Research On Parallel Processing Of Optical Fiber Vibration Signal Based On GPU

In recent years,distributed optical fiber sensing technology has become a hot research field at home and abroad due to its features such as low unit distance cost,strong environmental adaptability,safety without power supply,easy to extent,to lay and to preserve.The thesis is based on distributed vibration fiber intrusion detection system,a kind of positioning fiber perimeter security system based on distributed fiber sensor technology,which has matured internationally,while most of the domestic fiber perimeter security systems are still in the area type.The difficulty of distributed vibration fiber research mainly lies in the analysis and recognition of vibration events,reduction of false alarm rate,guarantee of practicability and real-time processing of massive data.Distributed optical fiber sensor technology takes the whole optical fiber as both sensing unit and signal transmission medium,and can measure the strain information at any position of the optical fiber.The sensing distance can be up to tens or even hundreds of kilometers,equivalent to thousands of sensors deployed.The data volume is huge and the real-time processing requirements are high.The traditional implementation method occupies large system resources and takes a long time of operation,which not only affects the real-time performance of the system,but also causes great system burden.In view of the above situation,this thesis proposes a parallel processing algorithm based on GPU for optical fiber vibration data to reduce false alarm rate of distributed vibration optical fiber intrusion detection system and solve intrusion interference problem.At the same time,GPU parallel operation can share the operation pressure of CPU side,so as to improve the real-time performance of distributed vibration fiber intrusion detection system and obtain better practical application.In this thesis,while designing the GPU-side parallel computing algorithm,the actual application effect of the algorithm is analyzed and tested in detail.Under the premise of ensuring practicability,this thesis uses the NSIGHT tools provided by NVIDIA to analyze the execution process and performance.The results show that the recognition accuracy of GPU-based parallel computing algorithm can reach about 95%,the performance of the key algorithm can be improved by 40 times than before,which greatly improves the real-time performance of the system.