The BP Neural Network Design Applied on the Classification of the Apples

A kind of artificial neural network (BP-ANN) was used to classify the different kinks of apples with the reflected spectrum. The spectrum (from 460nm to 720nm) of some specks on the surface of the apples was measured with the spectrometer. A kind of BP-ANN with single hidden layer was devised to identify two sorts of the apples (the scar and the rotten) by the biological surface spectrum. To get the optimized parameters of the neural network, it was studied that the performance of the ANN with the different neurons numbers of the hidden layer, the different ranges of the output, the error of network with the different MSE(Minimum Squares Error) goal. Finally a multi-stage ANN was devised to identify the four sorts of apples, the fleckless, the pumped, the scar, and the rotten. 20,20,20 and 20 were selected as training samples respectively, and 10,10,10 and 10 were selected as testing samples respectively. This BP-ANN can achieve 87.5% accuracy if 15% noise was added.