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。

To select characteristic NIR wavelengths for rapid and nondestructive detection of rapeseed oil content, a miniature near-infrared (NIR) spectrometer was used in combination with methods including competitive adaptive reweighted sampling (CARS), genetic algorithm (GA), successive projections algorithm (SPA), uninformative variable elimination (UVE), backward interval partial least squares (BIPLS), and synergy interval partial least squares (SIPLS). Subsequently, partial least squares regression (PLSR) and least squares-supported vector machine (LS-SVM) regression model were established and the parameters of LS-SVM model were optimized. The results showed that for PLSR model, 26 characteristic wavelengths selected by BIPLS + GA produced the optimum model, where the correlation coefficient (Rp) and root mean square error of prediction (RMSEP) for prediction sets were 0.9330 and 0.0075, respectively. For LSS-VM model, 13 characteristic wavelengths selected by SIPLS+GA produced the optimum model and the correlation coefficient (Rp) and root mean square error of prediction (RMSEP) for prediction sets were 0.9192 and 0.0055, respectively. The above results demonstrate that parameter optimization and wavelength selection can not only effectively simplify the quantitative analysis model for rapeseed oil content based on miniature NIR spectrometry, but also enhance prediction accuracy and prediction stability. These data provide useful technical references for further applications of miniature NIR spectrometry.

国家自然科学基金项目（31171697）；国家重大科学仪器开发专项(2014YQ491015)；江苏高校优势学科建设工程资助项目