Improvement of response speed and precision of distributed Brillouin optical fiber sensors using neural networks

•A high-precision real-time distributed optical fiber temperature and strain sensing system is achieved and demonstrated.•A nonlinear interpolation deep neural network (NIDNN) is proposed to interpolate and denoise for BGS.•A convolutional neural network based on depthwise separable convolution (BFSECNN) is proposed to shorten the BFS extraction.•The sweep time after using NIDNN is 1/6 of the hardware sweep time of 1-MHz interval for the 28.5-km long distributed Brillouin optical fiber sensor.•The BFSECNN extraction time is 3.7 s for 71,168 BGSs, 77 times faster than the Lorentzian curve fitting.