Applying supervised deep convolutional neural network to identify bones from archaeological assemblages
Section: Archaeological Sciences
My research aims to bring together current machine learning and computer vision techniques and archaeological finds processing. I am interested in applying machine vision particularly in zooarchaeology, as large portion of all faunal assemblages are fragmented, and they would benefit from a faster processing methodology. Moreover, faunal identification is considered by some as lacking in quality control and assessment, meaning that zooarchaeologists often have no practical means to verify and reproduce theirs and others' results. In relation to this reproducibility issue, a notable lack of inter-analyst consistency - let alone the lack of within-study consistency tests - leaves even the best zooarchaeological studies open for criticism. These reproducibility and consistency issues therefore make zooarchaeological analyses rather subjective. By using image pixel data of bones from modern reference collections as the baseline input for a deep convolutional neural network, it is hoped that software capable of zooarchaeological identification will provide consistency and comparability for future zooarchaeological analyses.
Koneen Säätiö - Kone Foundation
- BSc Archaeology, Bournemouth University, 2013
- MA Archaeology (research), Universiteit Leiden, 2016
- Conference papers
Sipilä, I.M.V., 2017. The Evolution of Neanderthal Mobility. Poster presented at the 7th Annual European Society for the study of Human Evolution Meeting 21-23 September 2017, Leiden, the Netherlands.