SmartWoodID: Smart classification of Congolese timbers: deep learning techniques for enforcing forest conservation

15 December 2020 → 15 March 2025
Federal funding: various
Research disciplines
  • Natural sciences
    • Machine learning and decision making
    • Plant morphology, anatomy and physiology
  • Engineering and technology
    • Data visualisation and imaging
  • Agricultural and food sciences
    • Forestry management and modelling
deep learning wood identification DR Congo
Project description

A substantial part of the timber trade is still illegal and illegal logging is the most profitable biodiversity crime. UN Environment estimates that illegal logging and the associated timber trade counts up to US$50 to $152 billion per year. Illegal logging involves a high risk of irreversible damage to ecosystems associated with the exploitation of highly sought after, sometimes protected, species. Timber regulations are already active (CITES, FLEGT, EUTR), but implementation and enforcement are a challenge. Currently, Belgium has the negative connotation of being the ‘hub of illegal timber trade’. 27.5% of the total EU28 imports of primary tropical timber products are imported via Belgium (mainly via the port of Antwerp). Wood identification is a key process in the enforcement that needs to check whether the shipment corresponds with the products mentioned on the accompanying documents. For this reason, there is a growing demand for timber identification tools that can be applied by law enforcement officers. 

The Tervuren xylarium is the Belgian governmental collection of wood samples. It is an internationally renown part of the federal scientific heritage, housed by the Royal Museum for Central Africa and comprises reference material of 13 000 different botanical species. One of the growing actual functions of the collection is supporting forensic research through verification of a species’ identity. The most common technique of timber identification is a wood anatomical assessment. Machine Learning methods are likely to be able to assist the wood identification process for non-specialists. Wood species have indeed characteristic features at different microscopical magnifications. However, some of those features are highly variable, which hampers the development of classical dichotomy identification keys that can be used by non-specialists. Moreover, many features seen on wood surfaces are to be understood as artifacts (fissures, traces of mechanical damage, fungi and insect attacks) and are not always easy to distinguish from diagnostic characteristics for the untrained eye. The Tervuren xylarium offers the most complete assemblage of reference material for the development of new wood identification  approaches.

The project proposal aims at automating part of the wood identification process by applying artificial intelligence techniques for the analysis of wood anatomical images of timber species of the Democratic Republic of the Congo. The tree flora of Central Africa comprises 3013 species, 27 of these belong to the class 1 commercial timber species of the DRC and are actually intensively logged and traded, 20 to class 2 (have potentially a big commercial value), 44 to class 3 (are considered to be promoted) and 879 to class 4 (commercial value is not yet known).  The project will use xylarium samples of all the species of the four classes and will take advantage of the power of modern deep learning approaches. The project will rely on expert wood anatomical descriptions which will serve as annotated training data to develop the software. The project will be unique because of the large number of African species, the application of deep learning and a database of standardized descriptions that will become available. In a first work package expert annotations of microscopic and mesoscopic images of transverse surfaces of 1000 Congolese wood species will be made. Work package 2 will develop an image processing pipeline for semi-automated annotation of microscopic and mesoscopic wood sections. Work package 3  will focus on the production of a user-friendly interface.