均方误差
特征选择
滑石
残余物
特征(语言学)
波长
近红外光谱
均方根
计算机科学
模式识别(心理学)
数学
光学
人工智能
材料科学
统计
算法
物理
语言学
哲学
量子力学
复合材料
作者
Changhao Bao,Changhao Zeng,Jinming Liu,Dongjie Zhang
出处
期刊:Applied Optics
[The Optical Society]
日期:2022-06-27
卷期号:61 (19): 5790-5790
被引量:16
摘要
Excessive illegal addition of talc in flour has always been a serious food safety issue. To achieve rapid detection of the talc content in flour (TCF) by near-infrared spectroscopy (NIRS), this study used a Fourier transform near-infrared spectrometer technique. The identification of efficient spectral feature wavelength selection (FWS), such as backward interval partial-least-square (BiPLS), competitive adaptive reweighted sampling (CARS), hybrid genetic algorithm (HGA), and BiPLS combined with CARS; BiPLS combined with HGA; and CARS combined with HGA, was also discussed in this paper, and the corresponding partial-least-square regression models were established. Comparing with whole spectrum modeling, the accuracy and efficiency of regressive models were effectively improved using feature wavelengths of TCF selected by the above algorithms. The BiPLS, combined with HGA, had the best modeling performance; the determination coefficient, root-mean-squared error (RMSE), and residual predictive deviation of the validation set were 0.929, 1.097, and 3.795, respectively. BiPLS combined with CARS had the best dimensionality reduction effect. Through the FWS by BiPLS combined with CARS, the number of modeling wavelengths decreased to 72 from 1845, and the RMSE of the validation set was reduced by 11.6% compared with the whole spectra model. The results showed that the FWS method proposed in this paper could effectively improve detection accuracy and reduce modeling wavelength variables of quantitative analysis of TCF by NIRS. This provides theoretical support for TCF rapid detection research and development in real-time.
科研通智能强力驱动
Strongly Powered by AbleSci AI