Classification of Potato Varieties Using Isoelectrophoretic Focusing Patterns, Neural Nets, and Statistical Methods

Automatic potato variety classification by a multivariate statistical classification method and a combined neural network model is described. Both classification methods were based on digitized isoelectrophoretic patterns of the soluble tuber proteins. The statistical classification algorithm was based on a model that allows for individual stretching as well as displacement along the pH axis. The neural network architecture consisted of two layers: a self-organizing feature map and a feed-forward classifier. Twelve potato varieties were classified. The mean value of the recognition rates were 84.5 and 87.5% obtained by the statistical classification method and the neural network model, respectively. The results confirm the theory stated in earlier classification studies, that the automatic classification systems are well-established, independent of the origin of the samples, and unaffected by pattern deformations and variations in the background level of the electrophoretic gels. Keywords: Potato; isoelectric focusing; image processing; classification; neural networks