A full-stack web platform that visualizes pollen-affected areas using crowdsourced plant images.
Built to help users—especially those with allergies—adjust daily routes based on real-time pollen activity.
This project integrates computer vision, GIS visualization, and scalable web architecture.
User-uploaded plant photos are classified by a fine-tuned CNN model, and results are reflected dynamically on a map.
The system bridges frontend (React.js), backend (Flask), and ML model (MobileNetV2) through a clean API pipeline.
- Model: MobileNetV2, fine-tuned on 7 plant species + “Others” category
- Tools: TensorFlow, Keras, NumPy, Matplotlib, Pillow, Pickle
- Flask REST API for model inference and database access
- SQLAlchemy ORM for data management
- Manages flower metadata and geographic tagging of classified images
- React.js interface with Leaflet for map rendering
- Exif-js for extracting GPS and metadata from uploaded images
- Responsive design for both web and mobile
- Frontend: Displays pollen map and handles image uploads
- Backend: Hosts model, processes requests, and updates pollen data
- Databases:
- Flower database (Latin name, bloom period, pollen radius, allergy risk)
- Location database (coordinates and recognized species)
- Demonstrated potential for public health and environmental monitoring
- Created a scalable architecture for future expansion to weather- and wind-aware pollen prediction
- Expand model classes and training data
- Integrate environmental factors such as wind, pollen size, and humidity
- Optimize inference latency for real-time deployment
Developed collaboratively under the guidance of Ritsumeikan University and Universitas Dinamika
Team Members
- Anjal Shrestha
- Brandon Pratama Kwee
- CHEN Dong Fang
- Abigail Excelsis Deo
- M. Haris M.K.A