Key Points
- The study, published in Food Chemistry, demonstrates that NIR spectroscopy combined with machine learning (ML) can effectively and rapidly determine flavonoid and protein content in Tartary buckwheat.
- Researchers tested 60 buckwheat seed samples using various ML models and found that SVR-based models outperformed others in predicting flavonoid and protein content due to their ability to handle complex, non-linear spectral data.
- This method could significantly benefit food producers and regulators by enabling real-time, non-destructive quality assessments with portable NIR devices.
A recent study explored a new method for determining the total flavonoid and protein content in Tartary buckwheat. This study, which was published in the journal Food Chemistry, showcases the utility of near-infrared (NIR) spectroscopy in testing food quality and detecting food adulteration (1). As the study shows, when NIR spectroscopy is combined with machine learning (ML), this method offers a promising alternative to traditional methods for ensuring the quality of buckwheat products.
Why is buckwheat valued by consumers?
Buckwheat is valued by consumers because of its nutritional components, making it a healthier ingredient in several key food products (2). In particular, it is valued for its high flavonoid content and protein density, making it a sought-after functional food (1,2). Tartary buckwheat is the most common type of buckwheat that consumers like. However, differentiating it from common buckwheat and accurately assessing its nutritional content has historically been a time-consuming and costly process (1).
The researchers explored this challenge in their study. To do so, they used three ML models, including partial least squares regression (PLSR), support vector regression (SVR), and backpropagation neural networks (BPNN), and applied them to the NIR spectral data obtained from both Tartary and common buckwheat samples (1).
What was the experimental procedure?
As part of the experimental procedure, the researchers first analyzed 30 samples of Tartary buckwheat seeds and 30 samples of common buckwheat seeds, all sourced from Zhengzhou Duofu Co., Ltd., a recognized supplier in the region (1). Each sample underwent a meticulous preparation process, including cleaning, grinding, drying, and cold storage, to ensure consistency in spectral measurements (1). Using the NIR1700 spectrometer from Ideaoptics in Shanghai, China, the researchers captured spectral data across a wavelength range of 900 to 1700 nm, employing a resolution of 7.8 nm and 32 scans per measurement to ensure high reliability (1).
What was the model development process like?
The model development process was extensive. During model development, the researchers used three parameter optimization algorithms, nine spectral preprocessing methods, and two feature selection techniques. Out of the models tested, the researchers identified the RAW-SPA-CV-SVR model as the best performer for flavonoid prediction, achieving a coefficient of determination (R²p) of 0.9811 and a root mean squared error of prediction (RMSEP) of 0.1071 (1). For protein content prediction, the top-performing model was MMN-SPA-PSO-SVR, which yielded an R²p of 0.9247 and RMSEP of 0.3906 (1).
Consistently, the SVR-based models performed better than the PLSR and BPNN ones. The researchers theorized that this was because it handled complex spectral data analysis better and was able to model non-linear relationships. This is a key advantage when evaluating overlapping spectral bands inherent in biological samples like buckwheat (1).
What is the impact of this study?
By effectively distinguishing between Tartary and common buckwheat and quantifying nutritional content in a fast, non-destructive manner, this method, presented by the research team could offer food producers and regulators a vital tool for authenticating food products and maintaining quality standards.
Such advances could be especially useful in supply chains where food fraud and adulteration threaten product integrity and consumer trust (1). The researchers showed that by integrating machine learning with portable NIR devices, producers and inspectors could perform real-time quality assessments with minimal sample preparation (1).
However, this study is not a panacea for this issue. The researchers acknowledged in their study that more work is needed to confirm the effectiveness of their model. They suggested that by enhancing the representativeness of training samples, optimizing preprocessing algorithms, and improving the robustness of machine learning models under variable conditions, their method could be improved (1).
The researchers also recommend that future work should examine developing field-deployable NIR instruments embedded with trained SVR models (1). Such systems would bring the power of laboratory analysis to the field, enabling real-time decisions that could benefit industries from agriculture to food safety inspection (1).
As consumer demand for food transparency grows, tools like these offer a promising solution for meeting both nutritional labeling standards and regulatory compliance.
References
- Yu, Y.; Chai, Y.; Li, Z.; et al. Quantitative Predictions of Protein and Total Flavonoids Content in Tartary and Common Buckwheat Using Near-infrared Spectroscopy and Chemometrics. Food Chem. 2025, 462, 141033. DOI: 10.1016/j.foodchem.2024.141033
- Luthar, Z.; Golob, A.; Germ, M. et al. Tartary Buckwheat in Human Nutrition. Plants (Basel). 2021, 10 (4), 700. DOI: 10.3390/plants10040700
