Explainable AI for the olive oil industry

Abstract

Understanding Machine Learning results for the quality assessment of olive oil is hard for non-ML experts or olive oil producers. This paper introduces an approach for interpreting such results by combining techniques of image recognition with knowledge representation and reasoning. The Design Science Research strategy was followed for the creation of the approach. We analyzed the ML results of fluorescence spectroscopy and industry-specific characteristics in olive oil quality assessment. This resulted in the creation of a domain-specific knowledge graph enriched by object recognition and image classification results. The approach enables automatic reasoning and offers explanations about fluorescence image results and, more generally, about the olive oil quality. Producers can trace quality attributes and evaluation criteria, which synergizes computer vision and knowledge graph technologies. This approach provides an applicable foundation for industries relying on fluorescence spectroscopy and AI for quality assurance. Further research on image data processing and on end-to-end automation is necessary for the practical implementation of the approach.