Scholarlink

Detecting gravitational waves (GWs) in measured data is a challenging task that demands advanced techniques for effective analysis due to the presence of intensive noise and the non-stationary nature of these signals. Quadratic time-frequency distributions (TFDs) from Cohen’s class provide valuable tools for analyzing various non-stationary signals simultaneously in the time and frequency domain. This chapter reviews a method that integrates deep convolutional neural networks (CNNs) with these quadratic TFDs to enhance the detection of GWs from binary black hole (BBH) mergers. The approach was validated on a comprehensive dataset of 100.000 signals combining actual Laser Interferometer Gravitational-Wave Observatory (LIGO) data with synthetically simulated GW injections. Twelve different two-dimensional (2D) TFD representations were calculated (resulting in 1.2 million TFDs) and used as inputs to three high-performance CNN models: ResNet-101, Xception, and EfficientNet. The proposed approach demonstrated superior detection performance, achieving high values across various classification metrics. Furthermore, it outperformed a CNN model using original time-series data, proving to be a viable solution for detecting GWs in low signal-to-noise ratio (SNR) environments.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Detecting Gravitational Waves From Binary Black Hole Mergers Using Deep Convolutional Neural Networks and Quadratic Time-Frequency Distributions

  • 章节

  • 发表日期：

    2025年

查看全文
  • Nikola Lopac,
  • Jonatan Lerga,
  • Franko Hržić

摘要

Detecting gravitational waves (GWs) in measured data is a challenging task that demands advanced techniques for effective analysis due to the presence of intensive noise and the non-stationary nature of these signals. Quadratic time-frequency distributions (TFDs) from Cohen’s class provide valuable tools for analyzing various non-stationary signals simultaneously in the time and frequency domain. This chapter reviews a method that integrates deep convolutional neural networks (CNNs) with these quadratic TFDs to enhance the detection of GWs from binary black hole (BBH) mergers. The approach was validated on a comprehensive dataset of 100.000 signals combining actual Laser Interferometer Gravitational-Wave Observatory (LIGO) data with synthetically simulated GW injections. Twelve different two-dimensional (2D) TFD representations were calculated (resulting in 1.2 million TFDs) and used as inputs to three high-performance CNN models: ResNet-101, Xception, and EfficientNet. The proposed approach demonstrated superior detection performance, achieving high values across various classification metrics. Furthermore, it outperformed a CNN model using original time-series data, proving to be a viable solution for detecting GWs in low signal-to-noise ratio (SNR) environments.