Biometric fingerprints of trees: log tracing from forest to sawmill and early estimation of wood quality – TreeTrace

Two sampling campaigns were conducted, one for each species in the project. The «spruce« campaign was conducted in close collaboration with the Austrian project partners in September 2018. It involved 100 4.5 m long logs collected in the Vosges mountains. End-of-log images were taken in the forest, then the logs were transported to Freiburg for further photos of the sections after transport. In order to test the log recognition algorithms, several photos were taken of each section by varying the tilt angle of the device. The logs were then scanned by an industrial X-ray tomograph at the Baden-Württemberg Forest Research Institute (FVA). Terrestrial LiDAR was also used to collect information on the three-dimensional shape of the logs to complement the information available on the sections to assess quality. Slices were cut from both ends of each log to be analyzed with hyperspectral cameras and then used to measure quality in the laboratory (wood density and ring widths).
The «Douglas« campaign primarily involved French teams, particularly LaBoMaP and AMVALOR for industrial testing. The BBF sawmill (in the commune of La Machine) agreed to equip its feed line with a camera and conduct the sawing experiment for board collection. After testing images at a logging site and then tracking the logs to the sawmill yard, we decided to harvest the logs from the yard by selecting four logging sites. Thirteen logs (12-13 m) were selected from each site. After ridging, 156 logs measuring 3.5 to 4 m were obtained. Slices were collected between each log. The logs were photographed in the park with a smartphone and then with the camera of the feed line, and in n sawn (with the exception of 4 logs which were used for peeling tests). The sawn products were collected on 32 logs (4 sites × 4 trees × 2 logs). The 346 boards were then transported to Cluny to be tested by a testing machine and then by static bending tests until failure.

The TreeTrace project has achieved excellent results on several levels. It has led to the initiation of interesting and fruitful collaborations with the University of Salzburg, in computer science and wood sciences, the project being very interdisciplinary. Also with the LaBoMaP of Arts et Métiers in Cluny for the connection with the downstream part of the sector, the project being rather positioned on the upstream part between the forest and the industry. The project has led to the creation of two databases deposited in open access on data.gouv.fr which have been used in several new projects since. TreeTrace has enabled numerous scientific advances both in the characterization of roundwood and in biometric traceability by analyzing photos of cross-sections of logs/bolts. The interest of artificial intelligence in addressing these issues has been demonstrated. TreeTrace has enabled us to become efficient and internationally recognized on these two themes and has enabled the emergence of new projects: EffiQuAss (ANR-21-CE10-0002) on the evaluation of wood quality; Biomtrace (With the ONF and financed by France Bois Forêt, in collaboration with the University of Salzburg) on ??the development of a biometric traceability method for oak logs; TraCertiBois (PIA France 2030 RéCLasSIF, associating ENSAM and the Institut Mines Télécom) on the traceability of sawn timber to certify its origin and quality.

Today, it is very important to continue the work initiated by first providing the missing algorithms and then testing these methodologies in a real-life setting in a given region to more precisely assess their potential. Projects complementary to the TreeTrace project are still underway, and others are under construction. The data collected during the project are freely accessible and will allow the research community to continue the work.

Prospects for the TreeTrace oak project with INRAE, LaBoMaP (Arts et Métiers Cluny), and Loria (on the quality assessment and traceability of oak logs and sawn timber).

Decelle, R.; et al. Ant Colony Optimization for Estimating Pith Position on Images of Tree Log Ends. Image Processing On Line (IPOL). 2022, 12, 558–581.

Decelle, R.; Jalilian, E. Neural Networks for Cross-Section Segmentation in Raw Images of Log Ends. In Proceedings of IPAS 2020 - Fourth IEEE International Conference on Image Processing, Applications and Systems, Gênes, Italy, 2020.

Decelle, R.; et al. Light U-Net with a New Morphological Attention Gate Model Application to Analyse Wood Sections. In Proceedings of ICPRAM 2023 - 12th International Conference on Pattern Recognition Applications and Methods. 2023.

Longuetaud, F.; et al. TreeTrace project database on traceability and quality assessment of round woods: Douglas fir sampling. Ann. For. Sci. 2022, 79(46).

Longuetaud, F.; et al. Traceability and quality assessment of Norway spruce (Picea abies (L.) H. Karst.) logs: the TreeTrace_spruce database. Ann. For. Sci. 2023, 80(1), 9.

With the increasing amount of imaging devices installed at sawmills, the importance of using these data for improving workflow and for increasing revenues in the wood processing industries is growing. In this context, challenging questions with respect to imaging and image processing technology arise, several of which will be tackled in this joint project.
The project considers two application cases as follows: The first application case is the question of tracing tree logs from the forest harvesting site to the sawmill by using biometrics related tree log recognition techniques based on image processing of cross-section data only. This approach of course assumes the additional availability of imaging sensors in the forest. Since there is a trend for installing CT imaging devices at sawmills, which are of course not available in the forest, the challenging issue of cross modality matching arises. The second application case is the determination of wood quality from cross-section imagery, applicable already in the forest, and/or at the sawmill. Obviously, these two application cases share many aspects:. (1) They can be combined at application level, i.e. wood quality may be determined already in the forest due to imaging devices available for the tracing application, and further refined using the sensors available at the sawmill. Conversely, CT data or RGB images from the sawmill, acquired to analyse the wood quality, can be used for the tracing application; (2) Ddata preprocessing and many features extracted are required for both, matching cross-section images as well as automated wood quality analysis; (3) The questions which imaging sensors should be employed and how the resulting data can be combined effectively have to be answered.
Thus, synergies arise between these two application cases which will be efficiently exploited in the project. A common data set for experimental validation can be used (which implies also sharing employed sensors), ground-truth data established with respect to annotating images can be shared, many software components implementing preprocessing (e.g., pith detection, cross-section texture segmentation, contrast optimisation) as well as feature extraction techniques (e.g., annual ring detection, reaction wood detection, knot detection, rot detection) can be developed jointly and shared subsequently.
The project will break new grounds in the area of wood imaging and processing of corresponding data with advanced algorithms in vision and machine learning with particular focus on cross modality processing. While those techniques are being developed for two specific application cases, the developed algorithms will be applicable to a wide range of applications in wood imagery processing and analysis as well as for other domains where similar settings arise.