Discrimination of Hard-to-pop Popcorn Kernels by Machine Vision and Neural Networks

Popcorn is a highly commercial agricultural product. Repeated buying of a brand name of popcorn depends very much on its quality. One of the major quality aspects is the number of the hard-to-pop kernels. If the hard-to-pop kernels could be pre-screened prior to packaging, it would greatly enhance the quality of the popcorn and accordingly consumers’ satisfaction. In this study, experiments were conducted to discriminate the popcorn kernels which could be popped from those that could not. Neural networks and machine vision were used for this purpose. Results showed that the visible features gathered by the vision system used in this study could discriminate between the poppable and unpoppable kernels at a 75% accuracy rate using the neural-network approach. The discrimination rate could be enhanced by adjusting the number of neurons in the network, examining better morphological and colour features for input to the neural network, and increasing the sample size.