苹果近红外预测模型的传递研究

Study on near-infrared prediction model transfer for apples

中文关键词:  近红外  模型传递  直接校正法

英文关键词:Near-infrared  model transfer  direct standardization

中文摘要:

      目的 实现对相同型号和不同型号水果便携近红外仪进行模型传递。方法 采用标准变量变换(SNV)预处理, 无信息变量消除法(UVE)及与连续投影算法相结合(UVE-SPA)进行特征波长选择后, 再使用直接校正法(DS)进行模型传递。结果 对于相同型号的仪器K1, K2, 经SNV-UVE-SPA-DS算法处理后, 建立在K1上的苹果糖度校正模型成功传递到了K2上, R=0.9730, RMSEP=0.2375°Brix, 结果略优于截距法; 而经过SNV-UVE-DS算法, 传递到与K1不同型号J仪器上的模型预测结果为R=0.9375, RMSEP=0.3458°Brix。结论 合适的波长选择结合DS算法可以有效地用于相同型号及不同型号水果便携仪上的模型传递。

英文摘要:

      Objective To achieve the model transfer among the same type and different types of fruit portable near infrared instruments. Methods Standard normalized variant (SNV) was firstly used as a pre-treatment technique. Uninformative variables elimination (UVE), which also combined with successive projections algorithm (SPA) was used to select effective wavelengths before using direct standardization (DS) for model transfer. Results By using SNV-UVE-SPA-DS algorithm, the prediction model was successfully transferred between the same type instruments (K1, K2), R=0.9730, RMSEP=0.2375°Brix, which was slightly better than bias correction. The best transfer result of different type instruments was obtained by SNV-UVE-DS algorithm, R=0.9375, RMSEP=0.3458°Brix. Conclusion The result highlights that an apposite wavelength selection method combined with the direct standardization (DS) algorithm can be used for model transfer among the same type and different types effectively.