Speaker
Description
The discovery of gravitational waves as a major breakthrough in 21st-century physics has provided humanity with a novel observational tool to explore the mysteries of the Universe. However, in actual detection, gravitational wave detectors are often affected by various types of noise interference, particularly transient and non-Gaussian ‘glitch’ noise. These glitches not only obscure genuine gravitational wave signals but also generate false triggers owing to their striking similarity to real signals, significantly compromising detection accuracy. Consequently, developing efficient and accurate glitch classification methods to improve the gravitational wave data quality has become a key issue that urgently needs to be addressed in this field. In this study, we present a novel classification approach that involves: (1) constructing a fusion-view dataset to fully extract valuable information from the data and (2) optimizing a convolutional neural network (CNN) model to improve the glitch classification performance in gravitational wave time-frequency representations.
To further enhance the model’s capability with minority-class samples, we incorporate a long shortterm memory (LSTM) network, creating a hybrid CNN-LSTM architecture. Our experimental results show that this hybrid model achieves higher classification accuracy while demonstrating
superior feature-learning capabilities, especially for categories with limited samples. Additionally, we address the issues of class imbalance and low data quality through a data augmentation technique, effectively reducing interclass confusion and improving dataset quality.
| Research Area | Gravitational Waves |
|---|