Implementation of LOCAL algorithm with near-infrared spectroscopy for compliance assurance in compound feedingstuffs

Optimisation of the predictive ability of NIR models to estimate nutritional parameters in elephant grass through LOCAL algorithms

Elephant grass is a tropical forage widely used for livestock feed. The analytical techniques traditionally used for its nutritional evaluation are costly and time consuming. Alternatively, Near Infrared Spectroscopy (NIRS) technology has been used as a rapid analysis technique. However, in crops with high variability due to genetic improvement, predictive models quickly lose accuracy and must be recalibrated. The use of non-linear models such as LOCAL calibrations could mitigate these issues, although a number of parameters need to be optimized to obtain accurate results. The objective of this work was to compare the predictive results obtained with global NIRS calibrations and with LOCAL calibrations, paying special attention to the configuration parameters of the models. The results obtained showed that the prediction errors with the LOCAL models were between 1.6 and 17.5 % lower. The best results were obtained in most cases with a low number of selected samples (n = 100-250) and a high number of PLS terms (n = 20). This configuration allows a reduced computation time with high accuracy, becoming a valuable alternative for analytical determinations that require ruminal fluid, which would improve the welfare of the animals by avoiding the need to surgically prepare animals to estimate the nutritional value of the feeds.

Fraud Detection in Batches of Sweet Almonds by Portable Near-Infrared Spectral Devices

One of the key challenges for the almond industry is how to detect the presence of bitter almonds in commercial batches of sweet almonds. The main aim of this research is to assess the potential of near-infrared spectroscopy (NIRS) by means of using portable instruments in the industry to detect batches of sweet almonds which have been adulterated with bitter almonds. To achieve this, sweet almonds and non-sweet almonds (bitter almonds and mixtures of sweet almonds with different percentages (from 5% to 20%) of bitter almonds) were analysed using a new generation of portable spectrophotometers. Three strategies (only bitter almonds, bitter almonds and mixtures, and only mixtures) were used to optimise the construction of the non-sweet almond training set. Models developed using partial least squares-discriminant analysis (PLS-DA) correctly classified 86–100% of samples, depending on the instrument used and the strategy followed for constructing the non-sweet almond training set. These results confirm that NIR spectroscopy provides a reliable, accurate method for detecting the presence of bitter almonds in batches of sweet almonds, with up to 5% adulteration levels (lower levels should be tested in future studies), and that this technology can be readily used at the main steps of the production chain.

Reduction of the Number of Samples for Cost-Effective Hyperspectral Grape Quality Predictive Models

Developing chemometric models from near-infrared (NIR) spectra requires the use of a representative calibration set of the entire population. Therefore, generally, the calibration procedure requires a large number of resources. For that reason, there is a great interest in identifying the most spectrally representative samples within a large population set. In this study, principal component and hierarchical clustering analyses have been compared for their ability to provide different representative calibration sets. The calibration sets generated have been used to control the technological maturity of grapes and total phenolic compounds of grape skins in red and white cultivars. Finally, the accuracy and precision of the models obtained with these calibration sets resulted from the application of the selection algorithms studied have been compared with each other and with the whole set of samples using an external validation set. Most of the standard errors of prediction (SEP) in external validation obtained from the reduced data sets were not significantly different from those obtained using the whole data set. Moreover, sample subsets resulting from hierarchical clustering analysis appear to produce slightly better results.

Pre-harvest screening on-vine of spinach quality and safety using NIRS technology

The study sought to perform a non-destructive and in-situ quality evaluation of spinach plants using near infrared (NIR) spectroscopy in order to establish its suitability for different uses once harvested. Modified partial least square (MPLS) regression models using NIR spectra of intact spinach leaves were developed for nitrate, ascorbic acid and soluble solid contents. The residual predictive deviation (RPD) values were 1.29, 1.21 and 2.54 for nitrate, ascorbic acid and soluble solid contents, respectively. Later, this predictive capacity increased for nitrate content (RPDcv = 1.63) when new models were developed, taking into account the influence on the robustness of the model exercised by the simultaneity between the NIR and laboratory analyses. Subsequently, using partial least squares discriminant analysis (PLS-DA), the ability of NIRS technology to classify spinach as a function of nitrate content was tested. PLS-DA yielded percentages of correctly classified samples ranging from 73.08-76.92% for the class 'spinach able to be used fresh' to 85.71-73.08% for the class 'preserved, deep-frozen or frozen spinach, both for unbalanced and balanced models respectively, based on NH signal associated with proteins. Overall, the data supports the capability of NIR spectroscopy to establish the final destination of the production of spinach analysed on the plant, as a screening tool for important safety and quality parameters.

Long-Length Fiber Optic Near-Infrared (NIR) Spectroscopy Probes for On-Line Quality Control of Processed Land Animal Proteins

This research was conducted using a spectral database comprising 346 samples of processed animal proteins (PAPs) with a range of compositions, analyzed using a Fourier transform near-infrared spectroscopy multichannel instrument (Matrix-F, Bruker Optics) coupled to a 100 m fiber optic cable. Using both its static and dynamic operating modes (on a conveyor belt), simulating the movement of the product in the plant, the predictive capabilities of both modes of analysis were assessed and compared, for the purposes of predicting moisture, protein, and ashes. The results show that both exhibit highly similar degrees of precision and accuracy for predicting these parameters. This research provides a foundation of scientific-technical knowledge, hitherto unknown, regarding the "on-line" incorporation of an instrument (equipped with a 100 m fiber optic cable) into a processing plant of by-products of animal origin.

A New Local Modelling Approach Based on Predicted Errors for Near-Infrared Spectral Analysis

Over the last decade, near-infrared spectroscopy, together with the use of chemometrics models, has been widely employed as an analytical tool in several industries. However, most chemical processes or analytes are multivariate and nonlinear in nature. To solve this problem, local errors regression method is presented in order to build an accurate calibration model in this paper, where a calibration subset is selected by a new similarity criterion which takes the full information of spectra, chemical property, and predicted errors. After the selection of calibration subset, the partial least squares regression is applied to build calibration model. The performance of the proposed method is demonstrated through a near-infrared spectroscopy dataset of pharmaceutical tablets. Compared with other local strategies with different similarity criterions, it has been shown that the proposed local errors regression can result in a significant improvement in terms of both prediction ability and calculation speed.

Evaluation of local approaches to obtain accurate near-infrared (NIR) equations for prediction of ingredient composition of compound feeds

This research work investigated new methods to improve the accuracy of intact feed calibrations for the near-infrared (NIR) prediction of the ingredient composition. When NIR reflection spectroscopy, together with linear models, was used for the prediction of the ingredient composition, the results were not always acceptable. Therefore, other methods have been investigated. Three different local methods (comparison analysis using restructured near-infrared and constituent data [CARNAC]), locally weighed regression [LWR], and LOCAL) were applied to a large (N = 20 320) and heterogeneous population of non-milled feed compounds for the NIR prediction of the inclusion percentage of wheat and sunflower meal, as representative of two different classes of ingredients. Compared with partial least-squares regression, results showed considerable reductions of standard error of prediction values for all methods and ingredients: reductions of 59, 47, and 50% with CARNAC, LWR, and LOCAL, respectively, for wheat, and reductions of 49, 45, and 43% with CARNAC, LWR, and LOCAL, respectively, for sunflower meal. These results are a valuable achievement in coping with legislation and manufacture requirements concerning the labeling of intact feedstuffs.

Optimization of NIR spectral data management for quality control of grape bunches during on-vine ripening

NIR spectroscopy was used as a non-destructive technique for the assessment of chemical changes in the main internal quality properties of wine grapes (Vitis vinifera L.) during on-vine ripening and at harvest. A total of 363 samples from 25 white and red grape varieties were used to construct quality-prediction models based on reference data and on NIR spectral data obtained using a commercially-available diode-array spectrophotometer (380–1,700 nm). The feasibility of testing bunches of intact grapes was investigated and compared with the more traditional must-based method. Two regression approaches (MPLS and LOCAL algorithms) were tested for the quantification of changes in soluble solid content (SSC), reducing sugar content, pH-value, titratable acidity, tartaric acid, malic acid and potassium content. Cross-validation results indicated that NIRS technology provided excellent precision for sugar-related parameters (r² = 0.94 for SSC and reducing sugar content) and good precision for acidity-related parameters (r² ranging between 0.73 and 0.87) for the bunch-analysis mode assayed using MPLS regression. At validation level, comparison of LOCAL and MPLS algorithms showed that the non-linear strategy improved the predictive capacity of the models for all study parameters, with particularly good results for acidity-related parameters and potassium content.

Multivariate near-infrared reflection spectroscopy strategies for ensuring correct labeling at feed bagging in the animal feed industry

A key concern in animal feed factories is guaranteeing the correct labeling of compound feeds. Therefore, due to incorrect labeling, there is an urgent need for new control methods on the claims that can be made. In this study, this question has been tackled with different multivariate classification algorithms based on the near-infrared spectral fingerprint obtained from a given compound feed analyzed in its original physical market presentation form (i.e., cubes, coarse meals, pellets). The objective of this paper is the evaluation of different methods for establishing a separation among 24 feed types. Two linear methods, soft independent modeling of class analogy (SIMCA) and partial least squares (PLS) with two approaches to classification (PLSD and PLS-LDA); and one nonlinear method, support vector machines (SVM), were studied. The database used had the following structure: a first division was made between granules and meals; within these two groups, there was a second division according to three animal species to which the feed was marketed (bovine, ovine, and porcine); within each species there was a third division according to the age or physiological status of the animal (i.e., lactating dairy cattle, starters, etc.). Given the database structure, all the methods were evaluated following two strategies: (1) development of a model composed of the nine classification models corresponding to the structure of the data; and (2) development of a unique model that discriminates among the 24 classes of different feeds. With both strategies the lowest percentage of misclassified samples was achieved with the SVM method (3.96% with strategy 1 and 2.31% with strategy 2). Among the linear methods evaluated, SIMCA yielded the best results, with a percentage of 8.47% misclassified samples with strategy 1 and 4.05% misclassified samples with strategy 2. The results in this study show the ability of near-infrared spectroscopy to make acceptable classifications of feed types based only on spectral information, with differences in performance depending on the multivariate algorithm used.

Advanced nonlinear approaches for predicting the ingredient composition in compound feedingstuffs by near-infrared reflection spectroscopy

For quantitative applications, the most common usage of near-infrared reflection spectroscopy (NIRS) technology, calibration involves establishing a mathematical relationship between spectral data and data provided by the reference. This model may be fairly complex, since the near-infrared spectrum is highly variable and contains physical/chemical information for the sample that may be redundant, and multivariate calibration is usually required. When the relationship to be modeled is nonlinear, classical regression methods are inadequate, and more complex strategies and algorithms must be sought in order to model this nonlinearity. The development of NIRS calibrations to predict the ingredient composition, i.e., the inclusion percentage of each ingredient, in compound feeds is a complex task, due to the nature of the parameters to be predicted and to the heterogeneous nature of the matrices/formulas in which each ingredient participates. The present paper evaluates the use of least squares support vector machines (LSSVM) and two local calibration methods, CARNAC and locally biased regression, for developing NIRS models to predict two of the most representative ingredients in compound feed formulations, wheat and sunflower meal, using a large spectral library of 7523 commercial compound feed samples. For both ingredients, the best results were obtained using CARNAC, with standard errors of prediction (SEP) of 1.7% and 0.60% for wheat and sunflower meal, respectively, and even better results when the algorithm was allowed to refuse to predict 10% of the unknowns. Meanwhile, LSSVM performed less well on wheat (SEP 2.6%) but comparably on sunflower meal (SEP 0.60%), giving results very similar to those reported previously for artificial neural networks. Locally biased regression was the least successful of the three methods, with SEPs of 3.3% for wheat and 0.72% for sunflower meal. All the nonlinear methods improved on the standard approach using partial least squares (PLS), which gave SEPs of 5.3% for wheat and 0.81% for sunflower meal.

Use of artificial neural networks in near-infrared reflectance spectroscopy calibrations for predicting the inclusion percentages of wheat and sunflower meal in compound feedingstuffs

The use of near-infrared reflectance spectroscopy (NIRS) calibrations to predict the ingredient composition in compound feeds (i.e., inclusion percentage of each ingredient) is a complex task, regarding both the nature of the parameters to be predicted, since they are not well-defined chemical entities, and the heterogeneousness of the matrices/formulas in which each ingredient participates. The present paper evaluates the use of nonlinear regression methods, such as artificial neural networks (ANN), for developing NIRS calibrations to predict these parameters. Two of the most representative ingredients in the Spanish compound feed formulations (wheat and sunflower meal) were selected for evaluating ANN possibilities, using a large spectral library comprising a total of 7523 commercial compound feed samples; 7423 were used as training set and 100 as validation set. Three general models of networks were studied: multilayer perceptron with back-propagation training (BP), multilayer perceptron with Levenberg-Maquartd training (LM), and radial basis function nets (RBF); moreover, in accordance with a factorial design, more complex architectures were evaluated gradually, changing the number of hidden layers and hidden neurons, for the determination of the optimal network topology. For both ingredients, the best results were obtained using ANN with BP training, showing prediction error values (SEP) of 2.72% and 0.66% for wheat and sunflower meal, respectively. These SEP values showed a significant improvement (19%-49% for sunflower meal and wheat, respectively) in comparison with those obtained using calibrations developed with linear methods.