班菲尔脐橙可溶性固形物近红外光谱特征谱区选择

为探讨快速无损检测班菲尔脐橙可溶性固形物(TSS)含量的方法,利用多元散射校正对脐橙1 000 ～2 500 nm近红外光谱进行了预处理,并用偏最小二乘法(PLS)、区间偏最小二乘法(iPLS)和联合区间偏最小二乘法(siPLS)分别建立预测模型.结果表明,采用siPLS将光谱划分为17个子区间,利用其中的第4(1 267～1355 nm)、5(1 356 ～1 443 nm)、9(1708～1795nm)、15(2 236 ～2 323 nm)号4个子区间联合建立的TSS模型效果最佳,其校正集决定系数和均方根误差分别为0.9109和0.331 2.预测集决定系数和均方根误差分别为0.878 9和0.448 7,主因子数为6个.研究表明,近红外光谱技术结合siPLS可优选出表征班菲尔脐橙TSS含量信息的特征光谱区间简化预测模型,同时提高模型预测能力和精度.

Selection of NIR Efficient Regions for Total Soluble Solids of Bamfild Navel Orange

In order to explore a nondestructive approach to predict total soluble solids (TSS) of Bamfild nave orange, multiplicative scatter correction (MSC) was used to preprocess the NIR spectra (1000~2500nm) of navel orange. Partial least square (PLS), interval partial least square (iPLS) and synergy interval partial least square (siPLS) methods were employed to establish the calibration models of TSS. The results showed that siPLS model was obtained with 17 intervals and the optimal combinations of intervals selected were No.4(1267~1355nm), No.5(1356~1443nm), No.9(1708~1795nm) and No.15(2236~2323nm). The coefficient of determination r was 0.9109 and RMSECV was 0.3312 in calibration set, and r was 0.8789 and RMSEP was 0.4487 in prediction set. It was achieved with six factors. This work demonstrated that NIR spectroscopy with siPLS model could not only select the efficient regions for TSS of orange and simplify the model, but also improve precision.