Wetlands Prototype MVP

Index

1. Introduction
2. Backend Overview
3. Frontend Overview
4. EXIF Metadata Extraction
5. Prompt options- Identifying Wetland Organisms
6. Lookup Tables vs ENUM
7. React Frontend Setup
8. How to Start the Frontend
9. Recommended Photo Sites
10. Diagnosing Performance Issues

---

Introduction

This repository contains the Wetlands Prototype MVP, a project designed to analyze wetland images, extract metadata, and provide insights using AI models. The project includes a FastAPI backend, a React frontend, and PostgreSQL for data storage.

---

Backend Overview

Architecture

• API Layer: Handles HTTP requests and routes them to appropriate services.
• Service Layer: Contains business logic and interacts with the database and AI models.
• Database: Stores structured data for the application.
• AI Model Integration: Embeds AI models for inference tasks, such as image recognition and data analysis.

Key Features

• Frameworks: Built using FastAPI for lightweight and efficient API handling.
• Database: Utilizes PostgreSQL for relational data storage.
• AI Models: Integrated using pre-trained models for specific tasks.
• Environment Configuration: Managed through .env files for secure and flexible deployment.

---

Frontend Overview

Key Features

• Responsive Design: Ensures compatibility across devices of various screen sizes.
• React with Vite: Provides a fast development environment and optimized builds.

---

EXIF Metadata Extraction

Why EXIF?

• Widespread Support: Embedded in most modern image formats (e.g., JPEG, TIFF).
• Rich Metadata: Includes date-time, camera settings, GPS coordinates, and more.
• Ease of Use: Libraries like Pillow make it straightforward to extract EXIF data.

Limitations

• Read-Only: EXIF metadata is primarily read-only.
• Not Universal: Some images (e.g., screenshots) may lack EXIF metadata.

---

Prompt options - Identifying Wetland Organisms

Prompt Versions

1. Multiple Choice: Provides 1-3 possible identifications with a fact about the organism.
2. Most Likely Identification: Suggests the most likely organism visible in the image.
3. Hidden Inhabitant: Suggests an organism likely present but not visible based on habitat and metadata.

---

Lookup Tables vs ENUM

Why Lookup Tables?

• Scalability: Add new categories without schema changes.
• Flexibility: Supports regional variations and external GIS integration.
• Analytics: Easier to analyze evolving categories.

---

React Frontend Setup

Recommendations

• Start with JavaScript and React Compiler for simplicity and modern tooling.
• Transition to TypeScript in later phases if needed.

---

How to Start the Frontend

1. Install Dependencies:

   npm install

2. Start the Development Server:

   npm start

3. Open in Browser: Navigate to the URL displayed in the terminal (e.g., http://localhost:3000). #Options through development:

---

Recommended Photo Sites

• Wolf River Greenway: Explore the Greenway
• Wolf River Interactive Map: View Map
• Bridges over culverts

---

Diagnosing Performance Issues

Common Issues

• High response times during large image uploads.
• Increased failure rates on metadata extraction endpoints.
• Fluctuating request rates causing slowdowns.

Steps to Resolve

1. Identify Log Patterns:
   • Use Azure Monitor and Application Insights to analyze logs.
   • Example KQL Query:

     requests
     | where timestamp > ago(1h)
     | summarize count() by resultCode, bin(timestamp, 5m)

2. Set Alerts:
   • Define alert rules for high CPU usage, response times, and failure rates.
3. Propose Corrective Actions:
   • Short-term: Scale out resources.
   • Long-term: Optimize database queries and improve AI model efficiency.