HealTree

The Problem

Farmers and forestry workers often can't identify tree diseases until significant damage is done. Expert consultation is slow and expensive. A tool that provides instant disease identification from a phone camera could enable early intervention and save crops.

The Solution

I built HealTree as a Flutter mobile app backed by a YOLOv8 object detection model.

• Real-time detection — Point the camera at a tree and get instant disease identification with confidence scores.
• YOLOv8 model — Trained on a curated dataset of common tree diseases (leaf spot, blight, rust, canker) with TensorFlow for model optimization.
• Firebase integration — User accounts, detection history, and monitoring dashboards for tracking disease spread over time.
• Offline inference — The model runs on-device via TensorFlow Lite for use in areas without internet connectivity.

What Went Wrong

The initial model had high false positive rates on healthy leaves with natural discoloration (autumn colors, sun damage). It would flag non-diseased trees as infected.

The fix: I expanded the training dataset with a dedicated "healthy with natural variation" class and added data augmentation (color jittering, brightness shifts) to teach the model the difference between disease symptoms and normal leaf variation.

Results

• Real-time disease detection via smartphone camera
• On-device inference for offline use in rural areas
• Firebase-backed monitoring history for tracking disease spread