What question did this study set out to answer?

The research aims to assess the effectiveness of a geometric feature extraction method for classifying terrain using polarimetric SAR data.

March 26, 2026Open Access

Geometric Features for Polarimetric SAR Terrain Classification: Benchmark Results on AIRSAR San Francisco

Key Points

The research aims to assess the effectiveness of a geometric feature extraction method for classifying terrain using polarimetric SAR data.
Applied a proprietary geometric feature extraction method to polarimetric SAR data.
Derived feature vectors from the Riemannian structure of coherency matrices.
Conducted classification on AIRSAR San Francisco dataset with a 5-class system using ResNet-SE.
Achieved 96.73% overall accuracy with the proposed method.
Competitive performance compared to state-of-the-art methods, with SDF2Net showing 97.13% accuracy.
Demonstrated a 9.3 percentage point advantage in a few-shot regime with limited training samples.

Abstract

OmniSciences Technical Report TR-2026-001 We present benchmark results for a proprietary geometric feature extraction method applied to polarimetric SAR (PolSAR) terrain classification. The method derives compact feature vectors from the Riemannian structure of Hermitian positive-definite coherency matrices on the symmetric space GL(3,ℂ)/U(3). Key results on AIRSAR San Francisco (5-class, L-band): 96.73% overall accuracy with ResNet-SE (15×15 patches, 1% training) — competitive with state-of-the-art complex-valued CNNs (SDF2Net: 97.13%) +9.3 percentage point advantage in few-shot regime (n=5 samples/class) Geometric features subsume raw coherency matrix input Feature extraction methods are patent-pending. This report presents results only; methods will be disclosed in a forthcoming publication.

Geometric Features for Polarimetric SAR Terrain Classification: Benchmark Results on AIRSAR San Francisco

Key Points

Abstract

Cite This Study