Jördis Sieburg-Rockel

The conversion of processes controlled by human expertise and know-how into automated operations is the goal of a wide range of modern research. This study reports on a blind test in which twelve unknown samples were analyzed by human experts and an AI system developed to check the genera of hardwoods used in paper production. Trained to detect hardwood cells in microscopic images of maceration slides, it assigns them to the nine most commonly used hardwood species. Softwoods were also added to the samples, to make the test as realistic as possible. Human experts achieved an accuracy of 0.96 for recognizing a genus contained in a sample. The two-stage machine recognition system achieved an accuracy of 0.79 at the defined threshold values of 0.7 for object recognition and 0.75 for genus probability. Human experts failed to recognize some genera contained in the samples, whereas the AI models did not miss any genus at these thresholds. A comparison of the results revealed the strengths and weaknesses of the independent methods. This study is an important step towards optimizing the automated image recognition system developed to support the implementation of the European Union Deforestation Regulation (EUDR) with regard to deforestation-free products/supply chains.

Deep understanding of the structural composition and growth of biological specimens is becoming increasingly important for the development of bio-based and sustainable material systems. Full-field nano-computed tomography is particularly suitable for this purpose as it allows for non-destructive 3D imaging at high spatial resolution. However, most biological samples are functionalized by water and respond sensitively to any changes in climate conditions, specifically relative humidity, by adjusting their material moisture content. To date, only a limited number of tomography instruments offer an in situ climate control option to users. These, however, are limited either by the range of relative humidity states, the long times required to change the climate state, or obstruction or attenuation of the beam. Here, the first fully automatized climate cell for in situ full-field nanotomography is presented. It has been designed, built and integrated at the nanotomography station at the P05 imaging beamline, operated by Hereon at the DESY storage ring PETRA III, Germany. The highly flexible and windowless design allows the humidity dependent swelling and shrinking of lignified plant cell walls to be studied in situ , using phase contrast nanotomography. The concept of this climate chamber can easily be integrated into other setups. It operates in the relative humidity range of 0–90% and can be controlled in a temperature range of 10–50°C. Climate conditions can be adjusted at any time, remotely from the control hutch by using a humidity generator. Results show that the developed setup maintains a stable climate during the entire duration of a tomographic scan at different humidities and does not obstruct the sample or hinder the imaging conditions. During the tomographic investigation the sample remains stable in the flow of the air stream and shows typical cell wall swelling and shrinking behaviour depending on the equilibrium moisture content. This new climate cell is now available to all users of the P05 nanotomography instrument for conditioning samples, serving a wide range of scientific applications.

Wood species identification plays a crucial role in various industries, from ensuring the legality of timber products to advancing ecological conservation efforts. This paper introduces WoodYOLO, a novel object detection algorithm specifically designed for microscopic wood fiber analysis. Our approach adapts the YOLO architecture to address the challenges posed by large, high-resolution microscopy images and the need for high recall in localization of the cell type of interest (vessel elements). Our results show that WoodYOLO significantly outperforms state-of-the-art models, achieving performance gains of 12.9% and 6.5% in F2 score over YOLOv10 and YOLOv7, respectively. This improvement in automated wood cell type localization capabilities contributes to enhancing regulatory compliance, supporting sustainable forestry practices, and promoting biodiversity conservation efforts globally.

Wood species identification plays a crucial role in various industries, from ensuring the legality of timber products to advancing ecological conservation efforts. This paper introduces WoodYOLO, a novel object detection algorithm specifically designed for microscopic wood fiber analysis. Our approach adapts the YOLO architecture to address the challenges posed by large, high-resolution microscopy images and the need for high recall in localization of the cell type of interest (vessel elements). Our results show that WoodYOLO significantly outperforms state-of-the-art models, achieving performance gains of 12.9% and 6.5% in F2 score over YOLOv10 and YOLOv7, respectively. This improvement in automated wood cell type localization capabilities contributes to enhancing regulatory compliance, supporting sustainable forestry practices, and promoting biodiversity conservation efforts globally.

We have developed a methodology for the systematic generation of a large image dataset of macerated wood references, which we used to generate image data for nine hardwood genera. This is the basis for a substantial approach to automate, for the first time, the identification of hardwood species in microscopic images of fibrous materials by deep learning. Our methodology includes a flexible pipeline for easy annotation of vessel elements. We compare the performance of different neural network architectures and hyperparameters. Our proposed method performs similarly well to human experts. In the future, this will improve controls on global wood fiber product flows to protect forests.

In this study, we explore the explainability of neural networks in agriculture and forestry, specifically in fertilizer treatment classification and wood identification. The opaque nature of these models, often considered 'black boxes', is addressed through an extensive evaluation of state-of-the-art Attribution Maps (AMs), also known as class activation maps (CAMs) or saliency maps. Our comprehensive qualitative and quantitative analysis of these AMs uncovers critical practical limitations. Findings reveal that AMs frequently fail to consistently highlight crucial features and often misalign with the features considered important by domain experts. These discrepancies raise substantial questions about the utility of AMs in understanding the decision-making process of neural networks. Our study provides critical insights into the trustworthiness and practicality of AMs within the agriculture and forestry sectors, thus facilitating a better understanding of neural networks in these application areas.

We have developed a methodology for the systematic generation of a large image dataset of macerated wood references, which we used to generate image data for nine hardwood genera. This is the basis for a substantial approach to automate, for the first time, the identification of hardwood species in microscopic images of fibrous materials by deep learning. Our methodology includes a flexible pipeline for easy annotation of vessel elements. We compare the performance of different neural network architectures and hyperparameters. Our proposed method performs similarly well to human experts. In the future, this will improve controls on global wood fiber product flows to protect forests.

Wood and its processing into particles are, combined, the largest cost factor in the production of particleboard, followed by the cost of adhesive. Thus, reducing their cost is a goal of process optimization. This study investigated whether possible savings could be identified and quantified by determining the particle surface using automated three-dimensional laser-scanning technology (3D Particleview, Fagus-Grecon). The focus was on saving adhesive by sieving out adhesive-consuming fines. It was shown that, currently, with the actual prices for wood (89 €/t), particle preparation (37 €/t), and adhesive (570 €/t), the resulting additional costs for particles are overcompensated by the savings for adhesive with high adhesive content (e.g., 19%). The assumption of uniform distribution of adhesive on the total surface of all particles was checked for correctness using digital reflected light microscopy (VHX-5000, Keyence). Since urea-formaldehyde (UF) adhesive commonly used in particleboard production can only be detected with increased effort, phenol-formaldehyde (PF) adhesive was applied for the tests. Ultraviolet microspectrophotometry (UMSP) was used to rule out excessive penetration of the adhesive into the wooden tissue of the particles. The examination of the distribution of the adhesive over the surface showed that smaller particle sizes tended to be more heavily coated with adhesive. This means that the calculated savings still underestimate the real-life potential or that potential savings exist even with lower adhesive prices or higher prices for wood.

Wood and wood products including particleboard imported into the European Union (EU) are subject to the EU Timber Trade Regulation (EUTR) since 2013. With regard to the correct and complete required declaration, many such products not produced in the EU are currently investigated independently. The particles processed for the manufacture of particleboard are significantly smaller than solid wood samples. Thus, the preparation for anatomical analysis is more complex, as is their identification. The composition of the taxa identified allows a distinction between certain assortments according to the origin of the processed wood: temperate Europe; plantation-grown (worldwide); tropical and subtropical Asia; temperate Asia; tropical Africa. Important information for the evaluation of EUTR-subjected timbers in particleboard is provided.