All projects
04
AI & VisionCompleted
Vision · Model · API
AI Plant Health Assessment
A deep-learning image-classification service, built on ResNet18 and served through a FastAPI endpoint on AWS, that assesses plant health from images and returns structured, actionable responses to connected systems.
Status
Completed
Timeline
2025
Role
ML Engineer
Category
AI & Vision
Technology Stack
PyTorchResNet18FastAPIAWS EC2REST APIImage Classification
Problem
Assessing plant health visually is subjective and doesn't scale across many plants.
Constraints
- Limited labeled data
- Inference latency budget
- Deployable on modest cloud instances
Solution
A transfer-learned ResNet18 classifier exposed as a REST endpoint that returns health class and recommended action.
Key Decisions
- Transfer learning to work with limited data
- Simple REST contract for integration
System Architecture
01Image input
02Preprocessing
03ResNet18 inference
04Class → action mapping
05REST response
Diagram placeholder — replace with a detailed architecture diagram when available.
Hardware
- AWS EC2
Software
- PyTorch
- FastAPI
- REST API
Challenges
- Generalization across lighting/conditions
- Keeping the model lightweight
Results
- Deployed classifier integrated with the indoor agriculture system.
Lessons Learned
- A clean API contract makes the model reusable across projects.
Media
Media placeholder
Media placeholder