Abstract:
The extraction and classification of glitches in gravitational wave detector data have traditionally relied on complex machine learning architectures and large-scale datasets, often resulting in high computational costs. In this talk, we propose a novel and efficient method that integrates Q-Transform with Score-CAM (Score-weighted Class Activation Mapping) to enable accurate glitch localization and extraction with significantly reduced data and computational demands. By focusing on localized features in the time-frequency domain, our approach effectively isolates glitches while maintaining high precision. This streamlined framework not only improves the interpretability and accuracy of glitch detection but also offers a promising path toward real-time analysis in gravitational wave astronomy.
