Wood anatomy meets machine vision as a tool to fight illegal logging
Automated species identification has been a dream for scientists for centuries. For timber identification, the wealth of wood anatomical variation between species (see picture) remains one of the most important methods, since DNA extraction of wood is particularly challenging due to the majority of dead cells. However, training wood anatomy experts is time-consuming and expertise is rapidly disappearing world-wide. Therefore, in collaboration with prof. Fons Verbeek (LIACS, Leiden University), we want to test how computer-assisted features from microscopic wood images can be automatically extracted and used in a classifier to improve wood-based species identification.
Based on transverse and tangential microscopic wood sections, we are applying standard texture methods and more state-of-the art convolutional neural networks (CNN) as a deep learning strategy, with a focus on the ebony wood genus Diospyros. Based on a published dataset including transverse microscopic pictures from 112 species, we found that CNN produced the best results and considerably increased the identification success up to 95%. This remarkably high success rate in computer-assisted identification using only transverse sections highlights the fundamental importance of wood anatomy in species identification, and motivates us to expand the existing database to an extensive, world-wide reference database with transverse and tangential microscopic images from the most traded timber species and their look-a-likes, which would be a valuable tool for all stakeholders involved in combatting illegal logging.