Use of Near Infrared Hyperspectral Imaging (NIR-HSI) and chemometric tools to study the kinetics of wheat kernel imbibition

One of the major seed qualities is the potential of the seed to germinate. The first phase of germination, so-called "imbibition phase", consists in water absorption and is a critical step in the development of the plant. Due to some abiotic and morphologic criteria such as climate, kernel position in spikelet…, the history of seeds is depending of the kernel itself. The kinetics of germination may be different between seeds and a good understanding of this phenomenon is crucial.
The most used technique for determining the imbibition rate is the weight method (ISTA 1985) [1]. This technique is quantitative, simple to perform, and nondestructive; however, the way of water diffusion inside the grain cannot be observed. One objective of the French Plant Phenotyping Network (Phenome) is to develop phenotyping methods, based on new sensors, able to answer to biological questions and environmental stresses.
Due to the properties of absorption of water in the Near-Infrared range, NIR hyperspectral imaging (NIR HSI) is particularly indicated to track diffusion of water in single wheat kernels. However, the pathway used by water to penetrate into the grains was not yet described and the imbibition process still remains misunderstood [2]. As the seeds must be generally completely immerged in water for hours to observe imbibition, we propose here a different method to provide an accurate description of water uptake and distribution in wheat grain tissues using NIR HSI in real time.
Five wheat kernels were cross-sectioned at 75% and the part containing the germ is placed in a specific sample Teflon holder, connected to a water tank. At initial time T0, water is added inside the sample holder so that the germ is immerged. NIR hyperspectral images of the surface are acquired before T0 and at regular time intervals.
Hyperspectral images were acquired with a pushbroom hyperspectral imaging system (BurgerMetrics®) covering the spectral range of 950-2500nm with a spectral resolution of about 7nm.
Data processing was performed by using PCA, multi-blocks PCA [3] and Non-Negative Least Square algorithms.
Onto the factorial map built with the first two principal components, pixels belonging to imbibed kernels were clearly separated from the others. PC1 was constructed by typical wavelengths of starch whereas PC2 was built from those of water.
Results obtained by multi-blocks PCA can be displayed as scores-images related to the different PC. PC1 scores-images show that the pixels associated to dry endosperm have high intensity, due to a signal associated to starch. This intensity decreased with imbibition time which was probably caused by the appearance of water, which covers the signal of starch. The introduction of weights in the algorithm led to the representation of samples in a factorial map, where samples are gathered according to the kinetics of imbibition.
By introducing spectral signatures as “reference components” into the Non-Negative Least Square algorithm, it became easier to accurately track the behavior of water at the surface of cross-sectioned kernel, and more generally those of the tissues.