Identification of Ultra High Frequency Acoustic Coda Waves Using Deep Neural Networks

• Published in: IEEE sensors journal 2021-09, Vol.21 (18), p.20640-20647
• Main Authors: Thati, Venu Babu, Smagin, Nikolay, Dahmani, Hatem, Carlier, Julien, Alouani, Ihsen
• Format: Article
• Language: English
• Subjects: Acoustic imaging; Acoustic noise; Acoustic propagation; Acoustics; Algorithms; Artificial neural networks; Coda waves; data processing and augmentation; Deep learning; deep neural network; Emitters; Engineering Sciences; Feature extraction; Imaging; Inspection; Machine learning; Neural networks; Nondestructive testing; Physics; Receivers; Sensitivity; signal processing; Silicon; source identification; Transducers; Ultrahigh frequencies; Wave propagation
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2021.3099078

Due to the multi-path propagation and extreme sensitivity to minor changes in the propagation medium, the coda waves open new fascinating possibilities in the non-destructive evaluation and acoustic imaging. However, their noise-like structure and high spurious sensitivity for ambient conditions (temperature, humidity, and others) make it challenging to perform localized inspection in the overall coda wave evolution. While existing deterministic solutions reach their limit in handling complex data, emerging techniques such as deep learning-based algorithms have shown a promising potential to overcome these limitations. This paper proposes a deep neural network that paves the way to make the complex features of coda waves more handleable to reliably exploit coda waves in several applications even in a changing or unstable environment. Specifically, designed a Coda-Convolutional Neural Network that is able to identify coda waves with 95.65% precision in a silicon chaotic cavity including 5 emitters and 16 receivers using ultra high frequency ultrasonic coda waves.