Human-in-the-Loop Signature Bootstrapping for UAV Hyperspectral PFM-1 Mine Detection

This paper evaluates PFM-1 landmine detection from UAV visible and near-infrared hyperspectral imagery using spectral angle mapper (SAM), matched filter (MF), adaptive coherence estimator (ACE), and constrained energy minimization (CEM). It compares a ground-measured SVC signature, a fully informed in-scene core-pixel signature, and a simulated human-in-the-loop bootstrap. Beyond ROC-AUC and average precision, the study measures target-discovery curves and candidate-review burden. Full-review bootstrapping reaches the fully informed in-scene case after verifying seven target regions. ACE confirms all regions in two rounds and nine candidate inspections, while SAM variants require thousands of reviews for their final targets.