Sparse UWB A-scan Landmine Detection: Code, Signal Caches, and Reproducibility Data

Reproducibility materials for “Landmine Detection and Type Recognition from Sparse UWB Measurements: A Controlled 3D FDTD Study” by Mike Schaeffer. The release supports evaluation of a learned separation–classification cascade and direct classifiers from five target-centered UWB A-scans. The study uses controlled three-dimensional electromagnetic simulations with heterogeneous soil and metallic clutter. Validation on a physical UWB/GPR instrument is planned. The archive contains research code, configurations, signal caches for 1638 development and 345 final environments, ten trained checkpoints, per-case predictions, training histories, numerical figure/table data and instructions for metric recalculation and repeated inference. It also includes the saved geometries, physical metadata and input decks for all 1983 environments, source models, simulation-generation code and the supplied solver source/patches. The caches contain the signals used by the classifiers; the archive does not duplicate the complete raw solver-output corpus. Earlier Phase-1 experiments are represented by the available historical summaries and supporting source extracts; their raw corpora and checkpoints are not included. The final evaluation pool was reused in earlier analyses, as documented in the manuscript and package. Author-owned research code is MIT; author-owned numerical data and weights are CC BY 4.0. Third-party components retain their original licenses, including CC BY-SA 4.0 and GPL-3.0-or-later where applicable. Consult the file-level license inventory and notices for exact scope. Related patent disclosure: Ukrainian application a202605448, filed on 14 September 2026. Funding disclosure: in-kind Microsoft Azure startup cloud-credit support. Version 1.0.0. See the README, manifest and validation reports for contents, commands, provenance and numerical tolerances.