Denoising-driven information extraction network (DENet) for Brillouin optical time-domain analyzer

Effective denoising techniques can significantly enhance the signal-to-noise ratio (SNR) in Brillouin gain spectrum (BGS) for Brillouin optical time-domain analyzer (BOTDA), thereby enhancing both measurement accuracy and sensing distance. These techniques also reduce the required number of averages times and hence increase measurement speed. However, most existing BGS denoising approaches treat denoising as a preprocessing step for Brillouin frequency shift (BFS) extraction. After denoising, sensing information such as temperature and strain are then typically extracted using methods like Lorentz curve fitting (LCF), resulting in a cumbersome and time-intensive process that is often unsuitable for real-time detection, not to mention the spatial resolution deterioration brought in. To address this issue, a novel denoising-driven information extraction network (DENet) based on a convolutional denoising autoencoder is proposed. DENet integrates denoising and information extraction into a unified real-time processing framework, thereby improving the measurement accuracy and speed of BOTDA, especially under a low SNR condition. Experimental results demonstrate that DENet could secure a 14.1 dB SNR improvement with accurate temperature extraction in only 1.2 s on a BGS with dimensions of 200 x 24000. Compared to traditional LCF methods, the root-mean-square error of the extracted temperature by DENet decreases from 1.38 degrees C to 0.64 degrees C, whereas the extraction time was reduced by approximately 168 times. Additionally, the proposed method also well preserves the spatial resolution, offering an efficient and accurate option for real-time BOTDA detection.