Sub-Pixel Target Anomaly Detection
Anomalies below the size of a pixel: artificial coatings, camouflage materials and low fill-fraction objects.
Anomaly detection against a locally estimated background, for man-made materials smaller than a pixel.
Sub-pixel anomaly response over mixed ground, where a target occupies part of a pixel rather than all of it.
WHAT IT FINDS
What it detects.
| Material | Typical setting | What can confuse it |
|---|---|---|
| Man-Made Materials & Camouflage Fabrics | Tactical terrain, infrastructure security, and border surveillance zones | Dense natural canopy shadowing and bright specular reflections |
METHOD
How it works.
Calibrated reflectance goes in, and every pixel is fitted against a public reference library.
Background covariance estimation
Local and scene-wide background covariance are computed across the contiguous channels. How well this works depends on how many distinct materials the background already contains.
Reed-Xiaoli anomaly detection
Measures how far each pixel sits from its surrounding background in spectral distance, with the false alarm rate held constant rather than the threshold.
Constrained energy minimisation
A filter that passes a chosen target signature at unit gain while suppressing everything else. It needs a signature to look for, which is the difference between this and plain anomaly detection.
Abundance and coordinates
Surviving pixels carry an estimated abundance and a location, and each export records the confidence attached to it.
SPECIFICATIONS
What you get, and what it runs on.
Conditions it needs
Method and checks are documented in Axine Applied Spectroscopy Technical Whitepaper · Series 6.
Scope
Exploratory. Detection degrades as the background gets more varied and as the target fills less of a pixel.
RUN IT OVER YOUR GROUND
Run this layer over your own area.
Order processing over your area of interest, or a tasked capture where the archive has nothing usable. Imagery access is never charged for: a plan buys the models.
COMPLEMENTARY ANALYTICS
Related layers.
Infrastructure Thermal Inertia
Day and night thermal differencing, normalised by albedo. Dense material gives up heat more slowly than loose fill.
Alteration Cluster Detection
Vectorised clustering that pulls out the core hydrothermal hubs and the nodes worth working outward from.
Mine Footprint & Disturbance
Multi-temporal surface disturbance, tailings boundaries and waste rock footprints, measured between dates.