22nd EANN 2021, 25 - 27 June 2021, Greece

Contaminated Soil Detection: A proposal using Machine Learning with Hyperspectral imaging

Fernando Oliveira Duarte, Levi de Resende Filho, Hector Azpurua, Andre Almeida Santos, Jefferson Souza, Gustavo Pessin, Rosa Correa Pabon


  Crude oil leakages can produce damages to the environment and the nearby population. Oil companies perform continuous soil and vegetation surveillance around susceptible areas, aiming to identify potential leakages and mitigate their impacts. Traditional methods for monitoring oil leaks are based on visual inspection and losses on the pipeline‚Äôs flow-rate signal. These methods allow identifying significant leaks, but small and progressive leaks could pass unperceived for a long time. Therefore, aid monitoring methods that allow more agile soil contamination identification becomes a need. Hyperspectral imaging can offer that agility in the monitoring when employed with library references of contaminated soil. In this study, we propose an intelligent system to identify oil contamination and contamination level, using hyperspectral data as an input for machine learning models. Three different system architectures are proposed and evaluated, aiming to improve identification. The results show the accuracy of 98% in the classification of soils, 94% in oil type, and 93% in contamination level. These observations indicate that this technology has great potential for environmental monitoring of bare soils along pipelines and refineries.  

*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.