Classification and analysis method development in order to predict wood coating quality and appearence, and industrial prototype – OPTIFIN

A multimodal imaging bench was set up to create an «image« database serving as a reference for the study. It combines a near-infrared hyperspectral imager with a light polarization system and a 3D color linear camera with multiscan function (color, grayscale, diffusion of a LASER profile, 3D profilometry). In order to exploit the NIR hyperspectral images, different online hyperspectral unmixing algorithms respecting the imager's acquisition scheme were developed. A physicochemical interpretation of the spectral signatures obtained was also carried out by analyzing typical samples by FTIR microscopy in the near and mid-infrared and by NIR hyperspectral microscopy. After defining all the relevant parameters for extracting images to characterize the appearance of the wood and distinguish the different classes, methods for image fusion and extraction of criteria as well as classification of wood products were developed. The proposed algorithms were tested with experimental data to evaluate their effectiveness in terms of accuracy, precision and speed for real-time classification.

Hyperspectral image unmixing algorithms have demonstrated their relevance and performance in the analysis of wood samples, particularly with regard to sapwood/heartwood discrimination, healthy knot/rotten knot discrimination, and the identification of backwood or reaction wood which is difficult to observe visually, but which has a particular impact on the quality of a finish. Particular attention was paid to the detection of sapwood, which is of real economic interest: its reliable detection at an early stage of production (parquetry, furniture, cooperage, etc.) means that material flows can be better valorized. However, its colorimetric variations make detection by color imaging highly unreliable. PIR hyperspectral imaging enables reliable detection of sapwood: it is insensitive to its colorimetric variations. Generally speaking, color and hyperspectral images provide different information. Color imaging enables the detection of colorimetric singularities, while hyperspectral imaging reveals structures that are almost invisible to the naked eye. Coupled color/hyperspectral imaging therefore appears to be a promising solution for detecting singularities in wood materials.

The objectives set out at the start of the OPTIFIN project were ambitious both from an experimental point of view and from the point of view of image processing and exploitation. The multimodal imaging platform implemented made it possible to obtain a unique labeled database by the number of samples and the types of images obtained: color, grayscale, diffusion of a LASER profile, 3D profilometry and NIR hyperspectral. The exploitation of this data is still ongoing and could lead to other innovative results. The valorization of all the results within the framework of the development of a software application implantable in an industrial device is envisaged.

Dahbi, R. Conception d’une chaîne de traitements pour la segmentation texture d’images multimodales de pièces de bois en chêne. Application à la détection des singularités et la discrimination du grain du bois. Automatique / Robotique. Université de Lorraine. 2023.

Nus, L.; et al. A semi-supervised rank tracking algorithm for on-line unmixing of hyperspectral images. International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2020. Barcelone, Spain. May 2020

Prévost, C.; et al. Hyperspectral Super-Resolution with Coupled Tucker Approximation: Recoverability and SVD-based algorithms. IEEE Transactions on Signal Processing. 2020, 68, 931-946.

Wood is a natural material whose rendering is enjoyed in many applications (flooring, furniture, panels layout, etc..). However, its heterogeneity and variability of surface make it difficult to control harmony looking and quality. This quality and harmony looking are mostly looked through a finish applied on wood (varnish, paint, etc.). The finish quality result and its harmony looking is affected by both wood surface and finish. Physical texture, acidity, presence of molecules that can migrate or react with finish are potentially impacting factors.
These technical difficulties related to heterogeneity and variability of wood surfaces generate significant non-quality costs and manual sorting in wood industries. Current industrial technical solutions are confined to detect only fex singularities, such as knots and perform classification based on fuzzy logic. Some singularities of the wood as sapwood is not always detectable to the eye while their impact is significant on the quality and the rendering of the finished surface. OPTIFIN project aims to develop an analytical method to predict and classify the quality of finishing based on a real-time wood surface analysis. This surface analysis will require the coupling of several acquisition techniques, usable in industry,(single camera spectrum, hyperspectral camera, polarization, telemetry, another spectrometry, etc..) and the coupling of several methods of signal processing (usual signal separation or recursive method, classification by fuzzy logic suitable for wood, another classifier). OPTIFIN project lasts for 36 months, includes two laboratories, a technical center and several manufacturers. It aims to:
• Develop methods and tools for rapid wood surface analysis, physico-chemical analysis coupled with texture-color wood, mainly for oak,
• The study, characterization and understanding of the physicochemical phenomena and singularities of the wood that impact the quality of finishes applied
• Based on the previous points, the development of a method of analysis and classification able to predict quality and looking of finishes.
The project is mainly built around the knowledge and skills developed on scientific classification of wood and various materials multi-spectral analysis of laboratories LCPME and CRAN. The technology center, CRITT Bois, coordinate this project and provides wood knowledge. Industrials BERRY Wood and OGF Industrie will give the course in order to develop knowledge and processes related to economic opportunities.