Non-destructive classification and quality evaluation of proso millet cultivars using NIR hyperspectral imaging with machine learning

Millet is a small-seeded cereal crop with big potential and remarkable characteristics such as high drought resistance, short growing time, low water footprint, and the ability to grow in acidic soil. There is a need to develop nondestructive methods for differentiation and evaluation of the quality attributes of different of proso millet cultivars grown in the U.S. Current methods of cultivar classification are either subjective or destructive, time consuming, not allowing for the whole population to be tested, and requiring trained operators and special equipment. In this study, the feasibility of using near-infrared (NIR) hyperspectral imaging (900-1700 nm) to predict the quality attributes of proso millet (Panicum miliaceum L.) seeds as well to classify its different cultivars was demonstrated. Ten different cultivars of proso millet variety, which are the most popular in the US, investigated in this study included Cerise, Cope, Earlybird, Huntsman, Minco, Plateau, Rise, Snowbird, Sunrise, and Sunup. To reduce the large dimensionality of the hyperspectral imaging, principal component analysis (PCA) was applied, and the first two principal components were used as imaging features for building the classification models. The Classification performance showed a test accuracy rates as high as 99% for classifying the different cultivars of proso millet using gradient tree boosting ensemble machine learning algorithm. Moreover, using the partial least squares regression (PLSR) the coefficient of determination (R²) for quality prediction of proso millet seeds were 0.87, 0.80, 0.83, 0.93, and 0.92 for moisture content, crude protein, crude fat, ash, and carbohydrate, respectively. The overall results indicate that NIR hyperspectral imaging could be used to non-destructively classify and predict the quality of proso millet seeds.