spectraforce 0.1.0

Premium desktop workstation for Earth observation segmentation.

SpectraForge

A premium desktop workstation for Earth observation segmentation with GMM + uncertainty, multi-sensor feature selection, and interactive ROI labeling. Built to run locally on Windows/macOS/Linux.

Why it exists

• Multi-sensor preprocessing (Sentinel‑1/2/3, Landsat, ERA5, ENMAP, hyperspectral, PlanetScope, UAV)
• GMM segmentation with predict_proba and uncertainty maps
• Interactive ROI labeling to refine clusters or assign human labels
• Designed for patent-ready workflows (keep private until filing)

Features (v1)

• Clean, modern studio UI (not a QGIS clone)
• GeoTIFF, NetCDF, ENVI (hyperspectral), Sentinel‑2 folder support
• Sensor presets (Sentinel‑1/2/3, Landsat, PlanetScope, UAV, ERA5, ENMAP, hyperspectral)
• Auto feature selection across multi-band data
• GMM segmentation with probability maps and entropy-based uncertainty
• Index Builder with curated indices per sensor + custom JSON/YAML recipes
• ROI selection + cluster labeling workflow
• Run history saved to runs/
• Copernicus + Planet API panels (offline stub in this build)

Install

From PyPI:

python -m pip install spectraforce

From source:

python -m pip install -r requirements.txt

Run

spectraforge

Alternative:

python -m spectraforge

Data formats

• GeoTIFF: .tif, .tiff
• NetCDF: .nc (ERA5 and other gridded products)
• ENVI: .hdr + .img (hyperspectral)
• Sentinel‑2: SAFE folder with .jp2 band files
• NumPy: .npy arrays (2D or 3D)
• Tabular: .csv, .xlsx

Indices

SpectraForge ships with curated indices (NDVI, NDWI, etc.) per sensor. You can add your own:

• JSON/YAML recipe files (see samples/ for examples)
• Index layers appear in the layer stack like any other band Example custom indices file: samples/indices_example.json

NPY export

Export data to .npy directly from the UI:

• All bands in one file
• Selected bands in one file
• Individual band files
• Optional index layer exports

Offline samples

Synthetic samples live in samples/ so the repo runs without downloads. Real cropped samples live in samples/real/. Manipur example outputs live in samples/manipur/.

API keys (stored locally)

API keys are stored in ~/.spectraforge/config.json on your machine.

Why I Use SpectraForge

I use SpectraForge when I want a fast, repeatable path from raw satellite data to clean indices, segmentation, and uncertainty maps. I need it because I do not want to rebuild custom scripts for every dataset. It saves me time because I can run the same workflow across sensors and get consistent results. I use it easily by loading a folder, letting the app auto detect bands, choosing indices, running GMM, and labeling ROIs in one place.

Step by Step Outputs for Manipur Sentinel‑2

I ran the full pipeline on real Sentinel‑2 data from Manipur and saved the outputs below.

Step 1 — Load bands and true color

Step 2 — NDVI and NDWI indices

NDVI

NDWI

Step 3 — GMM segmentation with 8 clusters

Step 4 — ROI selection and labeling

ROI overlay

ROI labels

Step 5 — Uncertainty and confidence I use the same color rule for both maps: blue means low, red means high.
For uncertainty, blue means low uncertainty and red means high uncertainty.
For confidence, blue means low confidence and red means high confidence.

Uncertainty

Confidence

Saved outputs full resolution

• samples/manipur_full/manipur_full_stack.npy
• samples/manipur_full/manipur_full_stack_preview.tif
• samples/manipur_full/manipur_full_<index>.npy
• samples/manipur_full/manipur_full_<index>.tif

Color Legend for GMM labels

Clusters are unsupervised. Colors map to cluster IDs in order:

• 0 → maroon #800000
• 1 → darkblue #00008B
• 2 → darkgreen #006400
• 3 → cyan #00FFFF
• 4 → darkcyan #008B8B
• 5 → magenta #FF00FF
• 6 → indigo #4B0082
• 7 → grey #808080
• 8 → peru #CD853F
• 9 → slateblue #6A5ACD
• 10 → mediumspringgreen #00FA9A
• 11 → orangered #FF4500

If you run fewer than 12 clusters, only the first N colors are used.

Color Notes for indices

Index quicklooks use a viridis scale: brighter colors indicate higher values.

Uncertainty calibration made easy

• Run GMM → get predict_proba
• Use entropy + confidence to visualize uncertain regions
• Assign labels with ROI selections (no labeled data required)

GMM engine

If your environment has scientific stack conflicts, switch the engine to Safe mode in the UI.
Fast mode uses scikit‑learn when available.

SpectraForge vs QGIS (Pros & Cons)

Aspect	SpectraForge	QGIS
Focus	EO segmentation + indices + uncertainty	Full GIS for all domains
Setup	One command local run	Heavier install + plugins
GMM + uncertainty	Built‑in, turnkey	Requires plugins/workflows
Indices	Curated EO indices + custom recipes	Many tools, but more manual setup
UI style	Modern studio layout (not QGIS style)	Traditional GIS layout
Extensibility	Focused feature set	Huge plugin ecosystem
Geoprocessing breadth	Focused EO analytics	Broad GIS toolbox
Best for	Fast EO segmentation + research demos	Full GIS analysis & cartography

Pros of SpectraForge: fast EO‑first workflow, built‑in uncertainty, simple NPY export, easy to demo.
Cons vs QGIS: fewer GIS tools, smaller plugin ecosystem, less advanced cartography.

Contributions

See CONTRIBUTING.md and CODE_OF_CONDUCT.md.

Privacy note

Runs locally. No data leaves your machine.

License

Proprietary (permission required for any use)