Design and implement of variety discriminator of fragrant mushrooms based on Vis/NIR spectroscopy and BP-ANN

A new instrument for the variety discrimination of fragrant mushrooms was designed and fabricated. First, the principle of using visible and near infrared (Vis/NIR) spectroscopy for variety discrimination of fragrant mushrooms was introduced. Then, the method of using error back propagation artificial neural network (BP-ANN) in the Vis/NIR spectral data analysis was elaborated. Before applying BP-ANN in the data processing, principal components analysis (PCA) was used for spectral data compression. Due to its accumulative credibility up to 94.37%, the first group of three principal components (PCs) was chosen as the signal inputs of BP-ANN. The BP-ANN with three layers has been optimized for the node number in hidden layer. In the test, total 195 samples of three varieties of fragrant mushrooms were examined. Among them, 150 samples were picked randomly out as for the model-calibration and others for the model-verification. With only 4 samples misjudged, the total prediction rate reaches 91%. Finally, a new structure of variety discriminator of fragrant mushrooms based on microprocessor MSP430 CPU was illustrated. The result showed that the new microprocessor-based instrument integrating Vis/NIR spectroscopy with BP-ANN is practical as an approach of machine recognition of various fragrant mushrooms.