DronePrint: Acoustic Signatures for Open-set Drone Detection and Identification with Online Data
summary
DronePrint uses microphone audio from online videos, extracts MFCCs with peak normalization and data augmentation via frequency warping, and classifies audio with a 10-step two-layer LSTM to detect and ID drones. The implemented system achieves 96% closed-set detection, 86% open-set detection, and 90% cascaded make→model identification across 13 commercial models. Tests show robustness to Doppler, hardware modification and varying SNR, and evaluation uses three datasets (known drones, non‑drone backgrounds, unknown sounds).
AI-written summary · about
chapters (10)
- 0:00Introduction
- 0:19Why DronePrint
- 0:57System Architecture
- 2:10Peak Normalization
- 2:54Sound Types
- 3:07Classification
- 3:25Data Sets
- 4:06Detection
- 4:47Identification
- 5:23Robustness
related talks
45mThe Sound of Your Secrets: Teaching Your Model to Spy So You Can Learn to Defend
46mDrones, RemoteID and the Thrilling Future of UAS Flight
26mDeepfakes, audiofakes, and the future of media
27mTruth or Fiction The Art of Unveiling DeepFakes in Text, Audio, Image
51mSDRs for Real, or How to Catch a Drone?
26mDrone Penetration Testing and Vulnerability Analysis Framework
28mAlgorithmic Bias: Data as a Weapon with Ayodele Odubela
24mCommanderSong: A Systematic Approach for Practical Adversarial Voice Recognition