应用可见/近红外光谱法对赣南脐橙可溶性固形物进行了无损检测研究。通过主成分分析,获取光谱的有效信息,将其作为人工神经网络的输入变量进行非线性建模。90个建模样品训练结果是,样品参考值与预测值之间的相关系数为0.9147,训练均方差为0.5203;38个未知样品预测结果是:样品参考值与预测值之间的相关系数为0.9033,预测均方差为0.6964,相对预测偏差4.5709%。实验结果表明基于人工神经网络的可见/近红外光谱法无损检测赣南脐橙可溶性固形物是可行的。

Non-destructive measurement of soluble solid content in Gannan navel oranges was carried out by visible/near-infrared spectroscopy detection method. Effective information of spectra was obtained by principal component analysis, and was used as the input variables of artificial neural network for building the nonlinear model. The results, based on calibration for 90 samples, are 0.9147 and 0.5203 for calibration correlation coefficient and root mean square error of calibration. The results, based on prediction for 38 unknown samples, are 0.9033, 0.6964 and 4.5709% for prediction correlation coefficient, root mean square error of prediction, and relative standard deviation (RSD), respectively. Experimental results show that visible/near-infrared spectroscopy detection method, based on artificial neural network, for non-destructive measurement of soluble solid content in Gannan navel oranges is feasible.