Method for Detection of Wheat Grain Damage with Application of Neural Networks

摘要

In the present paper the application of backpropagation type neural networks to assessment of wheat grain quality is described. The contours of whole and broken grains have been extracted using the log-polar transform, precisely normalised and then used as input data for the neural network. The network optimisation has been carried out and then the results have been analysed in the context of response values worked-out by the output neurones. By evaluation of the obtained results it has been found that correct recognition of the grain quality is possible on the 97% level for the learning set, and 94% level for the test set. The achieved recognition level allows the utilisation of the proposed method in industrial devices dedicated to grain quality evaluation.