Unsupervised CP-UNet Framework for Denoising DAS Data with Decay Noise

• DOI: 10.48550/arxiv.2502.13395
• Publication Date: 2025-02-18
• Authors: Tianye Huang; Aopeng Li; Xiang Li; Jing Zhang; S. Xian; Xin Ma; Meng Kai Lü; Guodong Chen; Liangming Xiong; Xiangyun Hu
• Keywords: Signal Processing (eess.SP); FOS: Computer and information sciences; Computer Science - Machine Learning; Sound (cs.SD); Audio and Speech Processing (eess.AS); FOS: Electrical engineering, electronic engineering, information engineering; FOS: Physical sciences; Electrical Engineering and Systems Science - Signal Processing; Computer Science - Sound; Electrical Engineering and Systems Science - Audio and Speech Processing; Physics - Optics; Machine Learning (cs.LG); Optics (physics.optics)

Distributed acoustic sensor (DAS) technology leverages optical fiber cables to detect signals, providing cost-effective and dense monitoring capabilities. It offers several advantages including resistance extreme conditions, immunity electromagnetic interference, accurate detection. However, DAS typically exhibits a lower signal-to-noise ratio (S/N) compared geophones is susceptible various noise types, such as random noise, erratic level long-period noise. This reduced S/N can negatively impact data analyses containing inversion interpretation. While artificial intelligence has demonstrated excellent denoising capabilities, most existing methods rely on supervised learning with labeled data, which imposes stringent requirements the quality of labels. To address this issue, we develop label-free unsupervised (UL) network model based Context-Pyramid-UNet (CP-UNet) suppress noises in data. The CP-UNet utilizes Context Pyramid Module encoding decoding process extract features reconstruct enhance connectivity between shallow deep features, add Connected (CM) both section. Layer Normalization (LN) utilized replace commonly employed Batch (BN), accelerating convergence preventing gradient explosion during training. Huber-loss adopted our loss function whose parameters are experimentally determined. We apply 2-D synthetic filed Comparing traditional latest UL framework, proposed method demonstrates superior reduction performance.