Skip to content

README.md

Latest commit

 

History

History
432 lines (301 loc) · 31.8 KB

README.md

File metadata and controls

432 lines (301 loc) · 31.8 KB

Geospatial Artificial Intelligence (GeoAI) / Geospatial Machine Learning (GeoML) Compendium:

Resources 👨‍💻 focused on land cover monitoring 🏞️🛰️

🚨 This compendium is a work in progress ➡️ also see the repo's associated Updates doc. 🚨

DOI Update Visitor Badge

Overview

  1. Objectives
  2. Schematics
  3. Resources: Machine learning (classical)
  4. Resources: Deep learning
  5. Resources: Uncertainty assessment / accuracy assessment
  6. Resources: Other
  7. Software installers

1. Objectives

  • Document the various Machine Learning approaches available across various [desktop + cloud] platforms (e.g. ArcGIS Pro, ENVI, ERDAS Imagine, GEE, QGIS, R, SNAP, TerrSet)
    • Compile resources (including tutorials and sample scripts) regarding how to implement those approaches in the platforms listed ⬆️
    • Understand how to replicate ML workflows from one platform to another
  • Understand the limitations of the various ML approaches 🤔
    • Push the limits of the various approaches
    • For instance, what determines the taste of the cake? The quality of the ingredients used, or the technique of the baker? 🤔

2. Schematics

  • Overview of the distinctions between Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL)

source: Human Centered AI Lab (2017)

Therefore ⬇️ we will focus on the broader ML and the DL aspects of geospatial artificial intelligence. 🧐

source: SciKitLearn

3. Resources: Machine learning (classical / traditional)

These are resources for commonly-used software applications.

3.a. Overview of classical / traditional ML

Let's start with an overview of Machine Learning.

source: Google (2022)

By the way, how does classical / traditional ML differ from Deep Learning? 🤔

source: Google (2024)

3.b. Classification algorithms

3.b.1. Unsupervised classification algorithms

K-means

source: B. Howell (2006)

ISODATA

source: B. Howell (2006)

source: J. Jensen (2005)

source: J. Jensen (2005)

3.b.2. Supervised classification algorithms

Minimum Distance

source: B. Howell (2006)

Maximum Likelihood

source: B. Howell (2006)

Support Vector Machine

source: SciKit-Learn (2024)

Random Forest

source: RUS-Copernicus (2017)

Other: Miscellaneous supervised ML classification output options in Earth Engine ➡️ some classifiers output class probability maps

source: Google (2022) | for viewing class probabilities, see this

3.c. Classical / traditional ML in commonly-used software applications

ArcGIS

  • Classifying Images in ArcGIS Desktop 10.4 tutorial
  • Train a 'Random Trees' classifier in ArcGIS Desktop tutorial
    • Random Forest explainer video 🎥
  • Overview of image classification in ArcGIS Pro tutorial

QGIS

  • Supervised classification using the Semi-automatic classification Plugin (SCP)’s various algorithms and Sentinel-2 data in QGIS tutorial

ENVI

  • Classification (supervised + unsupervised ML) workflow overview
  • Classification tutorial

SNAP

  • Unsupervised classification of land cover using Sentinel-1 in SNAP tutorial
  • Supervised classification (Random Forests) of forests using Sentinel-2 in SNAP tutorial
  • Land cover classification (unsupervised, supervised) using Sentinel-1 in SNAP tutorial

Google Earth Engine (GEE)

4. Resources: Deep Learning (DL)

"Everybody else is doing [Deep Learning], so why can't we?"

source: modified from the Cranberries' "Everybody Else is Doing It, So Why Can't We?" album cover (1993)

TensorFlow resources

source: Google

4.a. General DL resources from organizations

Resources from Development Seed

Resources from SERVIR

  • SERVIR (2019): TensorFlow basics overview by Kel Markert
    • Slides ➡️ complementary to Google’s 2022 DL slide deck
    • Colab notebook
  • SERVIR GeoAI Working Group (alternatively the Geo-AI Working Group or Geo AI Working Group, and formerly the TensorFlow Working Group) resources (2019-present)
  • SERVIR GeoAI Working Group presentation videos (~2022-present) 🎥
  • Overview of how GEE connects to Google Cloud and AI (Sept. 2019): "Understanding the workflow between the Tensorflow library, Google Cloud Platform, Google Earth Engine, and Google Colab is a challenging undertaking..." (T. Mayer / SERVIR SCO)

Resources from Spatial Thoughts

source: U. Gandhi / Spatial Thoughts (2024)

Resources from Google

  • Google (2019): Overview of the connections between GEE, Google Cloud, and the AI Platform (now formally, "Vertex AI"):

source: N. Clinton, C. Brown / Google (2019)

source: Google (2022)

Resources from other sources

4.b. Software-specific DL resources

ArcGIS-related resources

source: Esri (2021)

source: screenshot from ArcGIS Pro v. 3.3

QGIS-related resources

  • General list of QGIS plugins referencing DL
  • QGIS Deepness: Deep Neural Remote Sensing plugin (2024) 🆕
  • QGIS Deep Learning Tools plugin (2021)
  • QGIS Deep Learning Datasets Maker plugin (2022)

ENVI-related resources

  • Overview of the ENVI Deep Learning Process page
  • ENVI Deep Learning Guide Map
  • What's New in ENVI Deep Learning 3.0 page
  • ENVI Deep Learning system requirements page
  • Train Deep Learning Models Using the ENVI Modeler page

R-related resources

  • NeuralNet package basic documentation
  • Torch (PyTorch) package basic documentation
    • ReddIt thread re: packaging of PyTorch / TensorFlow in R → “Has anyone else noticed that there do not seem to be many packages in R that allow for Neural Networks and Deep Learning (with the exception of "nnet")? It seems that any time and R user would like to fit Neural Networks, they are "forced" to use the R version of "keras" (through "reticulate") - ultimately, the fitting of the Neural Network is done behind the scenes in Python.”
  • DataCamp tutorial for building a neural network in R using NeuralNet, Keras, and TensorFlow packages
  • Udemy course preview video 🎥 about neural networks in R (2019) → Udemy course

5. Resources: Uncertainty assessment / accuracy assessment

source: Sara A. Metwalli / Towards Data Science (2020)

source: J. Brownlee (2020)

6. Resources: Other

GEE

Miscellaneous

Neural networks + deep learning in general

  • Neural networks explainer video 🎥 from Esri ➡️ great contrast w/ Random Forest explainer video 🎥

7. Software installers

Citation

If this documentation is used in publications, presentations, or other venues, please cite 📝 the following:

Cherrington, E. (2025). Geo AI Compendium (Version 1.0.0) [Document]. https://doi.org/10.5281/zenodo.16735094

DOI

Contact information

If you have any questions, feel free to contact Emil Cherrington by 📩 email: emil.cherrington [at] uah.edu.