Oil Pipeline Leak Detection Using Deep Learning: A Review on POC Implementation

Abstract

AbstractOil and gas production operations are key sources of environmental pollution which exposing the people and effect the human activity in the world. Petroleum Development Oman (PDO) is the leading exploration and production oil and gas companies in the Sultanate of Oman which lead to avoid adverse health effects in Oman. Oil pipline leakes could be undetected for a long time. However, the precise methods could help improve the oil leaking detection response process in channel required resources with more effectively to be concerned regions. Existing Synthetic-aperture radar (SAR) approaches are limited by their algorithm complexity which difficult to work with imbalanced data sets, doubts to select optimal features, and the relatively slow detection. Using deep learning approach could speed up the oil detection. convolutional neural network U-Net segmentation models based on oil leaking detection could be achieve promising automated results. However, there are insufficient features extraction due to loss of target to detect oil leaking or shadows in drone images that commonly appear in various size, shapes, and brightness levels, which the images that captured under many conditions. To overcome all these limitations, we utilized a deep learning model named Pyramid Scene Parsing Network (PSPNet). The proposed algorithm can probabilistically detect oil leak from drone imagery at the frequency of data collection. Thus, PDO Oman could reduce millions of Dollars when direct action from operators that received a quick true alarm of oil leaking. The effectiveness of the proposed method is demonstrated through A proof of concept (POC) based on a realistic dataset that collected history data that our deep learning algorithms achieved the perfect predict the oil leaking before occurs.