Determination of features and optimized model to classify the components of quality wheat seed using industrial digital camera image processing

The acquisition of basic knowledge in quality control of wheat seed using machine vision technology is important. The objective of this research was to develop hardware and appropriate software to determine seven-grain groups in wheat seed samples. Ninety-one features were extracted through 21000 single seed images and the shape, texture and color features were ranked. Five classification models were investigated. The highest classification accuracy was obtained by artificial neural network with two hidden layers and the first 35 superior features. In the test run of this model with independent data, classifying accuracy for big white wheat, small white wheat, broken white wheat, wrinkled white wheat, red wheat, barley and rye were 100, 96.7, 99.3, 90.3, 99, 99.7, and 98 percent respectively with the average of 97.6 %. Shape features were more prominent and textural and color characteristics followed it respectively. Average classification accuracy in models of linear discriminant analysis, quadratic discriminant analysis, K- nearest neighbor and artificial neural network with a hidden layer were 95, 96.7, 91.6 and 97.3 % respectively. In the context of this study, the machine vision system comprising an industrial digital camera and artificial neural network with two hidden layers was identified as a valuable system in the investigation of the visual qualities of wheat seeds.