LS-SVM和BP-ANN在草莓糖度NIR检测中的应用

Application of LS-SVM and BP-ANN to Quantify Soluble Solid Content in Strawberry by NIR

为了提高草莓糖度近红外光谱定量模型的性能,采用偏最小二乘法提取的潜在变量作为最小二乘—支持向量机和反向传播人工神经网络的输入变量,建立了草莓糖度的近红外定量模型,并与偏最小二乘模型结果进行了比较,建模所使用的光谱范围为6 000～9 000 cm-1。结果表明,所建立的最小二乘—支持向量机和反向传播人工神经网络定量模型的校正性能、预测性能和稳定性均优于偏最小二乘定量模型,最优模型为前10个潜在变量得分作为输入变量的最小二乘—支持向量机模型,其校正和预测相关系数分别为0.957和0.951,校正和预测均方根误差分别为0.279%和0.272%,剩余预测偏差为3.23,与以往研究文献相比,获得了较为理想的预测精度和稳定性能。

In order to improve performance of near infrared spectroscopy （NIR） models for quantitative analysis of solu- ble solid content （SSC） in strawberry, Least squares-support vector machine （LS-SVM） and back propagation-artificial neural networks （BP-ANN） with latent variables （LVs）, extracted by partial least squares （PLS）, as input were used to establish calibration models. And the performance were compared with PLS models. The spectral region used was 6000-9000 cm^-1.BP-ANN and LS-SVM models were superior to PLS model in calibration, prediction and robustness. Optimal models were obtained by LS-SVM with the first 10 LVs as input. The correlation coefficients and root mean square error of calibration and prediction were 0.957, 0.951, 0.279% and 0.272% , and the residual predictive deviation was 3.23, which were more satisfied in prediction accuracy and robustness than results reported by previous works. The results indicate that with LVs as input nonlinear methods of LS-SVM and BP-ANN offers more effective quantitative capability for SSC in strawberry.