Skip to Main Content
Recently developed optical inspection tools provide images from the inside of natural gas pipelines to monitor pipeline integrity. The vast amount of data generated prohibits human inspection of the resulting images. We designed an image processing and classification method to identify ab- normal events. Non-overlapping image blocks are classified into twelve categories: normal, black line, grinder marks, magnetic flux leakage inspector marks, single dots, small black corrosion dots, osmosis blisters, corrosion dots, longitudinal weld, field joint, cavity at a weld and longitudinal weld too close to field joints. Results compare different types of statistical classifiers. Features extracted from the pipeline image are designed to mimic the features humans use to identify the different classes. Difficulties include the large number of classes, the uneven costs associated with different errors, and training on a limited amount of expert classified data. Classification results show this to be a useful tool for pipeline monitoring.