Field Segmentation Project

Determine which agricultural fields are currently cultivated with plants and which are resting until next growing season using deep learning segmentation models.

Table of Contents

• Project Overview
• Installation
• API Documentation
• Notebook Timeline
• UI Interface
• Metrics

Project Overview

This project combines computer vision and deep learning to analyze satellite/aerial imagery of agricultural fields. The system identifies:

• 🌱 Growing Land: Fields currently under cultivation
• 🛌 Resting Land: Fields lying fallow for seasonal recovery

Key Features:

• U-Net and DeepLabV3 segmentation models
• FastAPI backend for processing requests
• Dockerized deployment
• Data augmentation pipeline
• UI with PySide6

Installation

Prerequisites

• Python 3.8+
• Docker 20.10+
• NVIDIA GPU (recommended for training)

Local Setup

git clone [email protected]:nonezonyx/field_segmentation.git
cd field_segmentation

# Open "saves" directory and download checkpoints from gdrive

# Install dependencies
pip install -r requirements.txt

# Start API server
uvicorn app:app --reload

# or run ui
python3 ui.py

Docker Deployment

docker-compose up --build

API

Endpoint: POST /process-land

Input parameters:

• image: JPEG/PNG file upload

• width: float > 0

• length: float > 0

Response format:

{
"processed_image": "base64_string",
"growing_land": float,
"resting_land": float
}

Access interactive docs at http://localhost:8000/docs after deployment.

Notebook Timeline

1. mask_creation.ipynb - Creating masks from annotations
2. data_augmentation.ipynb - Image augmentation strategies
3. unet_example.ipynb - U-Net model implementation
4. DeepLabv3.ipynb - DeepLabV3+ configuration
5. metrics.ipynb - UNet vs DeepLabv3 results comparison

UI

Simple UI was created using PySide6

Metrics

Overall DeepLabv3 showed better results (see metrics.ipynb)