A Real-Time Multifactor Risk Monitoring Method for the Gas Pipeline Operation Process Based on Mix-Supervised Target Recognition

With the rapid expansion of urban gas infrastructure, significant issues such as pipeline aging have arisen, leading to an increase in gas pipeline repair operations. However, this process has also resulted in numerous safety accidents. The traditional manual supervision mode for pipeline repair processes has several limitations, including incomplete identification of risk elements and the inability to estimate risks quantitatively. To address these challenges, a safety monitoring method was put forward in this study for the visible risk elements of the gas repair operation process. This method involves the identification of five types of risk elements and the establishment of a target detection data set for gas repair operations. Moreover, a data annotation method based on mix-supervised learning is proposed, which significantly enhances data annotation efficiency and saves 50% of marking time compared with manual annotation while maintaining an acceptable level of accuracy. Additionally, a visual risk element recognition model for the gas repair process was developed by using the YOLOv5 algorithm. The test results demonstrate that the detection accuracy of the visible risk element achieved in this research is 92.9%. These findings can assist in identifying potential safety hazards for personnel, equipment, and the environment during pipeline repair operations.