Soil micromorphological image classification using deep learning: The porosity parameter

• Published in: Applied soft computing Vol. 102; p. 107093
• Main Authors: Arnay, Rafael, Hernández-Aceituno, Javier, Mallol, Carolina
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.04.2021
• Subjects: Archeology; Deep learning; Image classification; Microstructures; Soil micromorphology; Deep learning; Archeology; Microstructures; Soil micromorphology; Image classification
• ISSN: 1568-4946; 1872-9681
• DOI: 10.1016/j.asoc.2021.107093

Identifying components and microstructures in soil and sediment thin sections is one of the many subjects of analysis in archeological research, as these features can provide information regarding the deposit from which they were extracted, such as its origin and nature, clues about their associated human contexts or alteration processes they might have undergone over time. This article presents a Deep Learning system based on Convolutional Neural Networks (CNN) to classify different porosity types of structures in photomicrographs from archeological soils and sediment thin sections, as a first step to build and expand a database that will boost research in this field of archeological research. The results obtained are encouraging and show that the presented models can be successfully applied to this classification task. The trained models have been used to estimate the quantity of the different microstructures in test images, obtaining a median error of around 2%. [Display omitted] •CNN models can accurately classify different porosity types of structures in images.•These models had not previously been applied to soil micromorphology.•This work can be the starting point to automatic soil micromorphology image labeling.