NIR Hyperspectral Imaging of legume seed flours

In Europe, legume seeds are mainly used for feeding livestock. As legumes are extremely rich in vegetable proteins, complex carbohydrates, fibers, vitamins and minerals and contain very little fat, they also constitute a food of choice for humans. In addition to their major nutritional role, legumes are very beneficial to disease prevention (anti-cancer, diabetes), their production is more respectful of the environment (reduction of pesticides, no addition of fertilizers ...) and they could facilitate the autonomy of Europe towards imports of protein-rich products. The aim of the European LEGATO project (LEGumes for the Agriculture of TOmorrow) is to contribute to the sustainable reintroduction of legume seeds for human consumption in European cropping systems. Working on the major seed legumes, pea and faba bean, the project will focus on the identification and testing of novel legume breeding lines possessing valuable characters, especially by developing fast and efficient tools based on spectroscopic models. NIR Hyperspectral Imaging (NIR-HSI) was chosen because it combines all the advantages of NIR spectroscopy, such as ease of use, accuracy, reproducibility, multiparametric analysis and visualization, providing simultaneously spectral and spatial information within a sample.
A collection of 200 pea and faba bean flours of different origins, subspecies and culture conditions has been constituted. Each flour has been analyzed individually with a pushbroom hyperspectral imaging system (BurgerMetrics®) covering the spectral range of 950-2500nm. Hypercubes of size 231X318X212 are then treated by different chemometrical algorithms specifically developed with Matlab. Unsupervised exploratory data analysis is first performed using Principal Component Analysis (PCA) and is completed with an Unweighted multi-blocks PCA that allows the reconstruction of score–images showing the distribution of the components for each sample. A clustering method (K-means) is also applied to find typical spectral signature of homogeneous groups of spectra.
Exploratory data analysis with PCA enables to gather the samples according to their internal structures indirectly related to particle size, starch, protein or cellulose content. A link could be established with the origin, subspecies or culture conditions of the different samples. The K-means algorithm enables to extract the typical spectral signature of the different tissues. Then, each pixel is attributed to a group and by assigning an arbitrary color to each group, it is possible to build a false-color image for each sample. The number of pixels belonging to each group is recorded in a contingency-table which is introduced into a correspondence analysis. The factorial map shows the relationship between samples and groups.
Thanks to the LEGATO project, the potential of NIR-HSI combined with chemometrics will be assessed for seed legumes and will contribute to promote this innovative technique in the food and agriculture field.