Near-Infrared Hyperspectral Imaging to Differentiate Wheat Classes

Abstract

Differentiation of wheat classes is one of the important challenges to the grain industry. Even though some classes of wheat may look similar, their chemical composition and consequently, the end-product quality can vary significantly. Visual differentiation of wheat classes suffers from disadvantages such as inconsistency, low throughput, and labor intensiveness. A near-infrared (NIR) hyperspectral imaging technique was used to develop a classification model to differentiate eight wheat classes grown in western Canada (Canada Western Red Spring (CWRS), Canada Prairie Spring Red (CPSR), Canada Western Extra Strong (CWES), Canada Western Red Winter (CWRW), Canada Prairie Spring White (CPSW), Canada Western Amber Durum (CWAD), Canada Western Soft White Spring (CWSWS) and Canada Western Hard Winter (CWHW)). Wheat bulk samples (11% moisture content wet basis), 50 g each filled in petri dishes, were scanned in the wavelength region of 960 to 1700 nm at 10 nm intervals using a long wavelength InGaAs NIR Camera. Seventy five mean reflectance features were extracted from the scanned images and used for the identification of wheat classes using an artificial neural network (ANN) and a statistical classifier. Classification accuracy was 100% in classifying CPSR, CWES, CWHW, CWRS, CWRW and CWSWS wheat classes and above 97% for the other two wheat classes (CPSW and CWAD) using linear discriminant analysis. Using quadratic discriminant analysis, the classification accuracy was above 97% for all wheat classes. The classification accuracy of ANN models of two different training patterns (60% training: 30% test: 10% validation and 70% training: 20% test: 10% validation) ranged from 68-100% and 61-100%, respectively.