| The pipeline has the advantages of safety and high efficiency when transporting gas,liquid and other objects.It is an important link to detect the change of the pipeline itself and the surrounding environment.The traditional manual survey operation has certain limitations in terms of efficiency and cost.How to realize the automatic and high-precision pipeline change detection is a subject worth studying.Based on the unmanned aerial vehicle(UAV)technology,this study acquires multiple images of the pipeline coverage area,splices them into complete image data of the detection target,and then uses deep learning methods to carry out research on multi-phase detection of pipeline changes.This research includes:(1)Since one UAV image can’t cover all detection targets and regions,how to accurately splice multiple continuous target images is a prerequisite for high-precision pipeline change detection.To solve this problem,based on SIFT operator,the RANSAC algorithm is used to eliminate outliers,the Levenberg-Marquardt algorithm is used to refine the single response matrix,and the image multi-level grouping method is used to splice images etc.,are considered to construct a high-precision and robust optimized sift matching and splicing algorithm to achieve accurate splicing of pipeline images.(2)Based on deep learning methods,the pipeline change detection model based on UAV remote sensing images is designed and implemented,and the Mask R-CNN network is optimized with the Adam algorithm.The deep learning-based pipeline change detection is implemented in the Keras framework.Using this model,the whole image after splicing in above-mentioned(1)can be used as input to realize pixel-level pipeline change detection.(3)Using PyQt5 as the system development platform,a prototype system of pipeline change detection based on the OpenCV library is designed and implemented.The system integrates the optimized SIFT matching and splicing algorithm.The change detection model of petroleum pipeline is based on deep learning methods,and changes detection of surface objects within a certain range of pipelines above and below the ground,and it realizes the automation and visualization of change detection process,analysis.Experiments show that the optimized SIFT algorithm proposed in this paper has obvious advantages in the registration and splicing of UAV remote sensing images.The deep learning model of pipeline change detection can effectively improve the change detection accuracy.This research has certain reference values for enhancing the automation and accuracy of the change detection of pipelines. |
