QLearn - Neural Network Training for QGIS

QLearn is a QGIS plugin that allows you to train a UNet segmentation neural network architecture to segment and classify raster data. It can also use pretrained models to make predictions on raster data.

GitHub Repo QGIS Plugin Repository

Key Features

• Train neural networks directly from QGIS

• Use the UNet architecture for segmentation tasks

• Process and prepare raster data for training

• Make predictions using trained models

• Visualize results within QGIS

The plugin integrates with QGIS’s Processing Framework, allowing you to easily incorporate machine learning into your geospatial workflows.

Requirements

• QGIS 3.26+ (earlier versions are untested but may work)

• torch and torchvision Python packages

• OsGeo4w (Reccomended)

• Windows 10+ (Linux and MacOS are untested but may work)

Contents:

• Tutorial
  • Getting Started with QLearn
  • Installation
  • Basic Training Workflow
• Settings & Features
  • Basic Training Parameters
  • Model Architecture Parameters
  • Data Processing Parameters
  • Training Process Parameters
  • Advanced Options
  • Prediction Parameters
• FAQ / Help
  • Frequently Asked Questions
  • Troubleshooting
  • Tips and Tricks
  • Getting Help
• Examples
  • Example 1: Water Classification
  • Example 2: Land Cover Classification
• About the Project
  • Introduction
  • Features
  • Methods