Underground Oil Pipeline Leak Detection Using CNN and SVM

Abstract

Abstract

    Oil leaks on land and water surfaces from pipeline cracks cause severe damage to the environment. Pipeline leak detection is an important and necessary step for pipeline safety management. This paper presents synthetic aperture radar (SAR) images as an approximate representation of the target scenes. The present study presents a transfer learning framework that uses deep learning convolutional neural networks (CNNs) for the purpose of pipeline leak detection. The investigation of CNN transfer learning is carried out using two distinct approaches: parameter-based CNN transfer learning and hybrid feature-based mechanisms. The optimal transfer learning model is selected using ResNet-50, DenesNet-201, Xception and AlexNet algorithms pre-trained on ImagNet. The results demonstrate that the feature-based CNN transfer learning approach Xception  combined with SVM proposed  in this study exhibits superior performance compared to parameter-based CNN transfer learning methods. The method Xception withe SVM achieved maximum accuracy of 99.8% ,recall 100%, precision 99.5 %and F1scor 99.7%.