PIPENet: A Semantic Segmentation Approach to Pipeline Component Detection from Magnetic Flux Leakage Readings

We developed a Semantic Segmentation model that is the weighted average of the four different models to detect the components of a pipeline using the Magnetic Flux Leakage (MFL) data. To the extent of our knowledge, this is the first time Semantic Segmentation has been used for this purpose and we were able to use the model not only to detect and classify the objects but also annotate the boundaries of them. The result of this work has reduced the work that our data analysts spend in detecting, classifying and boxing pipeline anomalies and replaced it with a confirmatory task. This work also showed promising results for future use of this approach in more complicated pipeline inspection tasks such as corrosion detection and weld counting.