Discrimination of Varieties of Red Wines Based on Independent Component Analysis and BP Neural Network

Abstract A new method of using Vis/NIR spectroscopy technique to identify the varieties of red wines was studied. Through comparing modeling performance built by different amounts of independent components, 20 independent components (ICs) extracted by independent components analysis (ICA) were employed as the inputs of the BP neural networks and were consider to be important parameter for calibration and validation, a three layers of BP neural network was built, category analysis was performed as well as the work of building mathematics model and optimizing the algorithm was completed. 5 samples from each variety and a total of 25 samples were selected randomly as the prediction sets. The remaining 150 samples were used as the training sets to build the training model which is validated by the samples of the prediction sets. The recognition rate of 100%, in addition, based on the independent component analysis, we selected two characteristic wavebands of red wines in reference to vector loading map of mixed matrix. So the pattern recognition methods developed in this paper not only played a good role in the classification and discrimination, but had a capability to extract the finger feature of red wine, and offered a new way to detecting and product developing in red wines.