Comparative performance of NIR-Hyperspectral imaging systems

Authors: MA, TE - Nagoya University; SCHIMLECK, LAURENCE - Oregon State University; DAHLEN, JOSEPH - University Of Georgia; Yoon, Seung-Chul; INAGAKI, TETSUYA - Nagoya University; TSUCHIKAWA, SATORU - Nagoya University; SANDAK, JAKUB - University Of Primorska; SANDAK, ANNA - University Of Primorska

Submitted to: Foundations
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/16/2022
Publication Date: 6/22/2022
Citation: Ma, T., Schimleck, L., Dahlen, J., Yoon, S.C., Inagaki, T., Tsuchikawa, S., Sandak, J., Sandak, A. 2022. Comparative performance of NIR-Hyperspectral imaging systems. Foundations. https://doi.org/10.3390/foundations2030035.
DOI: https://doi.org/10.3390/foundations2030035

Interpretive Summary: The use of near-infrared hyperspectral imaging (NIR-HSI) in the spectral range either between 900 nm and 1700 nm or 1000 nm and 2500 nm is rapidly expanding in a variety of fields, ranging from plants and biological materials to food quality and safety. Similarly, wood-related NIR-HSI research is being increasingly reported in the literature. Conventional NIR spectroscopy (NIRS) provides very high spectral resolution, but the spectral information is only limited to a relatively small spatial area. This may be disadvantageous when heterogenic materials such as large wood specimens (e.g., logs, lumber, or even cross-sectional discs) are analyzed because it is difficult to adequately represent a within-sample variation without collecting spectra from multiple sampling locations with a spectrometer. It has been demonstrated that spectral data provided by NIR-HSI images produce wood property calibrations whose performance is equivalent to those obtained using a benchtop NIRS spectrometer. Although this result was encouraging as compared to NIRS technology, NIR-HSI cameras still have a lower spectral resolution compared with NIR benchtop spectrometers. Furthermore, because NIR-HSI cameras vary considerably in terms of their spectral and spatial resolutions and wavelength ranges among different camera manufacturers, the performance of wood property calibrations may be influenced by these factors. Hence, this study utilized three different NIR-HSI cameras to acquire images from the transverse surface of one-hundred Douglas-fir (Pseudotsuga menziesii) samples and develop and compare calibration models and their predictive performance. The NIR-HSI camera-based models to predict specific gravity (SG) and stiffness (MOE ) were compared with those based on a benchtop NIRS spectrometer. The results indicated that the 900 - 1700 nm range of a NIR-HSI camera provided the best models for MOE, whereas all three HSI cameras provided similar SG results. The models based on the NIRS spectrometer heavily favored wavelengths greater than 1900 nm.

Technical Abstract: Near-infrared (NIR) spectroscopy (NIRS) allows for the rapid estimation of a wide range of wood properties, and it is now one of a suite of tools available for nondestructive evaluation. Typically, NIRS studies on wood have utilized benchtop spectrometers where the sample (e.g. solid wood, milled wood, chips) is either placed in front of the NIR beam, or the spectrometer is fitted with a fiber-optic probe. Over the last 20 years, efforts to utilize NIR hyperspectral imaging (NIR-HSI) to examine wood and wood products have increased in a variety of applications. Compared to benchtop NIR systems, NIR-HSI has several advantages (e.g. speed, visualization of spatial variability), but the data typically has a lower spectral resolution, or spectral sampling interval (SSI), and signal-to-noise ratio, compared to NIRS. Furthermore, the SSI and wavelength range varies considerably among different NIR-HSI cameras. NIR-HSI systems based on indium gallium arsenide (InGaAs) detectors have a wavelength range typically from 900-1700 nm, while short-wave infrared hyperspectral imaging (SWIR-HSI) systems based on mercury cadmium telluride (MCT) detectors have the ‘full’ NIR wavelength range from 1000-2500 nm similar to benchtop NIRS spectrometers. These factors may influence the performance of wood property calibrations. Thus, in this study, we compared commercially available one NIR-HSI (900 – 1700 nm) and two SWIR-HSI (1000 – 2500 nm) cameras with a NIRS benchtop spectrometer (900 – 2500 nm). Performance of calibration and prediction models on specific gravity (SG) and stiffness (MOE) was compared with one-hundred Douglas-fir (Pseudotsuga menziesii) samples. The limited wavelength range of a NIR-HSI camera provided the best models for MOE, whereas all three NIR-HSI and SWIR-HSI cameras provided similar SG results. The models based on the NIRS spectrometer heavily favored wavelengths greater than 1900 nm.