Research On Fast Denoising And Feature Extraction Of Scattering Spectrum In Botda System

Brillouin optical time domain analysis(BOTDA),as the representative of distributed optical fiber sensing technology,has the advantages of long distance,high measurement accuracy,high spatial resolution,corrosion resistance and anti electromagnetic interference compared with traditional electrical and mechanical sensors,which can realize the real distributed measurement.Because BOTDA can realize the distributed measurement of temperature and strain at the same time,it is widely used in the structural health monitoring of large-scale infrastructure such as dams,bridges and tunnels,submarine optical cables and oil and gas transportation pipelines.In recent years,relevant scholars have been committed to improving the performance of BOTDA system,but there are still some problems:(1)the non local mean algorithm in digital image processing technology can solve the similarity and redundancy of the scattering spectrum of BOTDA system without improving the system hardware configuration,and improve the sensing performance of BOTDA system to a certain extent.Because of the high complexity of non local mean algorithm,although it can improve the signal-to-noise ratio of the system,the real-time performance is not good,which is not conducive to the online monitoring of the system;(2)using the generalized regression neural network to extract the Brillouin scattering spectrum features,its accuracy depends on the smooth factor,the artificial selection is more complicated and the accuracy is unstable.In view of the above problems,the specific research contents of this paper are as follows:Firstly,aiming at the problems of high complexity and real-time performance of NLM algorithm,discrete cosine transform(DCT)is introduced to improve NLM algorithm,and compared with DCT and NLM algorithm in Brillouin scattering spectrum characteristics,signal-to-noise ratio,measurement accuracy and measurement time.The experimental results show that the optimized algorithm can keep the basic characteristics of Brillouin gain spectrum while denoising,improve the operation efficiency of the algorithm,and is conducive to the realtime online monitoring of BOTDA system.Then,this paper studies the feature extraction method of scattering spectrum in BOTDA system.Aiming at the problem that the accuracy of GRNN algorithm depends on the smoothness factor,and the artificial selection is complicated and the accuracy is unstable,particle swarm optimization is used to optimize GRNN,and the smoothness factor is globally optimized.The optimal value is used to establish the model,and then the model is simulated with Lorentz The characteristics of Brillouin scattering spectrum extracted by combined algorithm,cross-correlation algorithm and particle swarm optimization generalized regression neural network are compared and analyzed.The experimental results show that the Brillouin scattering spectrum features extracted by particle swarm optimization generalized regression neural network model are more accurate,and the algorithm has the fastest extraction speed and good real-time performance.