Plant Disease Detection - Research Project

Student: Warre Snaet | Institution: Howest MCT

Semi-supervised learning for plant disease classification on embedded edge devices using Rust.

---

📁 Structure

Source/
├── plantvillage_ssl/      # SSL implementation (pseudo-labeling) ✅ WORKS
├── incremental_learning/  # Add new classes (5→6, 30→31 experiments)
├── pytorch_reference/     # Python reference for comparison
├── benchmarks/            # Framework comparison scripts
└── research/              # Literature study, contract, meeting notes

---

🚀 Quick Start

1. Download Dataset (Once)

./download_plantvillage.sh

2. SSL Training

cd plantvillage_ssl
cargo build --release
./target/release/plantvillage_ssl train \
    --data-dir data/plantvillage \
    --labeled-ratio 0.2 \
    --epochs 30 --cuda

3. Incremental Learning

cd incremental_learning
cargo build --release
./target/release/plant-incremental experiment \
    --method lwf \
    --base-classes 5 \
    --new-classes 1 \
    --data-dir ../plantvillage_ssl/data/plantvillage

---

🎯 Research Questions

1. SSL: How efficient is pseudo-labeling on edge devices?
2. Incremental: Is 5→6 harder than 30→31 classes?
3. Data efficiency: How many images needed per new class?

---

📖 Documentation

• plantvillage_ssl/docs/ - Installation & user guide
• research/literatuurstudie.md - Literature review