Xu Zhuopin, Li Xiaohong, Cheng Weimin, Zhao Guangxia, Tang Liwen, Yang Yang, Wu Yuejin, Zhang Pengfei, Wang Qi
Anhui Key Laboratory of Environmental Toxicology and Pollution Control Technology, Hefei Institutes of Physical Science, Chinese Academy of Sciences, Hefei, 230031, China.
Anhui Key Laboratory of Environmental Toxicology and Pollution Control Technology, Hefei Institutes of Physical Science, Chinese Academy of Sciences, Hefei, 230031, China; University of Science and Technology of China, No. 96 Jinzhai Road, Hefei, 230026, China.
Talanta. 2024 Jan 1;266(Pt 1):125004. doi: 10.1016/j.talanta.2023.125004. Epub 2023 Jul 29.
Compound fertilizer occupies a dominant position in the structure of fertilizer products in China. The contents of nitrogen, phosphorus and potassium are the key indicators affecting the fertilization efficiency and the price of compound fertilizers. Laser-induced breakdown spectroscopy (LIBS) and near-infrared spectroscopy (NIRS) are two rapid analytical techniques suitable for online monitoring of the above components in compound fertilizer. However, accurate LIBS analysis needs to overcome matrix effects and interference from environmental elements, and NIRS also has the limitation of not being able to directly detect inorganic components in compound fertilizers. The combination of LIBS and NIRS techniques, namely LIBS-NIRS data fusion, has the potential to reduce interferences in the detection of single spectroscopic techniques and further improve the analysis accuracy. This study compared the LIBS-NIRS data fusion methods under different optimization conditions, and found that CARS-OPF (competitive adaptive reweighted sampling combined with outer product fusion) and CARS-EWF (competitive adaptive reweighted sampling combined with equal weight fusion) are two effective intermediate data fusion methods which can achieve better quantitative analysis results than single spectroscopic methods. The root mean square errors of prediction (RMSEP) for nitrogen, phosphorus, and potassium contents in compound fertilizers by using CARS-OPF are 0.901, 0.693, and 1.52, respectively, and the RMSEP for those indicators by using CARS-EWF are 0.934, 0.719, and 1.60, respectively. In these two methods, the LIBS and NIRS characteristic variables of compound fertilizers are firstly screened by CARS algorithm, and then intermediate data fusion was carried out by using equal weight fusion or outer product fusion. Redundant variables in the original data can be well removed in the data fusion process to ensure the accuracy of the analysis. Therefore, the combined methods of LIBS-NIRS based on CARS-OPF and CARS-EWF could be well applied to the rapid and accurate detection of main elemental contents in compound fertilizers.
复合肥在中国肥料产品结构中占据主导地位。氮、磷、钾含量是影响复合肥施肥效率和价格的关键指标。激光诱导击穿光谱法（LIBS）和近红外光谱法（NIRS）是两种适用于在线监测复合肥中上述成分的快速分析技术。然而，准确的LIBS分析需要克服基体效应和环境元素的干扰，NIRS也存在无法直接检测复合肥中无机成分的局限性。LIBS和NIRS技术相结合，即LIBS-NIRS数据融合，有可能减少单一光谱技术检测中的干扰，并进一步提高分析准确性。本研究比较了不同优化条件下的LIBS-NIRS数据融合方法，发现CARS-OPF（竞争性自适应重加权采样结合外积融合）和CARS-EWF（竞争性自适应重加权采样结合等权融合）是两种有效的中间数据融合方法，与单一光谱方法相比，它们能获得更好的定量分析结果。采用CARS-OPF法对复合肥中氮、磷、钾含量的预测均方根误差（RMSEP）分别为0.901、0.693和1.52，采用CARS-EWF法对这些指标的RMSEP分别为0.934、0.719和1.60。在这两种方法中，首先通过CARS算法筛选复合肥的LIBS和NIRS特征变量，然后采用等权融合或外积融合进行中间数据融合。在数据融合过程中可以很好地去除原始数据中的冗余变量，以确保分析的准确性。因此，基于CARS-OPF和CARS-EWF的LIBS-NIRS组合方法可很好地应用于复合肥主要元素含量的快速准确检测。
