Research on intelligent identification of third-party damage of oil and gas pipelines based on improved YOLOv5 algorithm

Abstract

Abstract In pipeline integrity management, third-party damage is an important cause of oil and gas pipeline failure and leakage. Among them, the safety threat to long-distance buried pipelines caused by excavators and other engineering vehicles is more prominent. For this reason, this paper proposes an automatic identification method of YOLOv5 engineering vehicles for video monitoring in high consequence areas. Firstly, for the variability of sunlight conditions and construction environment under actual working conditions, Mosaic9 data enhancement is used to create a more complex background and improve the robustness of the model. Second, the coordinate attention mechanism is introduced in the prediction layer to enhance the feature extraction capability of the model, thus solving the problem of complex background information and inconspicuous features of detection. Finally, the EIOU loss function is used to replace the CIOU function to improve the accuracy of the predicted border regression, which solves the problem of calculating the difference value of width and height instead of the aspect ratio. A dataset consisting of construction site photos obtained from surveillance cameras in the high consequence area of a section of pipeline in Guangdong Province is used to recognize four types of common construction vehicles, namely, excavators, loaders, rollers and heavy trucks, and the model is trained and tested on the actual engineering dataset. The experimental results show that the overall average detection accuracy mean value of the improved YOLOv5 algorithm is 81.3%, which is 3.6% higher than the original algorithm. Compared with other algorithms, the average mean precision is improved by 3.6%, the accuracy is improved by 0.5%, and the recall is improved by 3.0%, which obviously solves the problem of leakage detection that occurs in the recognition, and provides an effective solution for practical engineering applications.