研究目的
To classify transient signals (glitches) in gravitational wave detectors using a wavelet-based method combined with XGBoost for improved detector characterization and noise reduction.
研究成果
The WDF and XGBoost pipeline effectively classifies transient signals with high accuracy (over 95% for binary classification and over 82% for multi-label classification), demonstrating its potential for real-time application in gravitational wave detectors. However, optimizations are required to handle window edge effects and improve overall robustness.
研究不足
The method has lower accuracy compared to deep learning approaches. Issues include signals falling between analysis windows due to small overlap, leading to reduced classification accuracy for shorter signals. Further tuning of parameters is needed to improve performance and avoid edge effects.
1：Experimental Design and Method Selection:
The study uses a pipeline called WDFX, which combines the Wavelet Detection Filter (WDF) for feature extraction and denoising with the XGBoost algorithm for supervised classification. The WDF employs wavelet decomposition with various wavelet types (e.g., Daubechies, Haar, spline wavelets) and a thresholding method for denoising. XGBoost is used for binary and multi-label classification tasks.
2：Sample Selection and Data Sources:
Simulated data sets of glitches were generated, including six families of signals (Gaussian glitches, sine-gaussian, ringdown, whistle-like, scattered light-like, and chirp-like transients). These were added to Gaussian noise colored by the sensitivity curve of the LIGO Hanford detector (H1). Data was sampled at 8192 Hz.
3：1). Data was sampled at 8192 Hz.
List of Experimental Equipment and Materials:
3. List of Experimental Equipment and Materials: A virtual machine with 16 VCPU Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.4GHz and 32GB RAM was used for processing. Software tools include implementations of WDF and XGBoost.
4：4GHz and 32GB RAM was used for processing. Software tools include implementations of WDF and XGBoost.
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: Data conditioning involved whitening using an Auto Regressive (AR) model with 4000 parameters. WDF was applied with a window size of 1024 points (0.125 seconds) and small overlapping. Detected triggers were processed by XGBoost for classification, with training, validation, and test sets split in a 70/15/15 ratio.
5：125 seconds) and small overlapping. Detected triggers were processed by XGBoost for classification, with training, validation, and test sets split in a 70/15/15 ratio.
Data Analysis Methods:
5. Data Analysis Methods: Performance was evaluated based on accuracy, confusion matrices, and loss functions. Hyper-parameter tuning for XGBoost was done using grid search cross-validation.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
