Monitoring Method For Seismic Excitation And Reception Operation Quality Based On Deep Learning

The construction quality of excitation and reception operations in seismic acquisition is the key to obtaining high quality data.The construction personnel capture videos and photos of the entire construction process,and the quality inspectors monitor the construction quality based on the videos and photos taken,so that problems can be found and reworked in time to guarantee the accuracy and integrity of seismic data.As the seismic exploration develops further,the acquisition points are getting denser and the number of captured data is increasing exponentially.However,manual monitoring has some weakness,for instance,low efficiency,long time consumption,and subjectivity.To address these issues,an intelligent monitoring approach for the quality of seismic excitation and reception field operations is proposed,utilizing object detection and object tracking algorithms to replace traditional manual monitoring.This method meets the monitoring needs of field operations and can shorten exploration cycles,reduce acquisition costs,timely detect construction problems,and ensure the quality of seismic data.For the monitoring needs of construction personnel operation standards and seismic acquisition device installation quality,the YOLOv8 algorithm is used to identify the object in the captured data,including work clothes,safety helmets,warning lines,explosive boxes,seismic acquisition device installations that are qualified and unqualified.To address the challenges of object detection in seismic data,an modified YOLOv8 algorithm is put forward.Introducing CBAM convolutional attention modules into YOLOv8 to strengthen the attention of the front-end detection network,which weakens the background information and extracts the deep features of specific regions,and uses WIo U to replace the loss function of the classification part.This enables the detection network to focus on detecting normal quality detection boxes,thereby adapting to the complex seismic exploration environment,enhancing the target detection capabilities in mountainous and hilly backgrounds,and reducing missed,multiple detection boxes and misidentification issues in object detection.In response to the monitoring needs for the depth of stimulation and the amount of explosives used,the Byte Track object tracking algorithm is selected to track and count the placement rods and explosives in the well videos and stimulation videos.To address the issue of ID-switch caused by obstruction during the tracking process,an improved Byte Track algorithm is proposed.Based on Byte Track,a Re ID(re-identification)branch is added,utilizing the Res Ne St50 to train a re-identification model on an external dataset.The Re ID model is then added to the first association matching stage of Byte Track,where the effective information of the motion model and appearance model is used to measure the similarity between the detection target and the tracking trajectory,thereby improving the tracking accuracy and achieving correct counting of the placement rods and explosives in the video sequence.This approach is helpful for quality inspectors assess whether the depth of stimulation meets the requirements of exploration design and strictly regulate the use of explosives.Finally,detection and tracking experiments are conducted on the capture data,and the availability of the modified algorithm is validated through evaluation metrics and visualization.The results demonstrated that the proposed method can meet the requirements for quality monitoring of seismic excitation and reception operations.