Laboratoire en Intelligence des Données (LID), Département de Mathématiques et Génie Industriel, Polytechnique de Montréal, Montreal, Quebec, Canada.
Laboratoire Poly-Industrie 4.0, Département de Mathématiques et Génie Industriel, Polytechnique de Montréal, Montreal, Quebec, Canada.
J Food Sci. 2024 Sep;89(9):5674-5688. doi: 10.1111/1750-3841.17252. Epub 2024 Aug 10.
Near infrared spectroscopy (NIRS) is an analytical technique that offers a real advantage over laboratory analysis in the food industry due to its low operating costs, rapid analysis, and non-destructive sampling technique. Numerous studies have shown the relevance of NIR spectra analysis for assessing certain food properties with the right calibration. This makes it useful in quality control and in the continuous monitoring of food processing. However, the NIR calibration process is difficult and time-consuming. Analysis methods and techniques vary according to the configuration of the NIR instrument, the sample to be analyzed and the attribute that is to be predicted. This makes calibration a challenge for many manufacturers. This paper aims to provide a data-driven methodology for developing a decision support tool based on the smart selection of NIRS wavelength to assess various food properties. The decision support tool based on the methodology has been evaluated on samples of cocoa beans, grains of wheat and mangoes. Promising results were obtained for each of the selected models for the moisture and fat content of cocoa beans (Rcv: 0.90, Rtest: 0.93, RMSEP: 0.354%; Rcv: 0.73, Rtest: 0.79, RMSEP: 0.913%), acidity and vitamin C content of mangoes (Rcv: 0.93, Rtest: 0.97, RMSEP: 17.40%; Rcv: 0.66, Rtest: 0.46, RMSEP: 0.848%), and protein content of wheat-DS2 (Rcv: 0.90, Rtest:0.92, RMSEP: 0.490%) respectively. Moreover, the proposed approach allows results to be obtained that are better than benchmarks for the moisture and protein content of wheat-DS1 (Rcv: 0.90, Rtest: 94, RMSEP: 0.337%; Rcv: 0.99, Rtest: 0.99, RMSEP: 0.177%), respectively. PRACTICAL APPLICATION: This research introduces a practical tool aimed at determining the quality of food by identifying specific light wavelengths. However, it is important to acknowledge potential challenges, such as overfitting. Before implementation, it is crucial for further research to address and mitigate the issues to ensure the reliability and accuracy of the solution. If successfully applied, this tool could significantly enhance the accuracy of near-infrared spectroscopy in assessing food quality attributes. This advancement would provide invaluable support for decision-making in industries involved in food production, ultimately leading to better overall product quality for consumers.
近红外光谱（NIRS）是一种分析技术，由于其运营成本低、分析速度快和非破坏性采样技术，相对于实验室分析在食品工业中具有真正的优势。许多研究表明，NIR 光谱分析对于通过正确的校准来评估某些食品特性具有相关性。这使其在质量控制和食品加工的连续监测中非常有用。然而，NIR 校准过程困难且耗时。分析方法和技术根据 NIR 仪器的配置、要分析的样品和要预测的属性而有所不同。这使得许多制造商的校准具有挑战性。本文旨在提供一种基于智能选择 NIRS 波长的决策支持工具的开发数据驱动方法，用于评估各种食品特性。基于该方法的决策支持工具已在可可豆、小麦粒和芒果样本上进行了评估。对于每个选定模型的水分和脂肪含量的可可豆（Rcv：0.90、Rtest：0.93、RMSEP：0.354%；Rcv：0.73、Rtest：0.79、RMSEP：0.913%）、芒果的酸度和维生素 C 含量（Rcv：0.93、Rtest：0.97、RMSEP：17.40%；Rcv：0.66、Rtest：0.46、RMSEP：0.848%）和小麦-DS2 的蛋白质含量（Rcv：0.90、Rtest：0.92、RMSEP：0.490%），都得到了有希望的结果。此外，所提出的方法可以获得优于小麦-DS1 的水分和蛋白质含量基准的结果（Rcv：0.90、Rtest：94、RMSEP：0.337%；Rcv：0.99、Rtest：0.99、RMSEP：0.177%）。
本研究引入了一种实用工具，旨在通过识别特定的光波长来确定食品的质量。然而，需要认识到潜在的挑战，例如过拟合。在实施之前，进一步研究解决和减轻这些问题对于确保解决方案的可靠性和准确性至关重要。如果成功应用，该工具可以显著提高近红外光谱法在评估食品质量属性方面的准确性。这一进展将为食品生产行业提供宝贵的决策支持，最终为消费者带来更好的整体产品质量。
