SatVector

Satellite imagery patch similarity search using spectral indices, texture features, and HNSW indexing.

Note: Forked from terravector and adapted for Earth observation data.

Overview

SatVector converts satellite imagery into searchable vector embeddings. Each image patch gets a "fingerprint" based on spectral indices, texture analysis, and signal decomposition. An HNSW index enables O(log N) similarity queries across millions of patches.

Use cases:

• Find all patches similar to a known land cover type
• Discover anomalies in satellite imagery
• Transfer classification signatures across different regions
• Rapid land cover classification without manual labeling
• Change detection by comparing temporal signatures

Installation

git clone https://github.com/bshepp/satvector.git
cd satvector
pip install -r requirements.txt

Quick Start

Desktop Viewer (Recommended)

Launch the napari-based desktop viewer:

python viewer.py path/to/image.tif path/to/index.idx

Web UI

Launch the Gradio web interface:

python app.py

Command Line

# Build index from satellite image
python cli.py build image.tif --patch-size 64 --output satellite.idx

# Build with different signature types
python cli.py build image.tif --config vegetation --output veg.idx
python cli.py build image.tif --config urban --output urban.idx

# Query for similar patches
python cli.py query satellite.idx --patch 42 --k 10

# Query with visualization
python cli.py query satellite.idx --coords 500,200 --k 8 --visualize results.png

Signature Types

Spectral Indices

Vegetation and water indices from multispectral bands:

• NDVI: Normalized Difference Vegetation Index
• NDWI: Normalized Difference Water Index
• EVI: Enhanced Vegetation Index
• SAVI: Soil Adjusted Vegetation Index
• NDBI: Normalized Difference Built-up Index
• Band Ratios: Custom band ratio features

Texture Features

Image texture analysis:

• GLCM: Gray-Level Co-occurrence Matrix
• LBP: Local Binary Patterns

Decomposition

Signal processing from terravector:

• Gaussian, Bilateral, Wavelet, Morphological filters

Presets

Preset	Features	Use Case
default	Spectral + Texture	General purpose
vegetation	NDVI, EVI, SAVI + Texture	Vegetation mapping
urban	NDBI, texture, edges	Urban detection
water	NDWI, blue band stats	Water body detection
agriculture	NDVI temporal + texture	Crop classification

Supported Formats

• GeoTIFF (.tif, .tiff) - single or multi-band
• JPEG2000 (.jp2) - Sentinel-2 format
• NumPy (.npy) - preprocessed arrays

Requirements

• Python 3.8+
• numpy, scipy, scikit-image
• rasterio (geospatial I/O)
• nmslib (HNSW implementation)
• gradio, napari (UI)

License

Apache 2.0

Citation

@software{satvector2025,
  author = {Shepp, B.},
  title = {SatVector: Satellite Imagery Patch Similarity Search},
  year = {2025},
  url = {https://github.com/bshepp/satvector}
}