satellite-image-deep-learning/techniques

What it is

A curated, open-source repository that collects deep learning techniques, models, and algorithms specifically for analyzing satellite and aerial imagery. It's a reference hub — a long, searchable list of techniques and linked example projects.

What it covers

The repo organizes techniques into categories, including:

Classification — assigning a label to an entire image (e.g., 'urban', 'forest'). The README notes this is different from pixel-level classification (segmentation).
Segmentation — assigning a class label to every pixel, turning an image into a labeled grid. Examples given: road/building segmentation, land use and crop type classification.
Object detection
Regression
Cloud detection & removal
Change detection
Time series
Crop classification
Crop yield & vegetation forecasting
Generative networks
Autoencoders, dimensionality reduction, image embeddings & similarity search
Few & zero shot learning
Self-supervised, unsupervised & contrastive learning
SAR (synthetic aperture radar)
Explainable AI (XAI)
Large vision & language models (LLMs & LVMs)
AI agents
Foundational models

Each category includes links to specific GitHub projects with brief descriptions. For example, the Classification section lists projects like WaterNet (identifies water in satellite images) and Road-Network-Classification (classifies road networks using ResNet-34).

Why it exists

The README explains that deep learning has changed how satellite and aerial imagery is analyzed, but these images bring unique challenges — they're very large and contain many different object types. This repo is meant to give an exhaustive overview of techniques suited to those challenges.

How to use it

The README suggests using Command + F (Mac) or CTRL + F (Windows) to search the page directly — for example, typing 'SAM' to jump to relevant entries.

Companion website

The repository points to a companion site: satellite-image-deep-learning.com.

Notable details

The README includes a note that many articles labeled 'hyperspectral land classification' are actually describing semantic segmentation.
It links to introductory articles on satellite image classification and segmentation with neural networks.
The project is actively maintained and includes very recent entries (e.g., foundation models, AI agents, large vision & language models).