🐟 FIRMATECH-BACKEND-AI

Smart aquaculture AI backend built with FastAPI — enabling fish counting, disease detection, biomass estimation, and more. Designed for real-world farm automation with modular service architecture and AI integration.

---

📁 Table of Contents

• Features
• Architecture
• AI Models & Evaluation
• Endpoints Overview
• Project Structure
• Getting Started
• Environment Setup
• Future Work
• License

---

✅ Features

Module	Description
🎥 Fish Counting	Real-time video-based fish counting using YOLOv8 + Centroid Tracking
⚖️ Weight Estimation	Biomass calculation via length-based prediction formula
🩺 Disease Detection	Classify fish sickness from image (e.g., Aeromonas, Tilapia Virus)
🧪 Farm Monitoring	Generate natural language reports from sensor data using Gemini API
🍽️ Food Prediction	(Planned) Predict fish feeding needs based on AI heuristics

---

🧱 Architecture

FastAPI
🔼 Routers (API Endpoints)
🔼 Controllers (Logic Handlers)
🔼 Modules (Pydantic Schemas)
🔼 Services (Roboflow / Agent)
🔼 Utils (Tracking, Math utils)

Uses Roboflow for AI inference and Gemini Pro for report generation.

---

📊 AI Models & Evaluation

🤖 1. Fish Detection & Counting (YOLOv8)

• Model: Custom YOLOv8m (Roboflow)
• Input: Video frames
• Output: Bounding boxes of detected fish
• Post-processing: CentroidTracker avoids double counting

🧬 2. Disease Classification

• Model: Roboflow classification model
• Classes:
  • Healthy-Fish
  • Tilapia Lake Virus
  • Bacterial Aeromonas Disease
  • Streptococcus
  • null (uncertain)

{
  "isSick": true,
  "sicknessType": "Streptococcus"
}

⚖️ 3. Weight Prediction

• Formula:

  Weight (g) = 0.0196 × Length(cm)^2.9868
  

• Calibration: Static for testing (pixel → cm)

• Assumption: Sampled fish weight is applied to whole tank

---

📱 Endpoints Overview

Route	Method	Description
/fish-count	POST	Real-time fish count from video
/weight	POST	Predict fish weight
/disease	POST	Detect fish sickness
/fish-farm/report	POST	Generate sensor-based report
/food	🔒	(Planned) Food AI estimation

---

🧠 Diagram Snapshots

Fish Counting

Weight Prediction

Disease Detection

Daily report generation

---

📁 Project Structure

FIRMATECH-BACKEND-AI/
👉 main.py                 # App entry
👉 controllers/            # Endpoint logic
👉 routers/                # FastAPI route definitions
👉 modules/                # Schemas and models
👉 services/               # Agent/Roboflow integrations
👉 utils/                  # Tracking, calibration
👉 diagram/                # Diagrams
👉 Dockerfile, docker-compose.yml
👉 .env, .env.example
👉 requirements.txt

---

🚀 Getting Started

🔧 Local (Python)

git clone https://github.com/your-username/FIRMATECH-BACKEND-AI.git
cd FIRMATECH-BACKEND-AI
python -m venv venv
source venv/bin/activate   # or venv\Scripts\activate on Windows
pip install -r requirements.txt
uvicorn main:app --reload

🐳 Docker (Recommended)

docker-compose up --build

---

⚙️ Environment Setup

Copy .env.example to .env and set values:

ROBOFLOW_API_URL=your_key
ROBOFLOW_API_KEY=your_model_url
ROBOFLOW_DETECTION_MODEL_ID=your_model_url
ROBOFLOW_DISEASES_MODEL_ID=your_model_url
GOOGLE_API_KEY=your_gemini_key

---

🔬 Future Work

• Improve fish length estimation using adaptive calibration and pixel-to-cm mapping
• Enhance prediction accuracy with multi-angle sampling or better lighting
• Expand the dataset for disease detection across seasons