论文部分内容阅读
为实现油菜籽含油率快速无损检测,采用微型近红外光谱仪,结合竞争性自适应重加权(CARS)、遗传算法(GA)、连续投影算法(SPA)、无信息变量消除法(UVE)、向后区间偏最小二乘法(BIPLS)、联合区间偏最小二乘法(SIPLS)等方法优选油菜籽含油率近红外光谱特征波长,建立偏最小二乘回归(PLSR)和最小二乘支持向量机(LS-SVM)定量分析模型,同时对LS-SVM模型参数进行优化。研究表明,对PLSR模型,BIPLS+GA优选的26个特征波长建模效果最好,其预测相关系数(Rp)和预测均方根误差(RMSEP)分别为0.9330和0.0075,对LS-SVM模型,SIPLS+GA优选的13个特征波长建模效果最好,预测相关系数(Rp)和预测均方根误差(RMSEP)分别0.9192和0.0055。证明了波长优选和参数优化可有效简化油菜籽含油率近红外光谱定量分析模型,提高模型预测精度和稳定性,为进一步拓展微型近红外光谱仪的应用提供技术参考。
In order to realize rapid and nondestructive detection of oil content in rapeseed, micro-NIR spectroscopy, combined with CARS, GA, SPA, UVE, Partial least square regression (PLSR) and least squares support vector machines (LS) were used to optimize the RLS characteristics of oilseed rapeseed by means of BIPLS, SIPLS, -SVM) quantitative analysis model, while the LS-SVM model parameters are optimized. The results show that the 26 feature wavelengths of BIPLS + GA are the best models for PLSR modeling, and the prediction correlation coefficient (Rp) and prediction root mean square error (RMSEP) are 0.9330 and 0.0075, respectively. For LS-SVM model, The best 13 characteristic wavelengths of SIPLS + GA are best modeled, and the predicted correlation coefficient (Rp) and predicted root mean square error (RMSEP) are 0.9192 and 0.0055, respectively. It is proved that wavelength optimization and parameter optimization can effectively simplify the quantitative analysis model of near-infrared spectrum of rapeseed oil content, improve the prediction accuracy and stability of the model, and provide technical reference for further application of the micro-NIR spectrometer.