研究了使用近红外漫反射光谱对不同品种草莓进行无损鉴别的方法, 并分析了各品种草莓品质指标的差异性。 在4 545～9 090 cm-1光谱范围比较了反向传播人工神经网络、 最小二乘支持向量机及判别分析的分类模型性能, 发现拓扑结构为12-18-3的反向传播神经网络模型分类结果最优, 校正集和预测集分类正确率分别为96.68%和97.14%, “甜宝”(n=99)、 “丰香”(n=100)和“明星”(n=117)样品的单独判别正确率分别为94.95%, 97%和98.29%。 对三个品种样品的可溶性固形物、 可滴定酸、 pH值及固酸比品质指标进行了单因素方差分析, 发现四个指标含量均存在明显差异, 分析成分指标数据的主成分得分发现不同品种草莓存在明显的聚类趋势。 结果表明, 近红外光谱与反向传播人工神经网络结合可有效鉴别不同品种的草莓, 且不同品种草莓化学成分含量的差异为近红外光谱分类提供了理化解释。

Strawberry variety is a main factor that can influence strawberry fruit quality. The use of near-infrared reflectance spectroscopy was explored discriminate among samples of strawberry of different varieties. And the significance of difference among different varieties was analyzed by comparison of the chemical composition of the different varieties samples. The performance of models established using back propagation-artificial neural networks (BP-ANN), least squares-support vector machine and discriminant analysis were evaluated on spectra range of 4 545～9 090 cm-1. The optimal model was obtained by BP-ANN with a topology of 12-18-3, which correctly classified 96.68% of calibration set and 97.14% of prediction set. And the 94.95%, 97% and 98.29% classifications were given respectively for “Tianbao” (n=99), “Fengxiang” (n=100) and “Mingxing” (n=117). One-way analysis of variance was made for comparison of the mean values for soluble solids content (SSC), titratable acid (TA), pH value and SSC-TA ratio, and the statistically significant differences were found. Principal component analysis was performed on the four chemical compositions, and obvious clustering tendencies for different varieties were found. These results showed that NIR combined with BP-ANN can discriminate strawberry of different varieties effectively, and the difference in chemical compositions of different varieties strawberry might be a chemical validation for NIR results.