the right intelligence in the right place.
🙏 Huge thanks to my co-author Ahmed Milien, our supervisor Hamza Sassi, and the entire team at MicroEdition for making this possible.
#AI #IoT #EdgeAI #MachineLearning #ComputerVision #SmartAgriculture #Hydroponic #EmbeddedSystems #Flutter #SpringBoot #MQTT #PFE #Engineering #Innovation n
1 Agosto 2026 à 18h15
Ver en LinkedIn →
