Hyperspectral imaging coupled with multivariate analysis and artificial intelligence to the classification of maize kernels

Abstract

Maize (Zea mays) is one of the key crops in the world, taking third place after wheat and rice in terms of cultivated area. This study aimed to demonstrate the potential of non-destructive hyperspectral imaging in the wavelength range of 400-1000 nm to discriminate between and classify maize kernels in three cultivars by using non-destructive hyperspectral imaging in the wavelength range of 400-1000 nm. Three cultivars of maize kernels were exposed to hyperspectral imaging with 20 replications. Predictor variables included 28 intensities of reflection wave for spectral imaging and 4 variables in terms of the weight, length, width, and thickness of a single kernel. The classification was successfully performed through Linear Discriminant Analysis and Artificial Neural Network methods, taking into account 32, 15, and 5 predictor variables. According to the results, Linear Discriminant Analysis with 32 predictor variables is characterized by a high degree of accuracy (95%). The most important predictor variables included the reflection wave intensity of the third peak, the wavelength intensity of 490 nm, the wavelength intensity of 580 nm, and the weight and thickness of a single kernel. Keywords: maize, classification, hyperspectral imaging, artificial intelligence.