An Acoustic Modeling System For Physics Guided Machine Learning
Abstract
This study aims to establish an acoustic modeling cyber infrastructure (CI) that can use physics of sound to simulate the sound wave propagation in real-life environments and provide tools to generate realistic datasets for training the deep learning algorithms. Unmanned air systems (UAS) are being increasingly manufactured and used in different applications. Their high technical capabilities allow adversaries to exploit these technologies and conduct different types of attacks against critical infrastructures. Acoustic sensors can realize a resilient low-cost and power sensing system that enables detection, localization, and classification of these UAS through capturing and processing their propeller noise in a variety of environments. However, the state-of-the-art sound source detection and classification techniques do not provide accurate prediction of sound wave propagation in complex environments. Deep learning-based methods have shown promises, but the costs of collecting large training datasets to cover a range of possible scenarios are prohibitive. A reliable solution to successfully detect a UAS flying in outdoor needs reproducing and testing a range of realistic scenarios. These include different atmospheric conditions (wind, temperature, turbulence, absorption), noise barriers, types of UAS and maneuvers, number of sound sources, ground effects, and irregular terrains. What if we had a highly capable CI that could realistically simulate the sound propagation under different conditions in outdoor environments and generate reliable data? Establishing such CI is a goal of this research. If successful, the CI can help researchers to generate realistic datasets that can be used for training and testing deep learning algorithms.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2024
- Accession Number
- AD1219669
Entities
People
- Patrick McNamee
- Zahra N. Ahmadabadi
Organizations
- San Diego State University