Research project P4S/251/TIMBIR (Research action P4S)
As wood species misdeclarations violate environmental laws, enforcement officials need scientific tools to verify species claims. The implementation of the European Deforestation Regulation (EUDR) will substantially increase control quotas, requiring faster identification methods. However, current approaches are time-consuming and rely on highly trained personnel. Wood identification is also critical for the international transport, conservation, and restoration of cultural heritage objects (e.g. the RMCA’s vast African cultural heritage collection), which often cannot be destructively sampled due to their high value.
Therefore, TIMBIR aims to innovate the RMCA’s wood identification protocols by improving speed and accuracy, enabling multi-modal and multi-resolution input, and incorporating non-destructive X-Ray-μCT imaging.
We will build a comprehensive wood image database using the RMCA, Yangambi, and World Forest ID wood collections. In the previous SmartWoodID project, we developed a semi-robotic sample preparation pipeline and created an open-access database of >3700 flatbed scans representing >1000 species. TIMBIR will extend this with images from six different imaging techniques: low-resolution flatbed scanning, medium-resolution photography, high-resolution digital light microscopy, conventional light microscopy of thin sections, scanning electron microscopy (SEM), and X-Ray-μCT scanning. We will apply all imaging techniques on polished surfaces of ~800 samples of 200 species commonly traded in Europe or commonly occurring in the RMCA’s African cultural heritage collection. Additionally, we will compile a database of biochemical fingerprints using mass spectrometry (DART-TOFMS), a promising yet underutilized tool for species-level wood identification.
Next, we will develop AI algorithms capable of simultaneously analysing anatomical images and chemical fingerprints. Images will be converted into vector embeddings to combine with DART-TOFMS data in a multi-modal AI framework. Building on our success with CNNs trained on flatbed scans in SmartWoodID, TIMBIR will expand input variability (e.g. SEM, X-Ray-μCT), resolution variability, and predictive performance. The techniques we propose are complementary: surface imaging retains natural colour and texture; thin sections enable precise measurement of internal features like cell wall thickness; SEM offers ultra-high-resolution greyscale detail; X-Ray-μCT provides non-destructive 3D visualization. These techniques are all available at the RMCA or partner institutes where they are used for wood anatomical studies. Combining all of them in a single multi-modal AI framework will allow flexibility in the choice of the technique used for visualization of wood structure, which highly depends on the object that needs identification (e.g. X-Ray-μCT for art objects, flatbed scans for objects where sanding is allowed, thin sections or SEM when cutting is allowed). A user-friendly interface (e.g. Python-based Streamlit app) will be developed to enable practical use of the databases and AI models.
The system will be applied and validated in collaboration with the RMCA’s ENFORCE wood forensic centre (https://enforce.africamuseum.be/en), which handles over 200 identification requests annually from government agencies (e.g. customs at the port of Antwerp), companies, and art galleries. We will compare traditional and AI-based methods for accuracy and speed. The tool will also be tested on the RMCA’s African Cultural Heritage collection, focusing on X-ray-µCT imagery for non-destructive identification.
Once developed, the AI-based approach will significantly enhance RMCA’s wood identification capacity and enable non-destructive identification within the museum’s heritage collections. This supports object mobility, conservation, and contextual understanding. TIMBIR fully aligns with the P4S research priorities, RMCA’s strategic plan, and the FED-tWIN profile CONGOFORCE.
