研究目的
To investigate the combination of infrared (NIR) and mid infrared (MIR) spectroscopy for the quantification of rapeseed oil in olive oil blends and to evaluate three data fusion methods (low, mid, and high-level) for improving the accuracy of quantitative analysis in detecting olive oil adulteration.
研究成果
NIR and MIR spectroscopy are effective, non-destructive tools for detecting rapeseed oil adulteration in olive oil. Low-level and high-level data fusion strategies significantly improved prediction accuracy compared to individual techniques or mid-level fusion, with high-level fusion showing the best performance (R2 validation of 0.988 and RMSEP of 2.86). Mid-level fusion, using SPA for feature extraction, did not enhance results due to loss of useful information. The findings suggest that data fusion, particularly high-level methods, can be reliably used for quantitative analysis in food authenticity testing, but the choice of fusion strategy and feature extraction method is critical for optimal outcomes.
研究不足
The study used a limited number of samples (36 total, with 24 for calibration and 12 for validation) and only two types of oils (rapeseed and olive oil), which may not represent all possible adulteration scenarios. The feature extraction method (SPA) in mid-level fusion did not improve results, indicating dependence on the selection algorithm. High-level fusion required removal of low-concentration data (below 20%) for better performance, suggesting issues with prediction accuracy at lower adulteration levels. The experiments were conducted at room temperature, and variations in environmental conditions were not explored.
1：Experimental Design and Method Selection:
The study used NIR and MIR spectroscopy combined with chemometrics methods, specifically partial least squares (PLS) regression, to quantify rapeseed oil adulteration in olive oil. Three data fusion strategies (low-level, mid-level, and high-level) were applied to integrate information from both spectroscopic techniques. Pretreatments such as baseline correction, standard normal variate (SNV), Savitzky-Golay (SG) smoothing, and vector normalization were evaluated for model optimization.
2：Sample Selection and Data Sources:
Two brands of rapeseed oil and two brands of extra-virgin olive oil were purchased from local supermarkets and mixed to create adulterated samples with rapeseed oil concentrations ranging from 1% to 80% (w/w). One sample was prepared for each concentration.
3：List of Experimental Equipment and Materials:
NIR Spectrometer (PerkinElmer Spectrum One NTS), MIR spectrometer (PerkinElmer Spectrum 100), quartz cuvette, zinc selenide crystal, solvents (acetone, dichloromethane, ethanol, distilled water), and software (QUANTC for NIR, PerkinElmer Spectrum 10 for MIR, Unscramble
4：7 for data pretreatment, Matlab R2014a for PLS analysis). Experimental Procedures and Operational Workflow:
Spectra were acquired for each sample multiple times and averaged. For NIR, spectra were recorded over 4000-10000 cm?1 with 16 cm?1 resolution and 32 scans; for MIR, over 4000-650 cm?1 with 8 cm?1 resolution and 64 scans. Between samples, equipment was cleaned with solvents and air-dried. Background spectra were collected before each measurement. Data were divided into calibration and validation sets with similar statistical properties.
5：Data Analysis Methods:
PLS regression was used for quantification, with the number of factors determined by five-fold cross-validation. Evaluation parameters included R2, RMSEC, RMSECV, and RMSEP. For data fusion, low-level involved concatenating raw NIR and MIR data, mid-level used SPA for feature extraction, and high-level used binary linear regression on predicted values from individual models.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
