Research On Defect Recognition Method Of Pipeline Magnetic Flux Leakage Internal Detection Based On Object Detection

As the main transportation mode of important energy such as oil and natural gas,the accident rate of pipeline transportation increases with the increase of pipeline mileage and quantity.Monitoring pipeline safety risk has become the primary issue to ensure national energy production.However,the pipeline detection path is long and the amount of data collected by magnetic flux leakage detection is huge.Using the traditional manual interpretation method to identify magnetic flux leakage signals will take a long time,waste of energy,missed detection,false detection and other problems.Therefore,the establishment of automatic defect identification model of pipeline magnetic flux leakage internal detection signal is of great significance for pipeline defect identification.Aiming at the problem of intelligent defect recognition of pipeline magnetic flux leakage signal,an image recognition algorithm of pipeline magnetic flux leakage signal based on improved YOLOv5 network is proposed in this paper.Based on the YOLOv5 network,the target detection model YOLOv5 is improved by introducing the loss function Distance-IOU and the improved non maximum suppression algorithm,so as to improve the recognition accuracy of defects in pipeline magnetic flux leakage signal.Through the saturation magnetization of the pipeline to be tested,the magnetic flux leakage signal is collected,the magnetic flux leakage signal of the pipeline to be tested is imaged by the method of point tracing imaging,and the magnetic flux leakage curve image is processed to establish the magnetic flux leakage signal defect image data set.Using the data set to train the improved YOLOv5 target detection model,adjust the model parameters and establish the optimization model according to the recognition results;Using the optimization model,the defect of the magnetic flux leakage signal image is recognized,and the classification results,category scores and location information of the prediction targets of the analyzed data are obtained.After removing some repeated prediction results,the target detection results are marked on the input image to realize the defect recognition of the local magnetic flux leakage signal of the pipeline.Finally,using the migration learning algorithm,through the model migration of the trained YOLOv5 network,and introducing the small target detection layer into the migrated model,the defect identification of the overall pipeline magnetic flux leakage signal is realized,and the automatic defect identification model of the pipeline magnetic flux leakage internal detection signal is established.The results show that the improved YOLOv5 algorithm realizes the automatic detection and identification of pipeline defect magnetic flux leakage signal.Under the same training conditions,the accuracy of the improved YOLOv5 algorithm has been significantly improved compared with the original algorithm.Its accuracy in identifying the number of defects reaches 92.8%,which is 3.22% higher than the original algorithm.The average loss rate of the improved model loss function is 3.6%,which is 2.2% lower than the original YOLOv5 model,indicating that this method is feasible in the automatic identification and detection of pipeline defect magnetic flux leakage data.The fusion model of improved YOLOv5 algorithm and small target detection algorithm is studied and established to realize the identification and detection of small target pipeline defects.