Spectrum Recognition with Three-Stage Neural Network

In this paper, a new kind of three-stage neural network was developed to identify the sorts of the biological surface. The visible spectrum (from 380nm to 780nm) of the micro areas with some specks on the surface of the apples was measured with the self-made fiber sensor spectrometer. To sort the apples, A kind of BP-ANN with single hidden layer was devised to identify the characteristics on the biological surface automatically. To improve the performance of BP, A three-stage BP-ANN was devised to identify the four sorts of the apples, the fleckless, the bumped, the scared, and the rotten. It was also studied that the performance of the ANN with the different ranges of the output, the influence to the ANN if the noise was added to the input signals. 25,10,10 and 10 samples of four sorts were selected as training samples respectively, and 10,10,10 and 10 respectively were selected as testing samples. It proved that this kind of BP-ANN can achieve 90% accuracy if 10% noise was added.