Food Analytical Methods最新文献

The adulteration of high-value vegetable oils in blended products poses significant challenges for quality control and consumer protection. While Raman spectroscopy offers a rapid and non-destructive analytical tool, conventional chemometric models such as partial least squares (PLS) and support vector machines (SVM) often struggle with complex spectral data due to their reliance on manual feature selection and limited capability in capturing nonlinear relationships. To address these limitations, this study introduces a deep learning-based approach combining Raman spectroscopy with three advanced neural network architectures—1D-CNN, ConvNext-ECA, and CNN-GRU-MHA—for the quantitative determination of camellia oil in ternary blends with rapeseed and corn oils. Compared to traditional machine learning models, all three deep learning models demonstrated superior predictive accuracy. The CNN-GRU-MHA model achieved the best performance, with an R²p of 0.9981 and RMSEP of 0.3714. These results underscore the potential of attention-enhanced deep learning models as a robust and efficient tool for the authentication of blended vegetable oils.

Oxidative deterioration of edible oils leads to rancidity, resulting in the formation of volatile and non-volatile degradation products that adversely affect sensory quality, nutritional value, and consumer acceptance. Conventional methods for rancidity detection, such as peroxide value, p-anisidine value, free fatty acid value, thiobarbituric acid test, chromatography, and spectroscopic techniques, are accurate but time-consuming, costly, and laboratory-based, making them unsuitable for rapid detection. This review comprehensively analyses the potential of electronic nose (e-nose) technologies for rapid detection of rancidity in various edible oils. An e-nose system typically comprises a sample transfer unit, a sensor array (e.g. metal oxide or polymer-based gas sensors) for detecting volatile oxidation products, and a pattern recognition or data analysis system. Recent developments integrating e-nose systems with artificial intelligence, machine learning, computer vision, and hybrid analytical techniques are critically discussed, demonstrating improved accuracy, sensitivity, and practical applicability. Overall, this review bridges the gap between conventional and advanced analytical approaches by highlighting e-nose technologies as cost-effective, user-friendly, and scalable tools for routine quality assessment. The e-nose thus holds great promise for applications in oil processing industries, retail outlets, restaurants, and even household kitchens, contributing to improved food safety, reduced waste, and informed consumer choices.

Graphical Abstract

Traditional “gold standard” methods for quantifying arsenic (As) in rice are not ideal for routine analysis due to their time-consuming, expensive, and complex nature. Therefore, the aim of this study was to develop a simple, rapid, and minimally destructive method applying X-ray fluorescence (XRF) spectroscopy and chemometric modeling to quantify As in rice. Milled rice was spiked with As (63.47 – 553.43 μg kg⁻¹), pelletized, and analyzed utilizing an energy-dispersive XRF spectrometer. The resultant spectra were used to generate a prediction model via partial least squares regression (PLSR), which showed strong predictive capabilities (R² = 0.99, RMSEC = 14.3 μg kg⁻¹) and strong sensitivity (LOD = 27.76 μg kg⁻¹, LOQ = 92.52 μg kg⁻¹). Validation was conducted using a certified reference material, yielding an error of prediction of only 8.96%. Analysis of rice and rice-based foods showed strong agreement between the current study and traditional methods, demonstrating its robust capabilities for routine analysis.

‘Sous vide’ is a cooking process that uses long time and low temperatures, affecting the quality and composition of the product. This study aimed to develop and validate prediction models for the proximate composition of sous vide beef tenderloin using near-infrared reflectance spectroscopy (NIRS) and to compare the performance of two portable NIR spectrometers. Original spectra were pretreated using the Standard Normal Variate (SNV) method, followed by the 1st derivative. Models using partial least squares regression (PLS) were constructed with all spectral variables, and after the jack-knife algorithm performed wavelength selection. The centesimal composition of the ground samples was accurately determined by the models, except for the carbohydrate content. The prediction errors resulting from external validation (RMSEP) for the InnoSpectra and NeoSpectra evaluated compact instruments were, respectively, 0.92% and 0.65% for moisture, 0.72% and 0.93% for fat, 0.93% and 0.60% for protein, and 0.12% and 0.03% for ash. Conversely, poor results were obtained for samples where readings were taken from whole roasters with or without packaging. A new method for assessing the quality and usefulness of multivariate models was proposed based on RMSEP and comparing the distributions of the reference and validation data. The NIRS-based method is fast, requires simple sample preparation, does not require the use of chemicals, and employs low-cost instruments.

Conventional laboratory testing of moisture content and water activity in honey is often time-consuming and costly. Therefore, alternative analytical methods that are non-destructive, rapid, efficient, and cost-effective are needed, particularly for field applications. This study aims to develop predictive models for determining moisture content and water activity in liquid honey (forest and acacia) and powdered honey using portable near-infrared (NIR) spectroscopy combined with multiple linear regression (MLR) analysis Spectral data were acquired using a NIR SpectraPod (850–1700 nm), followed by laboratory measurements for calibration. Data pre-processing with Standard Normal Variate (SNV) and Multiplicative Scatter Correction (MSC) improved model accuracy. A simple K-nearest neighbors (KNN) analysis was included only as a preliminary check of honey-type separability. The developed models demonstrated strong predictive performance across all honey types. For individual honey varieties, the coefficient of determination (R²) ranged from 0.75 to 0.96, while the root mean square error (RMSE) values were between 0.02 and 0.05 for water activity and 2.1–4.6% for moisture content. When acacia and forest honey data were combined, the models maintained good performance, with R² values of 0.58–0.78 and RMSE values of 0.03–0.05 for water activity and 4.0–6.7% for moisture content. These findings demonstrate that portable NIR spectroscopy, integrated with regression modelling, is a reliable and efficient tool for rapid honey quality assessment, offering significant advantages over conventional laboratory methods.

Pine nuts are highly valued for their nutritional content but are prone to mold contamination during transportation and storage, potentially leading to the production of carcinogenic aflatoxins.To address this issue, we propose a rapid, non-destructive mold detection method using hyperspectral imaging combined with a lightweight three-dimensional convolutional neural network (Light3DCNN). A novel differentiable band-selection layer—feature selection (FS)—is integrated directly into the network’s training pipeline. Leveraging the Gumbel-Softmax relaxation technique and a straight-through estimator, FS enables gradient back-propagation through discrete spectral-band selections.This end-to-end learning mechanism dynamically selects the ten most informative bands by aligning the selection strategy with the final classification objective. In parallel, a lightweight feature-extraction unit combines deep convolution with point-wise convolution to independently capture spectral signatures and fuse multi-channel information with minimal computational overhead. This research compares the performance of the feature-selection-based FS-Light3DCNN model with other models. While FS-Light3DCNN achieves 99.34% accuracy—slightly lower than the full-band Light3DCNN (99.69%)—it significantly reduces training parameters and processing time by nearly 70%. FS-Light3DCNN outperforms models using SVM with UVE and CARS feature-selection algorithms, GLCM texture features, and traditional 1DCNN, 2DCNN, HybridSN, and 3DCNN models. Additionally, compared to PCA-Light3DCNN and RF-Light3DCNN, which use ten key bands selected by PCA and RF, FS-Light3DCNN improves accuracy by 6% and 5%, respectively. Experimental results demonstrate that FS-Light3DCNN excels in both accuracy and efficiency, effectively distinguishing between healthy and varying levels of mold contamination. This model provides a fast, reliable, and non-destructive method for assessing mold contamination in pine nuts and offers potential for broader applications in food quality testing.

The aging process has a significant impact on the quality and market value of white tea (WT), as described in a traditional proverb: “One year’s tea, three years’ medicinal value, and seven years becomes a treasure.” However, the lack of a rapid and reliable method for age identification poses challenges for market regulation. This study proposes a novel approach that combines surface-enhanced Raman spectroscopy (SERS) with machine learning (ML) to accurately identify the age of white tea. SERS is used to obtain molecular fingerprint spectra from Fujian white tea samples covering seven different years (2011−2023). Seven machine learning models, including logistic regression (LR), support vector machine (SVM), and random forest (RF), were systematically evaluated. The LR model, after being preprocessed through principal component analysis—logistic discriminant analysis, demonstrated relatively excellent performance. This SERS-ML method can perform rapid analysis and requires very little samples preparation. Our work establishes a robust, efficient, and field-deployable strategy for identifying the age of white tea, which is of great significance for combating fraud and protecting consumers.

Graphical Abstract

This study employs digital image processing and machine learning techniques to predict all colorimetric characteristics (color intensity, density, tonality, and color index percentages) of colored wines in a novel, cost-effective, and rapid manner. To determine the values of colorimetric characteristics, ultraviolet-visible (UV-Vis) absorbance was measured at three key wavelengths (A420, A520, and A620) using UV-Vis spectrophotometry, corresponding to the yellow, red, and blue color percentages, respectively. Simultaneously, the pictures of 86 wine samples were acquired, and the corresponding RGB and HSV color values were extracted from the images to serve as input features for multiple regression models. The models developed included principal component regression, k-nearest neighbors, linear regression, decision tree, random forest, and partial least squares (PLS). Among the models, random forest outperformed PLS in predicting A620 absorbance value due to its ability to capture non-linear patterns, whereas PLS demonstrated greater accuracy (R² > 0.95) in predicting the A420 and A520 absorbance values. According to feature selection, hue and saturation had the biggest impact on prediction accuracy. By determining absorbance values using the developed models, the complete colorimetric characteristics of the wine samples can be calculated, enabling the evaluation of their physicochemical parameters during the fermentation process or post-fermentation. As a result, all the models, improved, offer a promising alternative for quick, easy, and scalable prediction methods by reducing measurement time, eliminating the need for laboratory instruments, and introducing a new methodology to complement conventional spectroscopic techniques, with potential applications in consumer-level analysis and the process of wine quality control.

Colorimetric sensors for volatile amine (VA) detection makes meat freshness monitoring visible and convenient. However, the sensing performance still needs to be improved due to low VA concentrations in the early stage of meat spoilage. In this study, we proposed the strategy of dye solution pH regulation to improve the sensing performance of the VA colorimetric sensor. Taking methyl red (MR) as the example, the pH of the MR solution was adjusted, and MR was then adsorbed on the aerogel to fabricate the aerogel-type VA colorimetric sensor. By adjusting the pH of the MR solution to 1 prior to dye adsorption, the sensor exhibited optimal sensing performance, with a sensitivity of 3.58 ppm⁻¹ and a maximum ΔE value of 108.90 ± 2.43. Compared with the sensors without pH regulation, the sensitivity and maximum ΔE of the newly developed sensor increased by 70% and 51%, respectively. This strategy showed good generality in various pH-sensitive dye systems. The sensor also successfully applied to the real-time monitoring of shrimp. A ΔE value higher than 62.16 indicated the spoilage of the shrimp. The enhanced performance was also validated in the freshness monitoring of other types of meat. The findings provided a facile and effective approach for optimizing colorimetric sensing performance in meat and aquatic products freshness monitoring.

Foodborne pathogens, such as Escherichia coli O157:H7, pose a significant global health threat, necessitating the development of rapid and decentralized detection methods. Conventional laboratory assays often require extensive sample processing and sophisticated instrumentation, which limits their use for on-site, point-of-care testing. This work addresses the need for a simplified and accelerated platform. A rapid bi-functional linker-based immunoassay for the detection of Escherichia coli O157:H7 was developed by accelerating gold nanoparticle (AuNP) aggregation and sedimentation. The bi-functional linker facilitated antigen–antibody binding and controlled AuNP aggregation, while a 1-min centrifugation step expedited sedimentation, significantly shortening the overall assay time. Under these conditions, the assay produced an instrument-free, eye-readable visual response based on the range exhibiting a visible color change (REVC). A portable hand-powered centrifuge was fabricated and applied to detect E. coli O157:H7 in spiked tomato samples, achieving a decision level of 10³ CFU/25 g within a total assay time of just 70 min. Compared with passive settling (≈180 min), the proposed system significantly reduced detection time while maintaining consistent signal patterns, demonstrating its potential as a rapid, field-deployable biosensor for on-site foodborne pathogen detection.