Appl Opt. 2021 Sep 20;60(27):8400-8407. doi: 10.1364/AO.439291.
Selecting the decisive characteristic variables is particularly important to analyze the soluble solids content (SSC) of an apple with visible/near-infrared spectroscopy (VIS-NIRS) technology. The multi-population genetic algorithm (MPGA) was applied to variable selection for the first time, to the best of our knowledge. A hybrid variable selection method combined competitive adaptive reweighted sampling (CARS) with MPGA (CARS-MPGA) was proposed. In this method, CARS was firstly used to shrink the variable space, and then the MPGA was used to further fine select the characteristic variables. Based on CARS-MPGA, a nondestructive quantitative detection SSC model of an apple was established and compared with the models established by different variable selection methods, such as successive projections algorithm, synergy interval partial least squares, and genetic algorithm. The experiments showed that the CARS-MPGA model was the best. The number of modeling variables was only 64, and the determination coefficients, root mean squared error, and residual predictive deviation for the prediction set were 0.853, 0.443, and 2.612, respectively. The results demonstrated that the CARS-MPGA is a reliable variable selection method and can be used for fast nondestructive detection SSC of an apple.
选择决定性特征变量对于使用可见/近红外光谱（VIS-NIRS）技术分析苹果的可溶性固形物含量（SSC）尤为重要。据我们所知，多群体遗传算法（MPGA）首次被应用于变量选择。本文提出了一种结合竞争自适应重加权采样（CARS）和 MPGA（CARS-MPGA）的混合变量选择方法。该方法首先使用 CARS 缩小变量空间，然后使用 MPGA 进一步精细选择特征变量。基于 CARS-MPGA，建立了一种苹果的无损定量检测 SSC 模型，并与不同变量选择方法（如连续投影算法、协同间隔偏最小二乘法和遗传算法）建立的模型进行了比较。实验表明，CARS-MPGA 模型是最好的。建模变量的数量仅为 64，预测集的决定系数、均方根误差和剩余预测偏差分别为 0.853、0.443 和 2.612。结果表明，CARS-MPGA 是一种可靠的变量选择方法，可用于快速无损检测苹果的 SSC。
