Search

We present the development of a bilinear regression model for multivariate calibration on the basis of maximum correntropy criteria (MCC) whose robustness can be easily controlled. MCC regression methods can be more effective when the assumption of normality does not hold or when data are contaminated with outliers. These methods are competitive when the degree of robustness against outliers should be controlled. By controlling the robustness, information from candidate outliers can be partially retained rather than completely included or discarded during calibration. Within the context of bilinear regression models, an MCC approach using statistically inspired modification of the partial least squares (SIMPLS) is proposed, which is named maximum correntropy‐weighted partial least squares (MCW‐PLS). Thanks to the controllable robustness of MCC models, observations are upweighted or downweighted during the calibration process, rendering robust models with soft discrimination of samples. Such a weighting represents an important advantage, especially for cases when samples are not drawn from a normal distribution. Applications to three real case studies are presented. These applications uncovered three main features of MCW‐PLS: robustness control between SIMPLS and robust SIMPLS (RSIMPLS), improvements in prediction performance of bilinear calibration models, and the possibility to detect the most informative samples in a calibration set. [ABSTRACT FROM AUTHOR]

Abstract: Owing to spectral variations from other sources than the component of interest, large investments in the NIR model development may be required to obtain satisfactory and robust prediction performance. To make the NIR model development for routine active pharmaceutical ingredient (API) prediction in tablets more cost-effective, alternative modelling strategies were proposed. They used a massive amount of prior spectral information on intra- and inter-batch variation and the pure component spectra to define a clutter, i.e., the detrimental spectral information. This was subsequently used for artificial data augmentation and/or orthogonal projections. The model performance improved statistically significantly, with a 34–40% reduction in RMSEP while needing fewer model latent variables, by applying the following procedure before PLS regression: (1) augmentation of the calibration spectra with the spectral shapes from the clutter, and (2) net analyte pre-processing (NAP). The improved prediction performance was not compromised when reducing the variability in the calibration set, making exhaustive calibration unnecessary. Strong water content variations in the tablets caused frequency shifts of the API absorption signals that could not be included in the clutter. Updating the model for this kind of variation demonstrated that the completeness of the clutter is critical for the performance of these models and that the model will only be more robust for spectral variation that is not co-linear with the one from the property of interest. [Copyright &y& Elsevier]

A class of multivariate calibration methods called augmented classical least squares (ACLS) has been proposed which combines an explicit linear additive model with the predictive power of inverse models, such as principal component regression (PCR) and partial least squares (PLS). Because of its use of the explicit linear additive model, ACLS provides an interesting framework to incorporate different sources of prior information, such as measured pure component spectra, in the model. In this study, the predictive power of ACLS models incorporating different amounts of prior information has been compared to that of PCR and PLS using two examples, a designed experiment and one with biological samples. In both cases, the ACLS models showed predictive power comparable to PLS under idealized validation conditions. When a different interferent structure was present in the validation samples, the predictive power of the inverse models (PCR and PLS) dramatically decreased, with an increase in root-mean-squared error of prediction by a factor of 3.5 for the first example and a factor of 2 in the second example. The incorporation of prior information in the ACLS framework was found to considerably reduce or even completely remove these dramatic effects, especially when the pure component contributions for the interferents were taken into account. [ABSTRACT FROM AUTHOR]

• Challenges in multivariate calibration of spectroscopic sensors are discussed. • Discussion of methods for quantitative and qualitative calibration. • All methods are illustrated on an example case from the postharvest domain. • Discussion of applications for postharvest quality evaluation. • Special attention is given to spectral preprocessing, careful validation and calibration transfer. Vibrational spectroscopy methods are widely investigated as fast and non-destructive alternatives for postharvest quality evaluation. As these methods measure spectral responses at a large number of wavebands correlated to the quality traits of interest, multivariate calibration equations have to be built to estimate the quality traits from the acquired spectra. This paper provides an overview of the most important multivariate data analysis techniques for exploring spectral data, detecting outliers and building calibration models for predicting the quality traits of interest. Both linear and non-linear calibration methods are discussed for quantitative (continuous) and qualitative (discrete) quality traits. For each of the presented methods the theory is explained, followed by illustration of an example case from the postharvest domain and a discussion of applications of this technique for postharvest quality evaluation based on spectral sensors. As spectral preprocessing, careful validation and calibration transfer are crucial aspects for successful implementation of spectral sensors for postharvest quality evaluation, special attention is given to these aspects. Finally, conclusions are drawn and recommendations are made with respect to the steps to take and points of attention for successful calibration. [ABSTRACT FROM AUTHOR]

Banana and its dried products are one of the most common foods consumed all over the world. Quality attributes of dried banana slices are typically evaluated using time-consuming and labor intensive methods which do not allow to evaluate each individual slice. Replacement of the current destructive methods for quality evaluation of dried banana slices by fast, non-destructive methods would provide large added value to the food industry. Therefore, the aim of this study was to monitor the moisture content, texture and color of banana slices during the drying process in a fast and non-destructive way. VNIR hyperspectral reflectance imaging in the 400–1000 nm range was selected for this purpose. Thanks to a cross-polarized configuration the effects of glare or specular reflection on the banana slice surfaces in the hyperspectral diffuse reflectance images was largely reduced. Reference quality attributes for all the slices (moisture content, texture and color ( L*, a*, b* values) obtained using conventional destructive methods and colorimeter were predicted from their corresponding average reflectance spectra extracted from the hyperspectral images by means of partial least squares regression (PLSR). The PLSR calibration models were validated on samples which had not been used for model calibration. The results were very good for water content ( R 2 P  = 0.97, RMSEP  = 0.05 kg water/kg DM), quite good for and b* value ( R 2 P  = 0.83, RMSEP  = 1.95), and reasonable for texture ( R 2 P  = 0.66, RMSEP  = 11.8 N), a* value ( R 2 P  = 0.53, RMSEP  = 1.32) and L* value ( R 2 P  = 0.61, RMSEP  = 5.92). Subsequently, these calibration models were used to predict those quality attributes at pixel level for the validation slices to visualize the spatial distribution of these quality parameters at different stages during the drying process. The obtained results clearly indicate the potential of cross-polarized hyperspectral reflectance imaging for non-destructive monitoring of the quality attributes of banana slices during drying. [ABSTRACT FROM AUTHOR]

Calibration transfer has been traditionally performed in the context of transferring models between instruments using standard samples. Recently, new methodologies and applications have shown that transfer techniques can be adopted to achieve calibration transfer between other types of domains, such as product form, variant or seasonality. In addition, to achieving a higher efficiency for calibration transfer, it is desirable to perform the transfer without the need for standard samples or new reference analyses. Therefore, we propose a method for unsupervised calibration transfer based on the orthogonalization for structural differences between domains. The method has been successfully applied to one simulated dataset and two real datasets. In the studied cases, the proposed methodology allowed to achieve a successful transfer of calibration models and enabled the interpretation of the interferences responsible for the degradation of the original calibration models when transferred to the new domain. [Display omitted] [ABSTRACT FROM AUTHOR]

Multivariate calibration models are one of the most predominant tools in analytical chemistry for the non-destructive quantification of constituents. While pipelines, methods, and software are widely available for training calibration models, the monitoring strategies to track the prediction performance in time are still limited and are one of the emerging research topics in chemometrics and machine learning. Moreover, for the sake of efficiency, the development of modern monitoring strategies requires tools that do not require new reference values at each time point. The current study addresses the problem of monitoring calibration models for regression. We present a monitoring paradigm and propose two methods to monitor the performance of the model. The two methods focus on bias drift and spectral variation drift. We present the derivation of the methods and their application to two real case studies from the agrofood industry involving milk composition and apple harvesting. Conclusions and recommendations are presented for practitioners, as well as views for future work in this domain of research. • Views and concepts about multivariate calibration monitoring are presented. • Two unsupervised metrics to monitor calibration models in regression were derived. • Focus on monitoring models for potential bias and spectral variation drift. • Two real cases from the agrofood industry were monitored. [ABSTRACT FROM AUTHOR]

Highlights: [•] A construction of bendable straw particles in DEM is proposed. [•] The particles are calibrated with realistic straw properties. [•] DEM of grain–straw separation with these particles matches with experimental data. [•] A sensitivity study of grain–straw separation w.r.t. crop properties is performed. [•] Crop property influence can be related to straw coverage and grain kernel diameter. [Copyright &y& Elsevier]

Highlights: [•] Performance of calibration techniques using ‘a priori’ information is evaluated. [•] Calibration models are built using NAP, IDC, SBC, ACLS and PLSR. [•] Robustness of models is evaluated in presence of changing interferent structure. [•] Among all techniques, ACLS is found to result in robust model with lowest RMSEP [Copyright &y& Elsevier]

Abstract: The potential of near infrared (NIR) reflectance spectroscopy over the range 780–1690nm was investigated to measure the soluble solids content (SSC) and firmness of bell pepper fruit. Partial least squares (PLS) calibration models were constructed based on a calibration dataset which included data from two cultivars (Solution and Ferrari) and two harvest times (2005 and 2006). The effect of Savitzky–Golay second derivative preprocessing and extended multiplicative signal correction (EMSC) on the accuracy of the calibration models was investigated and the best results were obtained with the former. The SECV were equal to 5.9N and 0.59°Brix for firmness and SSC, respectively. When the model was applied to an external data set including data from cv. Solution and a different harvest season, the satisfactory SEP values of 4.49N and 0.7°Brix were obtained, but for firmness a bias of 5.6N was observed. From these results it can be concluded that NIR spectroscopy can be used as a non-destructive technique for measuring the SSC in bell pepper, but that further research is needed to make it robust for firmness prediction. [Copyright &y& Elsevier]

Abstract: Continuous wave (CW) and time-resolved (TRS) near infrared (NIR) reflectance measurements were carried out on ‘Conference’ pear fruit. While the absorption coefficient spectra obtained from the TRS measurements were dominated by the overtone of the hydroxyl moiety at 975nm, the scattering coefficient spectra were almost flat. The spectral data were used to construct calibration models for the soluble solids content (SSC) and Magness Taylor firmness of ‘Conference’ pear fruit. Reasonable models for SSC (RMSEP=0.44 °Brix) were obtained when the CW spectra in the range 780–1700nm were used, but not when the range was limited to that of the TRS equipment (875–1030nm). No satisfactory calibration model could be established between the TRS absorption coefficient spectrum and SSC. With none of the spectroscopic techniques could significant calibration models for firmness be constructed, although a highly nonlinear relationship between scattering coefficients at 900nm and firmness was observed. [Copyright &y& Elsevier]

Indirect quantification of chemical composition through spectral measurements requires the establishment of multivariate calibration models. The reference analyses on the calibration samples typically form a major cost factor in the establishment of these multivariate models. Therefore, the aim of this study was to select the most informative calibration samples in an unsupervised way based on the spectral measurements. To this end, guidelines to address this challenge in PLSR model building have been developed. The recommendations include calculating a sample size that surpasses the model complexity by a factor of 12, performing the selection in the PCA score space spanned by a sufficiently large number of principal components and using methods such as Kennard-Stone, Puchwein, Clustering or D-optimal designs. We provide the data and methodology used in the present study for future use. • We aim to find a subset that would render a model of comparable performance to the model rendered by the full set of samples. • The optimal sample size for a multivariate calibration model depends on the complexity of such a model. • It is possible to evaluate the quality of the selected samples by comparing covariance matrices for PLSR calibration models. • Control over the sample size and the input dimensionality lowers the impact of the selection method. [ABSTRACT FROM AUTHOR]