WTTnet: a network combining wavelet transform and transformer for denoising microseismic signal

Noise suppression is a key component in microseismic monitoring technology. Accurate denoising of microseismic signals is crucial for ensuring reliable data for locating mining-related seismic events and analyzing the state of rock mass during mining operations. This paper proposes a network combining wavelet transform and Transformer for denoising microseismic signals (WTTnet). WTTnet leverages discrete wavelet transform to separate the high- and low-frequency components of the input signal. These components are concatenated to form full-frequency features, which are then used as query and value vectors in the Transformer, while the high-frequency features serve as keys. The multi-head self-attention mechanism captures cross-scale correlations. Finally, inverse discrete wavelet transform converts the frequency-domain output back to the time domain. The primary strength of this model is its ability to identify and distinguish noise components across varying frequencies. The proposed method is tested on synthetic data contaminated with various noise types and on field data. Its denoising performance is evaluated using appropriate metrics and compared with other denoising methods. Experimental results show that this method outperforms traditional denoising methods in terms of overall denoising performance across diverse noise conditions.