Technology for Wood and Natural Fiber-Based Materials

Research Project

Spectral Imaging – New possibilities in measurement and separation technology

Moisture distribution, adhesives, plastics, wood – there are many things which the human eye or a camera are unable to differentiate between, as the characteristic absorption band of the sought substances is not found in visible light, but in near-infrared (NIR). In order to determine the distribution of specific materials on a surface or in a material flow, methods of near-infrared spectroscopy, multivariate data analysis and image processing are combined.

© Fraunhofer WKI | Manuela Lingnau
Schematische Darstellung eines Zeilenspektrographen.

Measurement principle

Spectral data from any point of a moving surface can be obtained using a special optical system (line spectrograph, see illustration) with which the spectrum of an image of a narrow excerpt, transverse to the direction of conveyance (target), is dissected point-by-point. The spectra are then photographed using a NIR scan camera (matrix detector). This results in equidistant spectral lines on one image line (spatial axis) which represent the localised intensities in the respective wavelength area. The WKI testing facility can produce hyperspectral images with a spatial resolution of around 1cm² to 1mm² per pixel and a spectral resolution better than 10nm. The material to be measured is transported on a conveyor belt with variable speed and measuring geometry. The spatial resolution results from the operating distance of the camera, the frame rate (up to 100Hz) and the feed rate. Halogen spotlights are suitable for the lighting. Special software has been developed in-house for the chemometric evaluation of the hyperspectral images.

Application examples and results

© Fraunhofer WKI
© Fraunhofer WKI

Representation of moisture and adhesive distribution

The strong absorption bands of water at around 1400nm and 1900nm can be used to visualise moisture distribution even in scenes with pronounced particle structures. The illustration (left) shows a hyperspectral image with partially dampened OSB strands with the intensity, determined over the entire wave length area, as grey-scale value. The illustration on the right shows a segmented image in which the intensities are shown only for the absorption band of the water, so that a clear differentiation can be made between the dampened regions (green) and the dry image areas (blue).

As UF resins also show a strong absorption at around 1500nm, glued particles in a scene with OSB strands lower ill., left)  are easily recognisable at these wavelengths (lower ill., right).

© Fraunhofer WKI
© Fraunhofer WKI

Differentiation between different adhesives

In chemometrics, various methods of multivariate data analysis are implemented in order to evaluate spectra. For example, it is possible to reduce the information content of spectra, which is distributed over various wavelengths and absorption bands, to its major components, thereby enabling qualitative differentiation or quantitative assessment of the spectra. Using the software developed by the WKI, this principle of “chemical imaging” can also be applied to hyperspectral images.

The upper illustration shows a classified image with non-glued pinewood strands upon which strands with various types of adhesive (left: PMDI, to the right of centre: UF, far right: MUPF) lay. The analysis of the major components (lower ill.) shows that the scores for the spectra are clearly grouped in the component area. In this case, only two major components are necessary in order to differentiate between three types of adhesive and the non-glued strands and therefore enable allocation of the image pixels to the substances represented there (red: non-glued, brown: PMDI, blue: UF, green: MUPF).

© Fraunhofer WKI
© Fraunhofer WKI

Recognition of extraneous material during the sorting of recovered wood

NIR spectroscopy is already used in separation technology in order to differentiate between the varying plastics in lightweight packaging. If, in the future, particles must be separated for e.g. the material recycling of waste wood, spectral imaging would be a suitable detection method. The upper illustration shows a mixture of broken pieces of coated and uncoated particle boards as well as solid wood and plastic objects, whilst the lower illustration shows (left) the score chart for two major components. In the image on the right, the pixels are classified according to their position in the component area, thereby enabling recognition of the coatings (green) and plastics (blue) and their subsequent removal as undesirable substances. 

Project funding

"Bildgebende und ortsauflösende Kontrolle des Klebstoffauftrags bei der OSB-Herstellung" – AiF Project 15242N (2007 to 2009) in co-operation with the Reutlingen Research Institute RRI

"ChOP-SIN – Chemometrische Online-Prozessanalyse durch Spectral-Imaging im erweiterten NIR-Bereich bis 2.2 µm mit Hauptkomponentenanalyse sowie Neuronalen Netzen" – ZIM Project KF2015010RR1 (2012 to 2014) in co-operation with inno-spec, Nürnberg

"DEMOWOOD – Optimierung des Stoffrecyclings und der Energiewiederverwendung aus Abfall und Abfallholz in verschiedenen Wertschöpfungsketten – Teilprojekt Produktion von Holzwerkstofplatten aus recycelten Spänen" WoodWisdom-Net, funded by BMBF, Ref. 033R060B (2010 – 2013)

"Molecular Sorting for Resource Effiency" (2011 - 2014) – funding within the framework of the Fraunhofer Society internal programme "Märkte von Übermorgen"