基于近红外光谱技术的水蜜桃糖度检测模型性能研究
Research on the Performance of Juicy Peach Sugar Content Detection Model Based on Near Infrared Spectroscopy

通过数据融合技术充分利用不同近红外光谱仪器间的信息，提高水蜜桃定量检测模型的可靠性。实验通过flame-NIR和USB2000+光谱采集系统分别采集相同水蜜桃样本的近红外光谱数据。之后，利用阿贝尔折光仪测出所有样本糖度真实值，分别构建糖度定量模型。在其基础上，构建两种光谱数据的融合模型，融合数据经过无信息变量消除(UVE)和遗传算法(GA)变量选择后构建UVE-PLS、GA-PLS。结果显示，相比于未进行数据融合构建的糖度定量预测模型，最优融合模型GA-PLS交叉验证均方根误差减少了22.19%、33.79%，均方根误差分别减少了14.24%、49.67%。结果表明，数据融合模型能充分利用flame-NIR和USB2000+光谱仪器的糖度信息，具有更好的检测能力。
The data fusion technology makes full use of the information between different near-infrared spectroscopy instruments, improves the reliability of the peach quantitative detection model, and realizes the rapid non-destructive detection of the peach sugar content. In the experiment, the near-infrared spectrum data of the same peach samples were collected by flame-NIR and USB2000+ spectrum collection system. After that, the true sugar value of all samples was measured by Abel refractometer, and a quantitative detection model of peach sugar content was constructed respectively. On the basis of it, a fusion model of two kinds of spectral data is constructed, and the fusion data is subjected to the elimination of information-less variables and the selection of genetic algorithm variables to construct UVE-PLS and GA-PLS. Results showed that the fusion model developed on two kinds of spectral data performed better than the quantitative model on the basis of the univocal spectral data, and the root mean square errors of cross validation of the optimal GA-PLS fusion model reduced 22.19% and the 33.79% compared with the univocal model, as well as that the root mean squared error of prediction reduced 14.24% and 49.67%. The results show that the data fusion models are superior to the pre-fusion models and have better detection capabilities. They can make full use of the sugar content information of flame-NIR and USB2000+ spectroscopic instruments to achieve quantitative detection of peach sugar content.