Adaptive Compressive Sensing: An Optimization Method for Pipeline Magnetic Flux Leakage Detection

Abstract

Leakage from a submarine oil pipeline would have a great impact on the environment and ecological balance. Accurate detection of pipeline defects can ensure safety in the transportation of oil resources. The traditional detection optimization algorithm may lead to the absence of effective features. An adaptive compressive sensing image data augmentation method that analyzes the pixel distribution of small defect features has been proposed to solve these issues. On the basis of Focal-EIoU, a new box loss of Focal-GIoU is proposed which is suitable for pipeline defect detection. Furthermore, the incorporation of bi-level routing attention diminishes the reliance of Yolov5 on specific inputs effectively, thereby enhancing the generalization ability of the detection model. Comparative experiments show that compared with the conventional Yolov5 model, this method improves mAP50 and mAP50:95 by 6.4% and 15.1%, respectively, with mAP50 reaching 91.5% and mAP50:95 reaching 52.9%.