Application of a hybrid variable selection method for the classification of rapeseed oils based on 1H NMR spectral analysis

¹H nuclear magnetic resonance (NMR) spectroscopy was utilized to distinguish the brands of rapeseed oils. As there are more than four hundreds of NMR variables which can cause the discrimination model redundancy, it is necessary to do effective variable selection. Successive projections algorithm (SPA) executed on the full spectrum only improved a few correct answer rate (CAR) and Cohen’s kappa coefficient (K) compared to full spectrum-least-square support vector machine (LS-SVM) model. The better results of uninformative variable elimination (UVE)-based SPA calculation show that it is necessary to do UVE before SPA. Because the cutoff threshold selection in UVE algorithm using an artificial random noise cannot obtain the optimal results, we applied simulated annealing (SA) algorithm to estimate the optimal cutoff threshold. The discrimination results show that UVE-SA did better works than conventional UVE. Only 13 variables were obtained by UVE-SA-SPA while the conventional UVE-based SPA selected 77 variables. The best 97.5% CAR and K of 0.967 result of UVE-SPA-LS-SVM model show that it is feasible to distinguish different brands of rapeseed oils using ¹H NMR spectra. It shows that a combination of SA, UVE, and SPA is effective method for the classification of rapeseed oils. Final result shows that all acyl chains, linolenyl and linoleyl chains, and triglycerides were most important for the classification.     

基于多源光谱分析技术的鱼油品牌判别方法研究

多源光谱分析技术被用于鱼油品牌快速无损鉴别。采用可见光谱分析技术、短波近红外光谱分析技术、长波近红外光谱分析技术、中红外光谱分析技术和核磁共振光谱分析技术采集了7种不同品牌的鱼油的光谱特征,并应用偏最小二乘判别分析法(partial least squares discrimination analysis,PLS-DA)和最小二乘支持向量机(least-squares support vector machine,LS-SVM)建立判别模型并比较判别结果。基于长波近红外光谱的PLS-DA模型和LS-SVM模型取得了最高识别正确率,建模集和预测集识别正确率均达到100%。采用中红外光谱和核磁共振谱分别建立的LS-SVM模型,也可以获得100%的判别正确率。而可见光谱和短波近红外光谱则判别准确率较差。且LS-SVM算法较PLS-DA更加适合用于建立光谱数据和鱼油品牌之间的判别模型。研究结果表面长波近红外光谱技术能够有效判别不同鱼油的品牌,为将来鱼油品质鉴定便携式仪器的开发提供了技术支持和理论依据。     

Discrimination of Producing Areas of Auricularia auricula Using Visible/Near Infrared Spectroscopy

Visible and near infrared (Vis/NIR) spectroscopy combined with chemometric methods was applied for the discrimination of producing areas of Auricularia auricula. Four major varieties of commercial A. auricula were prepared for spectral acquisition. Some pretreatments were performed, such as Savitzky–Golay smoothing, standard normal variate, and the first and second Savitzky–Golay derivative. The scores of the top four latent variables, extracted by partial least squares, were considered as the inputs of back propagation neural network (BPNN) and least squares-support vector machine (LS-SVM). The performance was validated by 60 validation samples. The excellent recognition ratio was 98.3% by BPNN and 96.7% by LS-SVM model with the threshold prediction error ±0.1. The results indicated that Vis/NIR spectroscopy could be used as a rapid and high-precision method for the discrimination of different producing areas of A. auricula by both BPNN and LS-SVM methods.     

External Shape Differences between Sympatric Populations of Commercial Clams <Emphasis Type="Italic">Tapes decussatus</Emphasis> and <Emphasis Type="Italic">T. philippinarum</Emphasis>

The actual Italian production of clams is chiefly sustained by the native Tapes decussatus and the fortuitously imported Tapes philippinarum. Both species are commercialized as “Vongola verace”, but the commercial value of T. philippinarum is lower. The discrimination of species by sight is usually difficult and it cannot be done by observation based on shell morphology but only when animals open their valves hence displaying the two siphons. In this study, we propose a new, noninvasive method to discriminate individuals of both species based on the analysis of the external shape of their shells. Accordingly, in sympatric populations at two sites of the Po river outlet, we have chosen individuals (63 for T. decussatus and 57 for T. philippinarum) of comparable commercial size for which a certain genetic discrimination was previously done. Pictures of the left side valve were taken for all specimens. Their profiles were analyzed with the elliptic Fourier analysis. The mean outline for each species was graphically extracted. The coefficients of the harmonic equations were analyzed by multivariate classification (partial least squares discriminant analysis [PLSDA]). Results showed a high percentage of correct classification of individuals of both species (96.6%). Contour analysis reflected the overall shell shape and thus identified morphological aspects that were difficult to recognize and quantify in sight. The high percentage of correct classifications obtained by combining the analysis of elliptic Fourier harmonics with PLSDA demonstrated the feasibility of this method to discriminate species with a high level of resemblance.     

花生冻伤近红外光谱快速判别方法研究

目的 构建一个基于近红外光谱的花生冻伤判别模型。方法 采用移动窗口平均平滑(Moving Window Average, WMA)、标准正态变量校正(Standard Normal Variate Correction, SNV)及一阶导数(First Derivative, FD)的组合预处理方法提升光谱信号质量;分别采用无信息变量消除法(Elimination of Uninformative Variables, UVE)、竞争性自适应重加权法(Competitive Adaptive Reweighted Sampling, CARS)以及二者的联合算法(CARS-UVE、UVE-CARS)筛选特征波长;最后构建基于支持向量机分类算法(Support Vector Machine Classification, SVC)的花生冻伤分类模型。结果 使用UVE-CARS算法筛选特征波长效果最佳,筛选出7个特征波长,构建的判别模型准确率达95%。结论 该花生冻伤判别模型为花生冻伤快速、无损判别提供可行的技术方案,并为基于滤光片式近红外技术的花生品质色选机的开发提供参考。     

Using Wavelet Textural Features of Visible and Near Infrared Hyperspectral Image to Differentiate Between Fresh and Frozen–Thawed Pork

In this study, wavelet textural analysis was applied to hyperspectral images in the visible and near-infrared (VIS/NIR) region (400–1,000 nm) for differentiation between fresh and frozen–thawed pork. The spectral data of acquired hyperspectral images were analyzed using partial least squares (PLS) regression and five wavelengths (462, 488, 611, 629, and 678 nm) were selected as the feature wavelengths by the regression coefficients from the PLS model. The fourth-order daubechies wavelet (“db4”) was used to serve as the wavelet mother function for wavelet textural extraction of the feature images at the above selected feature wavelengths with the wavelet decomposition level from 1 to 4. Four textural features were calculated in the horizontal, vertical, and diagonal orientations at each level. Forty-eight textural features were extracted from each feature image and used to differentiate between fresh and frozen–thawed pork samples by least-squares support vector machine (LS-SVM) model. Wavelet texture extracted from all five feature images at first decomposition level was identified as optimal wavelet texture combination, resulting in the highest classification accuracy for the LS-SVM models (98.48 % for the training set and 93.18 % for the testing set). Based on the texture combination, the quality attributes of pork meat could be predicted with correlation coefficients of calibration (r _c ) of 0.982 and 0.913, and correlation coefficients of prediction (r _p ) of 0.845 and 0.711 for pH and thawing loss, respectively. The results showed the possibility of developing a fast and reliable hyperspectral system for discrimination between fresh and frozen–thawed pork samples based on wavelet texture in the VIS/NIR wavelength range.     

基于高光谱成像技术的金银花与山银花快速鉴别

利用高光谱成像技术,研究一种快速、准确、无损的鉴别金银花与山银花的方法。通过对比3种预处理方法对偏最小二乘算法(Partial Least Squares,PLS)建模效果的影响,得到SNV为建模最优预处理方法。使用回归系数法(Regression Coefficient,RC)和连续投影算法(Successive Projection Algorithm,SPA)选择经预处理后光谱的特征波长,并分别建立极限学习机(Extreme learning machine,ELM)和最小二乘支持向量机(Last Squares Support Vector Machine,LSSVM)的判别分析模型。结果表明,光谱经SNV预处理后,应用SPA提取特征波长并建立LS-SVM判别分析模型为金银花和山银花最优判别模型,其建模集与预测集识别率均达到了100.00%。因此,利用高光谱成像技术能够无损、有效地鉴别金银花与山银花,并且在全光谱和特征波长下均能实现金银花与山银花的快速判别分析。     

Rapid recognition of marine fish surimi by one-step discriminant analysis based on near-infrared diffuse reflectance spectroscopy

A rapid and nondestructive near-infrared diffuse reflectance spectroscopy method combined with multivariate analysis was developed to discriminate different species and grades of marine fish surimi. Principal component analysis and discriminant analysis were applied to classify the species and quality grades of surimi. The results showed that excellent classification was obtained after optimizing spectral pretreatment. For the discrimination of three species of surimi, the correct classification rate of the calibration as well as the validation datasets were 98.5% and 100%, respectively, using the discriminant analysis method after multiplicative scatter correction pretreatment. For four grades of surimi, discriminant analysis provided 98.9% and 100% correct classification rate for calibration and validation datasets, respectively, after multiplicative scatter correction pretreatment. It was demonstrated that near-infrared diffuse reflectance spectroscopy integrated with discriminant analysis had significant potential as a rapid and accurate approach for rapid discrimination of surimi species and grades.     

Discrimination of Bacillus cereus from Selected Bacillus spp. by Pyrolysis Gas-Liquid Chromatography

Pyrolysis gas-liquid chromatography (PGLC) and statistical analysis employing stepwise discriminant analysis (SDA), were used to classify five species and varieties in the genus Bacillus. Nine strains were harvested, lyophilized, and pyrolyzed under replicate conditions. The combination of PGLC and SDA enabled classification of these selected varieties and species of Bacillus at 96% accuracy. Stepwise discriminant analysis of the same PGLC data discriminated (100%) B. cereus, a foodborne pathogen, from the remaining selected nonpathogenic bacilli when the two groups were compared. One particular elution peak resulting from the PGLC-SDA analysis was prominent in both discriminations reported. The PGLC and SDA method offers potential as an accurate, objective procedure for the discrimination of these selected bacilli.     

遗传算法联合LS-SVM的苹果原醋成分定量分析

利用近红外光谱技术对苹果原醋中的重要指标进行定量分析,并进行模型优化以提高性能。采用遗传偏最小二乘法（GA-PLS）提取的特征波长作为最小二乘支持向量机（LS-SVM）的输入变量,先后建立苹果原醋中总酸、可溶性固形物的近红外定量模型,并与建立的偏最小二乘（PLS）模型结果进行比较。用决定系数（R2）、预测均方根误差（RMSEP）以及相对分析误差（RPD）对模型进行评价,确定最佳建模方法。结果表明,相比于PLS模型,总酸及可溶性固形物指标的LS-SVM定量模型的R2、RMSEP以及RPD值均有更好的表现,且在进行独立测试集验证时,LS-SVM模型的预测精度也明显优于PLS模型。说明遗传算法联合LS-SVM建立的定量模型有很高的准确度及稳定性,可以应用于苹果原醋总酸和可溶性固形物含量的快速检测。     

Detection of Organic Acids and pH of Fruit Vinegars Using Near-Infrared Spectroscopy and Multivariate Calibration

Near-infrared (NIR) spectroscopy was investigated to determine the acetic, tartaric, formic acids and pH of fruit vinegars. Optimal partial least squares (PLS) models were developed with different preprocessing. Simultaneously, the performance of least squares-support vector machine (LS-SVM) models was compared with three kinds of inputs, including wavelet transform (WT), latent variables, and effective wavelengths (EWs). The results indicated that all LS-SVM models outperformed PLS models. The optimal correlation coefficient (r), root mean square error of prediction and bias for validation set were 0.9997, 0.3534, and −0.0110 for acetic acid by WT-LS-SVM; 0.9985, 0.1906, and 0.0025 for tartaric acid by WT-LS-SVM; 0.9987, 0.1734, and 0.0012 for formic acid by EW-LS-SVM; and 0.9996, 0.0842, and 0.0012 for pH by WT-LS-SVM, respectively. The results indicated that NIR spectroscopy (7,800–4,000 cm⁻¹) combined with LS-SVM could be utilized as a precision method for the determination of organic acids and pH of fruit vinegars.     

A comparative study for the quantitative determination of soluble solids content,pH and firmness of pears by Vis/NIR spectroscopy

Visible and near infrared (Vis/NIR) spectroscopy was investigated to determine the soluble solids content (SSC), pH and firmness of different varieties of pears. Two-hundred forty samples (80 for each variety) were selected as sample set. Two-hundred ten pear samples (70 for each variety) were selected randomly for the calibration set, and the remaining 30 samples (10 for each variety) for the validation set. Partial least squares (PLS) and least squares-support vector machine (LS-SVM) with different spectral preprocessing techniques were implemented for calibration models. Different wavelength regions including Vis, NIR and Vis/NIR were compared. It indicated that Vis/NIR (400–1800 nm) was optimal for PLS and LS-SVM models. Then, LS-SVM models were developed with a grid search technique and RBF kernel function. All LS-SVM models outperformed PLS models. Next, effective wavelengths (EWs) were selected according to regression coefficients. The EW-LS-SVM models were developed and a good prediction precision and stability was achieved compared with PLS and LV-LS-SVM models. The correlation coefficient of prediction (r_p), root mean square error of prediction (RMSEP) and bias for the best prediction by EW-LS-SVM were 0.9164, 0.2506 and −0.0476 for SSC, 0.8809, 0.0579 and −0.0025 for pH, whereas 0.8912, 0.6247 and −0.2713 for firmness, respectively. The overall results indicated that the regression coefficient was an effective way for the selection of effective wavelengths. LS-SVM was superior to the conventional linear PLS method in predicting SSC, pH and firmness in pears. Therefore, non-linear models may be a better alternative to monitor internal quality of fruits. And the EW-LS-SVM could be very helpful for development of portable instrument or real-time monitoring of the quality of pears.     

Application of Competitive Adaptive Reweighted Sampling Method to Determine Effective Wavelengths for Prediction of Total Acid of Vinegar

In this paper, near-infrared (NIR) spectroscopy coupled with wavelength selection methods was used to predict total acid of vinegar. Three wavelength selection methods including competitive adaptive reweighted sampling (CARS), Monte Carlo uninformative variable elimination (MC-UVE), and moving window partial least squares (MWPLS) were employed to select the key wavelengths. Five wavelengths including 4,348, 4,694, 5,365, 7,104, and 7,236 cm⁻¹ were selected by CARS method. Least squares (LS) regression model was built on the selected wavelengths. Compared to the partial least squares regression models based on full spectrum and wavelengths selected by MC-UVE and MWPLS, the performance of LS model was better, with higher determination coefficient for test (r ²) of 0.997, and lower root mean square error of prediction of 0.13 g/100 ml. Based on the results, it was concluded that NIR spectroscopy combined with CARS methods seem to be a rapid and effective alternative to the classical methods for the prediction of total acid of vinegar.     

A Non-invasive Method for Screening Sodium Hydroxymethanesulfonate in Wheat Flour by Near-Infrared Spectroscopy

A rapid and non-invasive method, based on near infrared diffuse reflectance spectroscopy, was established for screening sodium hydroxymethanesulfonate in wheat flour. Successive projection algorithm was used for spectral variable selection. The selected variables were applied as inputs to partial least square discriminant analysis (PLS-DA) and advanced K-means dynamic clustering. The first two principal components extracted by PLS-DA had been applied as inputs to least squares support vector machine (LS-SVM). Three algorithms, including PLS-DA, advanced K-means dynamic clustering, and LS-SVM, were used to establish the calibration model. The results of LS-SVM outperformed that of the other two methods, with the classification accuracy of 92.0% for the validation and 94.7% for the prediction. The results of the study showed the potential of near-infrared spectroscopy as a non-invasive and environmentally acceptable method for the screening of sodium hydroxymethanesulfonate in wheat flour.     

CLASSIFICATION OF CHINESE RICE WINE WITH DIFFERENT MARKED AGES BASED ON NEAR INFRARED SPECTROSCOPY

The feasibility of near infrared (NIR) spectroscopy for discrimination between Chinese rice wine with different marked ages (1, 3 and 5 years) was presented in this research. NIR spectra were collected in transmission mode in the wavelength range of 800–2500 nm. Discriminant models were developed based on discriminant analysis (DA) together with raw and second derivative spectra. The calibration result for raw spectra was better than that for second derivative spectra. The percentage of samples correctly classified for raw and second derivative spectra was 100 and 78.4%, respectively. In validation analysis, the percentage of samples correctly classified was 94.4%. For 1‐, 3‐ and 5‐year‐old sample groups, the percentage of samples correctly classified was 88.3, 100 and 100%, respectively. The results demonstrated that NIR spectroscopy could be used as a rapid and reliable method for classification of Chinese rice wine with different marked ages.     

Detection of adulterants such as sweeteners materials in honey using near-infrared spectroscopy and chemometrics

Near-infrared (NIR) spectroscopy combined with chemometrics methods has been used to detect adulteration of honey samples. The sample set contained 135 spectra of authentic (n = 68) and adulterated (n = 67) honey samples. Spectral data were compressed using wavelet transformation (WT) and principal component analysis (PCA), respectively. In this paper, five classification modeling methods including least square support vector machine (LS-SVM), support vector machine (SVM), back propagation artificial neural network (BP-ANN), linear discriminant analysis (LDA), and K-nearest neighbors (KNN) were adopted to correctly classify pure and adulterated honey samples. WT proved more effective than PCA, as a means for variables selection. Best classification models were achieved with LS-SVM. A total accuracy of 95.1% and the area under the receiver operating characteristic curves (AUC) of 0.952 for test set were obtained by LS-SVM. The results showed that WT-LS-SVM can be as a rapid screening technique for detection of this type of honey adulteration with good accuracy and better generalization.     

Identification of seven species of the Lactobacillus acidophilus group by FT-IR spectroscopy

Fourier transform infrared (FT-IR) spectra of 102 strains of the seven species of the Lactobacillus acidophilus group were collected and investigated for their potential use in classification and identification on species level. The database built contains more than 370 spectra. Various procedures of pre-processing and classification methods have been compared with respect to their predictive ability. The most encouraging results were achieved with linear discriminant analysis (LDA) of the absorbance values of normalized spectra at selected wavenumbers. The rate of correct species assignment in cross-validation (Jackknife procedure with one spectrum left out for model building) were 95%, 95%, 69%, 100%, 88%, 100%, and 91% for L. acidophilus, L. amylovorus, L. crispatus, L. gallinarum, L. gasseri, L. helveticus, and L. johnsonii, respectively. Very distinct grouping was found for L. gallinarum and L. helveticus, the most difficult differentiation in LDA was between the pairs L. crispatus/L. amylovorus and L. gasseri/L. johnsonii.     

Southern Spain lamb types discrimination by using visible spectroscopy and basic physicochemical traits

The potential for using visible spectroscopy (400–700 nm) to classify six types (breed × production system) of lamb meat was investigated. Seven wavelengths namely 400, 410, 420, 450, 510, 610 and 670 nm were retained for the discriminant analysis. The basic meat physicochemical traits of Longissimus dorsi were also studied and a model including that information together with the spectra was developed to compare both accuracies. Then, Myoglobin content, water holding capacity, pH, a^*, 670 and 610 nm wavelengths, protein percentage, L^*, ash content, 450 and 420 nm wavelengths and moisture percentage were selected as variables for the development of the discriminant function. The data analysis showed that it was possible to discriminate the lamb types with accuracy around 83% using visible spectroscopy. However these results improved to 95% when using the reflectance together with basic physicochemical traits (12% better than using only the spectra).     

基于高光谱成像技术的工夫红茶数字化拼配

以祁门红茶5?级6?孔正子口、5?级8?孔正子口、6?级6?孔正子口和6?级8?孔正子口4?种原料拼配成工夫红茶,应用高光谱图像系统获取拼配后茶样的光谱和图像信息。采用连续投影算法筛选光谱特征值;通过对图像做主成分分析,提取5?个特征波长,采用灰度共生矩阵法提取5?个特征波长下的图像纹理特征值。分别以光谱特征值、纹理特征值以及融合特征值作为模型输入值,结合偏最小二乘、最小二乘支持向量机（least squares-support vector machine,LS-SVM）和反向传播人工神经网络方法建立茶叶拼配配比定量预测模型,并对模型的结果做比较。结果表明,以光谱特征值与纹理特征值融合后的值为输入参数,结合LS-SVM方法建立的模型,配比预测正确率达到了94.5%,预测结果较好。研究结果为出口茶叶数字化拼配的可行性提供理论依据。     

Feasibility of Discriminating Dried Dairy Ingredients and Preheat Treatments Using Mid-Infrared and Raman Spectroscopy

This study investigated the feasibility of mid-infrared (MIR) and Raman spectroscopy for (i) discrimination of three dried dairy ingredients, namely skim milk powder (SMP), whey protein concentrate (WPC) and demineralised whey protein (DWP) powder, and (ii) discrimination of preheat treatments of dried dairy ingredients using partial least squares discriminant analysis (PLS-DA). PLS1-DA models developed using MIR ranges of 800–1800 and 1200–1800 cm^?1 yielded the best discrimination (correct identification of 97.2% for SMP discrimination and 100% for WPC and DWP discrimination). The best PLS2-DA model using MIR spectroscopy was developed over the spectral range of 800–1800 cm^?1 and produced correct identification of 100% for dairy ingredient discrimination. Models developed using Raman 800–1800 and 1200–1800 cm^?1 spectral ranges correctly discriminated (100% correctly identified) each dairy ingredient. Although all PLS1-DA and PLS2-DA models developed using both spectral technologies for preheat treatment discrimination had good discrimination accuracy (86–100%), they employed a high number of factors (8–9 for the best model). The use of the Martens uncertainty test successfully reduced the number of factors employed (3–4 for the best models) and improved the performance of PLS1-DA models for preheat treatment discrimination (all 100% correctly identified). This feasibility study demonstrates the potential of both MIR and Raman spectroscopy for rapid characterisation of dried dairy ingredients.