Pretraining Initialization and Time-Frequency Representation Design for Closed-Set Passive RF Drone Classification under Wireless Interference: Revision 1 Reproducibility Package

Résumé fourni par la source

Audited code, final Colab notebooks, immutable protocol manifests, aggregate outer-test result files, compact prediction analyses, clustered statistical outputs, and manuscript artifact builders for Revision 1. The archive covers capture-group-disjoint nested evaluation on DroneDetect V2, exact-AST AudioSet/ImageNet/scratch initialization controls, matched RF front ends, corrected YAMNet controls, condition-aware held-out evaluation, sensitivity analyses, recording-level inference, representation diagnostics, and NVIDIA L4 efficiency measurements. Raw IQ recordings, derived HDF5 caches, checkpoints, and resume files are excluded; the public data and cache-generation code are identified in the archive.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.