Distributed Fiber Optic Vibration Measurement Image Identification Method Based On YOLOv3

To Distributed fiber optic sensing system is widely used in oil and gas pipeline operation status detection,perimeter security and pipeline cleaning robot positioning industries.To address the shortcomings of current fiber optic sensing signal analysis such as low recognition rate,this thesis takes φ-OTDR(Phase-sensitive optical time domain reflectometry,φ-OTDR)principle of a typical application of distributed fiber optic vibration sensing s+ystem DVS(Distributed Vibration In the context of distributed vibration sensing(DVS),we propose an analysis method for fiber optic sensing signal identification that is different from the traditional numerical analysis methods,in order to promote the technological development in the field of fiber optic sensing,by combining the machine learning method target detection network YOLOv3 for fiber optic sensing signal analysis.Firstly,the distributed fiber optic sensing φ-OTDR scattering model and performance indexes,the structure of DVS system with typical applications of distributed sensing theory and the technology in the development of target detection network are briefly discussed.In this thesis,the YOLOv3 model is modified according to the characteristics of the current research task to improve its recognition capability of fiber optic sensing signals,using data augmentation methods to expand the number of data sets,replacing higher performance activation functions to improve the generalization capability of the model,adding two attention mechanisms to improve the extraction capability,and adding two attention mechanisms for IOU expressions are added with an aspect ratio term to improve the performance of the model at the target border.Finally,a comparison experiment is conducted to verify the performance of the modified YOLOv3 model.The results show that the modified YOLOv3 model has better recognition ability than the origin YOLOv3 model according to the Loss curve,IOU curve,and PR curve.In comparison with the DVS system,the performance of the modified YOLOv3 model is about 5% higher than that of the DVS system for mechanical and manual excavation and shows the efficiency of research work.