Research On Deep Learning Image Detection For Distributed Acoustic Sensing

Distributed Acoustic Sensing(DAS)is an emerging technology in recent years that utilizes telecommunications fiber optic cables as seismic sensor arrays.It enables the sampling of vibration signals with meter-level spatial resolution along fiber cables that can span tens of kilometers,resulting in tens of thousands of observations from a single device.DAS offers advantages such as dense and continuous monitoring,low maintenance costs,and strong environmental adaptability.These advantages make DAS a promising technology,providing new solutions and opportunities for infrastructure monitoring and earthquake research.Dense observation with Distributed Acoustic Sensing generates a massive amount of data.Efficient and fast processing of DAS big data is crucial for the widespread application of DAS in infrastructure and earthquake monitoring.In this study,we significantly reduced the data volume by converting DAS data into images.We then applied the YOLO(You Only Look Once)deep learning algorithm for image object detection to quickly identify vibration signals in DAS images,thereby improving data processing efficiency.Based on this approach,we conducted research on traffic monitoring in Changping District,Beijing and detection and localization aftershock activities in the epicentral region of Ms 6.8Luding earthquake in Sichuan on September 5,2022.The achievements of this study are as follows:1.We used a real-time deep learning-based object detection framework,the YOLO algorithm,to estimate traffic flow and vehicle speed on a 500-meter optical fiber cable in the suburban area of Beijing.We compared the results with the slant stacking method and peak event detection method using 15 days data.The findings showed that all three methods exhibited similar patterns in terms of peak hours,daily traffic flow variations,and hourly variations.However,the YOLO algorithm detected higher traffic flow compared to the slant stacking method,and its results were closer to the actual traffic flow.Both the YOLO algorithm and the slant stacking algorithm provided similar daily speed variation characteristics,which corresponded to the peak hours of traffic flow,showing a decreasing trend from morning to evening peak hours.Additionally,we discovered that the vibration amplitude generated by vehicles is positively correlated not only with the weight of the vehicles but also with their speed.Therefore,when using vibration monitoring to assess the weight of vehicles,the influence of vehicle speed needs to be considered.2.We used 45 days of DAS monitoring data from a communication optical fiber cable line approximately 6.4 km long in the epicentral region of the September 5,2022 Ms6.8 Luding earthquake in Sichuan.We employed the YOLO image detection method to detect seismic events and used the image template matching method for event localization.Training data and template images were created based on the Luding seismic network catalog.After training the YOLO model,we rescanned the continuous waveforms and detected 4443 seismic events,which was approximately twice the number of events in the concurrent seismic network catalog.However,only 543 events could be matched with the template events,which was slightly higher than the number in the seismic network catalog.The number of small magnitude events that could be matched with the template was relatively low.Therefore,the large number of unmatched events may be due to factors such as a limited number of templates and the low signal-to-noise ratio in DAS recordings.In summary,this study explored the use of artificial intelligence image recognition techniques to analyze massive DAS data in the contexts of traffic monitoring and earthquake monitoring.The results demonstrated the effectiveness and feasibility of image processing techniques in handling DAS data,and their advantage in processing speed enables near-real-time applications.The applications presented in this study also indicate that using DAS for traffic monitoring can enhance existing traffic video surveillance networks and contribute to the development of smart cities.Furthermore,utilizing existing communication fibers and DAS can quickly improve the seismic monitoring capabilities in earthquake source areas,enhance earthquake detection and localization,and provide better support for earthquake monitoring and early warning systems.