The modelling near-infrared (NIR) spectroscopy data requires removal of scattering effects from the data before applying advanced chemometrics methods. Often different scatter-correction techniques are explored, and the scatter-correction technique with the best performance is selected. However, the information highlighted by different scatter-correction techniques may be complementary and their fusion may result in better models for predicting characteristics, such as meat quality. To test this, sequential and parallel preprocessing fusion approaches will be used in this work to fuse information from different scatter-correction techniques to try to improve the predictive performance of NIR models. Three different chemical properties, i.e., moisture, fat and protein content, were predicted. For comparison, partial least-squares regression (PLSR) was performed on standard normal variate (SNV) corrected data, as this is a widely used scatter-correction technique. Compared to this commonly used procedure, the scattering fusion approaches reduced the error and bias by up to 52% and 84%, respectively. The results suggest that fusion of scatter-correction techniques is essential to achieve optimal NIR prediction models for predicting meat characteristics such as moisture, fat and protein content.

建模近红外（NIR）光谱数据需要在应用高级化学计量学方法之前从数据中消除散射效应。通常会探索不同的散射校正技术，然后选择性能最佳的散射校正技术。但是，通过不同的散射校正技术突出显示的信息可能是互补的，并且它们的融合可能会形成用于预测特征（例如肉质）的更好模型。为了测试这一点，在这项工作中将使用顺序和并行预处理融合方法来融合来自不同散射校正技术的信息，以尝试改善NIR模型的预测性能。预测了三种不同的化学性质，即水分，脂肪和蛋白质含量。为了比较，对标准正态变量（SNV）校正的数据进行了偏最小二乘回归（PLSR），因为这是一种广泛使用的散射校正技术。与这种常用程序相比，散射融合方法分别将误差和偏差分别降低了52％和84％。结果表明，散射校正技术的融合对于获得用于预测肉类特征（例如水分，脂肪和蛋白质含量）的最佳NIR预测模型至关重要。