Audio vs. Vision Pretraining for RF-Based Drone Fingerprinting under Wireless Interference: Reproducibility Package

Résumé fourni par la source

Training scripts, Jupyter notebooks, fold definitions, configuration files, and compact result JSON/CSV artifacts for the paper "Audio vs. Vision Pretraining for RF-Based Drone Fingerprinting under Wireless Interference". The package supports verification and reproduction of the DroneDetect V2 experiments, including the representation/backbone ablation, AudioSet-pretrained YAMNet baselines, ImageNet-pretrained MobileNet-V2 architecture-control baseline, condition-aware robustness experiments, and three-fold robustness variance analysis. Raw DroneDetect V2 IQ recordings and large derived spectrogram caches are not included; the provided scripts and instructions regenerate the cache from the public dataset.