Identification of Pipe Flanges in GPR Images by Using Neural Networks

The simple-offset GPR reflection methodology has demonstrated very good detection efficiency of pipe flanges. While this method allows prospecting large sections of pipelines in relatively short times, data interpretation can take considerably longer time.

In this work, we explore the use of Artificial Neural Networks (ANN) -one of the Supervised Learning techniques most frequently applied to GPR data-, for automatically identifying pipe flanges reflections in SO-GPR images.

First, we trained several ANNs using simulated SO-GPR image patterns. The achieved performances varied with the network structures. Based on those results, an optimized ANN structure was determined. Then, this network was tested with synthetic and experimental data, providing satisfactory accuracy levels in both cases. This indicates than ANNs can be a very valuable tool for this type of application, especially useful in the case of large surveys. While data inspection by a qualified interpreter to check for false positive or false negative network predictions would still be necessary, processing time could be significantly reduced.