Holz
© Fotalia
Holz

Mathematics Supports Curb Illegal Timber Trade

Holzbretter
© iStockphoto
Holzbretter

Wood Species Identification Using Artificial Intelligence (AI)

Project KI-WOOD

In the research project, we are developing automatic image recognition systems for the identification of wood species using artificial intelligence in cooperation with the Thünen Institute for Wood Research in Hamburg.

In 2013, the Federal Republic of Germany committed itself to preventing the import and distribution of illegal wood species and to implementing the Washington Convention on International Trade in Endangered Species of Wild Fauna and Flora. Among other things, this requires a clear identification of the wood species in the various materials and products made from wood. Our research supports us in this process.

Within the framework of the project, we are developing an optical image recognition system together with the Thünen Competence Center Wood Origin for the identification of woods in fibrous materials (pulp-/paper products) in order to fulfil the required declaration obligations regarding species identification according to the European Timber Trade Regulation (EUTR) in trade and to check them on a large scale. The main focus at Fraunhofer ITWM is the development of the AI analysis software.

With this system, more extensive checks can be carried out using AI and material flows can be monitored more intensively. Every testing laboratory that is enabled to carry out simple checks on the international trade in paper with the developed AI systems is a multiplier in curbing illegal logging.

Neural Networks Support Analysis

The clear recognition and delimitation of structural characteristics for a doubtless wood species identification currently requires sound scientific training/expertise and, above all, access to documented reference preparations. Our cooperation partner - the Thünen Institute for Wood Research - has more than 50,000 reference preparations at its disposal.

Our joint task is to develop neural networks for the unambiguous identification of wood on the basis of high-resolution microscope images and to compare these with microstructure analysis methods. At present, it is still unclear which methods are more suitable.

The necessary tools for this project (including annotation and augmentation) have been developed in part in other AI projects of the Image Processing department over the past year.

Furthermore, Our Department Has Developed Many Software Tools to Support AI Methods:

  • An annotation tool that allows users to classify images or highlight areas of interest. The tool has a direct interface to the deep learning software package TensorFlow from Google.
  • Test tools to validate the training, including regression tests, outlier detection, etc. 
  • Augmentation algorithms to increase the size of the database
  • The standard image processing tools distributed by BV for 2D image processing (ToolIP) and 3D image processing (MAVI). The MAVI tool is used to analyse, among other things, wood materials based on computer tomography images.

 

Fiber Analysis of the Institute of Wood Research

Fiber Analysis Eucalyptus
© Institute of Wood Research
Fiber Analysis Eucalyptus
Fiber Analysis Pinus
© Institute of Wood Research
Fiber Analysis Pinus
Fiber Analysis Pinus
© Institute of Wood Research
Fiber Analysis Pinus