Performance Enhancement of BOTDA Based on the Image Super-Resolution Reconstruction

The acquisition time of Brillouin optical time domain analysis (BOTDA) sensor is relatively long since it requires multiple scans of probe tones to map the Brillouin gain spectra (BGS). To enhance the performance of BOTDA with fewer probe tones and lower time-domain sampling rate, we propose a method based on generative adversarial network (GAN) to reconstruct the super-resolution BGS from its low-resolution counterpart for 41 upscaling factors. We preliminarily prepare a dataset composed of ideal BGSs with different linewidths, spatial resolutions, and Brillouin frequency shifts (BFS) for network training and create a 2048-level colormap for the conversion between BGS and its RGB image. By cropping the measured BGS into consecutive frames and making cross-correlation to transform each frame to a new one with nearly ideal Lorentzian shape as input for the independent reconstruction, the 4x enlarged super-resolved BGS conforming to the real BGS distribution could be established. The validity of our method was shown through a conventional BOTDA experiment. The results show that 75% measurements were reduced and the resolution in spatial and frequency domain were raised by 4 times, and the measurement accuracy also got greatly improved.