Application of Artificial Neural Networks in NIRS Qualitative Analysis: Determination of Producing Area and Variety of Loquat

摘要

Near infrared (NIR) spectra of a sample can be treated as a signature, allowing samples to be grouped on basis of their spectral similarities. Near infrared spectroscopy (NIRS) combined with probabilistic neural networks (PNN) have been used to discriminate producing area and variety of loquats. Two varieties ofloquats (Dahongpao' and 'Jiajiaozhong') picked from two producing areas ('Tangxi' and 'Cunan' in Zhejiang province) were analyzed in this study. Principal component analysis (PCA) was appliedand the results indicated that the dimension of the vast spectral data can be effectively reduced and remaining as much information as possible. For each model, half of the samples were used to training the network and the remaining half were used to validate the network. The results of the PC-PNN models for discriminating the variety of samples from the same producing area or for discriminating the producing area of same variety samples are much better than those of the PC-PNN models for discriminatingthe variety or the producing area of all loquat samples. The results of this study show that NIRS combined with PC-PNN is a feasible way for qualitative analysis of discriminating fruit producing areas and varieties.