Carlos Esquerre

Preventing over-fitting in PLS calibration models of near-infrared (NIR) spectroscopy data using regression coefficients

Selection of the number of latent variables (LVs) to include in a partial least squares (PLS) model is an important step in the data analysis. Inclusion of too few or too many LVs may lead to, respectively, under or over‐fitting of the data and subsequently result in poor future model performance. One well‐known sign of over‐fitting is the appearance of noise in regression coefficients; this often takes the form of a reduction in apparent structure and the presence of sharp peaks with a high degree of directional oscillation, features which are usually estimated subjectively. In this work, a simple method for quantifying the shape and size of a regression coefficient is presented. This measure can be combined with an indicator of model bias (e.g. root mean square error) to aid in estimation of the appropriate number of LVs to include in a PLS model. The performance of the proposed method is evaluated on simulated and and real NIR spectroscopy datasets sets and compared with several existing methods. Copyright

Initial studies on the quantitation of bruise damage and freshness in mushrooms using visible-near-infrared spectroscopy.

Identification of mushrooms that have been physically damaged and the measurement of time elapsed from harvest are very important quality issues in industry. The purpose of this study was to assess whether the chemical changes induced by physical damage and the aging of mushrooms can: (a) be detected in the visible and near infrared absorption spectrum and (b) be modeled using multivariate data analysis. The effect of pre-treatment and the use of different spectral ranges to build PLS models were studied. A model that can identify damaged mushrooms with high sensitivity (0.98) and specificity (1.00), and models that allow estimation of the age (1.0-1.4 days root mean square error of cross-validation) were developed. Changes in water matrix and alterations caused by enzymatic browning were the factors that most influenced the models. The results reveal the possibility of developing an automated system for grading mushrooms based on reflectance in the visible and near infrared wavelength ranges.

Time series hyperspectral chemical imaging data: Challenges, solutions and applications

Hyperspectral chemical imaging (HCI) integrates imaging and spectroscopy resulting in three-dimensional data structures, hypercubes, with two spatial and one wavelength dimension. Each spatial image pixel in a hypercube contains a spectrum with >100 datapoints. While HCI facilitates enhanced monitoring of multi-component systems; time series HCI offers the possibility of a more comprehensive understanding of the dynamics of such systems and processes. This implies a need for modeling strategies that can cope with the large multivariate data structures generated in time series HCI experiments. The challenges posed by such data include dimensionality reduction, temporal morphological variation of samples and instrumental drift. This article presents potential solutions to these challenges, including multiway analysis, object tracking, multivariate curve resolution and non-linear regression. Several real world examples of time series HCI data are presented to illustrate the proposed solutions.

Use of near infrared hyperspectral imaging to identify water matrix co-ordinates in mushrooms (Agaricus bisporus) subjected to mechanical vibration

In this study, white mushrooms (Agaricus bisporus) were subjected to physical perturbation by mechanical vibration. Hyperspectral images were obtained after perturbation using a pushbroom line-scanning instrument operating in the wavelength range of 1000–1700 nm (7nm spectral resolution). Changes in sample spectra arising from perturbation were examined by observation of difference spectra and partial least squares regression (PLSR) coefficients. Different spectral pre-treatments [multiplicative scatter correction (MSC), extended multiplicative scatter correction (EMSC) and standard normal variate (SNV)] were employed in order to decrease spectral variability caused by scattering and differences in the optical path length due to physical changes in the mushrooms induced by the perturbation. Candidate water matrix co-ordinates were proposed at 950 nm, 1174 nm, 1398 nm, 1433 nm, 1454 nm, 1496 nm and 1510 nm. Mechanical vibration increased the concentration of weakly hydrogen-bonded water and decreased that of strongly hydrogen-bonded water in the mushrooms without causing changes in the bulk moisture content.

Wavelength selection for development of a near infrared imaging system for early detection of bruise damage in mushrooms ( Agaricus bisporus )

Previous research has demonstrated the potential use of near infrared (NIR) hyperspectral imaging for non-destructive monitoring of mushroom quality. The mushroom industry demands economical and high-throughput imaging systems that can reliably classify groups of mushrooms according to quality parameters. Multispectral imaging systems based on the acquisition of just a few (2–10) wavelengths fulfil these criteria. This research concerns the development of a low-cost robust multispectral system for mushroom quality control which can identify slightly damaged mushroom tissue using NIR spectral images. A three step approach was employed: HI the most suitable pre-treatment was selected; (2) wavelengths with the most stable normalised regression coefficients were identified using ensemble Monte Carlo variable selection (EMCVS); and (3) partial least square discriminant analysis (PLS-DA) models were built using the selected regions (49 nm bandwidth) to simulate a multispectral system. Minimum scaled reflectance spectra produced better results than maximum scaled, mean scaled, median scaled or raw spectra. Five key spectral regions were identified, centred around 971 nm, 1090 nm, 1188 nm, 1384 nm and 1454 nm. A PLS-DA model built using three spectral regions (1090 nm, 1188 nm, 1384 nm) and scaled by the 1454 nm band (minimum reflectance) correctly classified 100% of the physically damaged mushrooms.

A Study on the Application of Near Infrared Hyperspectral Chemical Imaging for Monitoring Moisture Content and Water Activity in Low Moisture Systems

Moisture content and water activity are key parameters in predicting the stability of low moisture content products. However, conventional methods for moisture content and water activity determination (e.g., loss on drying method, ‎Karl Fischer titration, dew point method) are time consuming, demand specialized equipment and are not amenable to online processing. For this reason they are typically applied at-line on a limited number of samples. Near infrared hyperspectral chemical imaging is an emerging technique for spatially characterising the spectral properties of samples. Due to the fast acquisition of chemical images, many samples can be evaluated simultaneously, thus providing the potential for online evaluation of samples during processing. In this study, the potential of NIR chemical imaging for predicting the moisture content and water activity of a selection of low moisture content food systems is evaluated.

A frame-based ANN for classification of hyperspectral images: assessment of mechanical damage in mushrooms

Imaging spectroscopy integrates conventional imaging and spectroscopy to attain both spatial and spectral information from an object. Processing and analysis of a hypercube can be a hard task. Robust methods for hyperspectral data classification are required, insensitive to imaging deficiencies and high input dimension. Mushrooms have a thin and porous epidermal structure, and are sensitive to handling and transportation practices. Mechanical damage triggers a browning process within the tissue changing its metabolic state. The objective is to quantify different levels of physical perturbation on the mushroom pilei, using near-infrared spectral images and machine learning approaches. An ANN classifier is implemented, whose input is a small set of vectors containing representative information, and output is the set of categorical labels that correspond to different levels of mechanical vibration. For obtaining a salient dataset for classifying the images, the Harris corner detection algorithm is employed. The advantage of using interest points is to replace an exhaustive search over the entire image space by a computation over a concise set of highly informative points. A frame-based classification approach is proposed and shown to produce an increase in the classification accuracy, since feature vectors regarded as single instances may not always carry sufficient discriminant information. Comparisons with statistical features computed from wavelet coefficients showed that interest points are more suitable in assessing mechanical perturbation. Comparisons on a classifier level with support vector machines showed that ANNs perform better for the specific application, implying a connection between the classification method and the underlying learning problem. Overall, the frame-based classification scheme reduced the misclassification rate. This approach is suited for challenging classification problems where the degree of class separation is variable, i.e., assessment of mechanical damage in mushrooms.

Selection of variables based on most stable normalised partial least squares regression coefficients in an ensemble Monte Carlo procedure

A modification of ensemble Monte Carlo uninformative variable elimination (EMCUVE) is proposed, which does not involve the use of random variables, with the aim of improving the performance of partial least squares (PLS) regression models, increasing the consistency of results and reducing processing time by selecting the most informative variables in a spectral dataset. The proposed method (ensemble Monte Carlo variable selection—EMCVS) and the robust version (REMCVS) were compared to PLS models and with the existing EMCUVE method using three near infrared (NIR) datasets, i.e. prediction of n-butanol in a five-solvent mixture, moisture in corn and glucosinolates in rapeseed. The proposed methods were more consistent, produced models with better predictive accuracy (lower root mean squared error of prediction) and required less computational time than the conventional EMCUVE method on these datasets. In this application, the proposed method was applied to PLS regression coefficients but it may, in principle, be used on any regression vector.

Penetration Depth Measurement of Near-Infrared Hyperspectral Imaging Light for Milk Powder

The increasingly common application of the near-infrared (NIR) hyperspectral imaging technique to the analysis of food powders has led to the need for optical characterization of samples. This study was aimed at exploring the feasibility of quantifying penetration depth of NIR hyperspectral imaging light for milk powder. Hyperspectral NIR reflectance images were collected for eight different milk powder products that included five brands of non-fat milk powder and three brands of whole milk powder. For each milk powder, five different powder depths ranging from 1 mm–5 mm were prepared on the top of a base layer of melamine, to test spectral-based detection of the melamine through the milk. A relationship was established between the NIR reflectance spectra (937.5–1653.7 nm) and the penetration depth was investigated by means of the partial least squares-discriminant analysis (PLS-DA) technique to classify pixels as being milk-only or a mixture of milk and melamine. With increasing milk depth, classification model accuracy was gradually decreased. The results from the 1-mm, 2-mm and 3-mm models showed that the average classification accuracy of the validation set for milk-melamine samples was reduced from 99.86% down to 94.93% as the milk depth increased from 1 mm–3 mm. As the milk depth increased to 4 mm and 5 mm, model performance deteriorated further to accuracies as low as 81.83% and 58.26%, respectively. The results suggest that a 2-mm sample depth is recommended for the screening/evaluation of milk powders using an online NIR hyperspectral imaging system similar to that used in this study.

Influence of Polymer Packaging Films on Hyperspectral Imaging Data in the Visible–Near-Infrared (450–950 nm) Wavelength Range

Hyperspectral imaging (HSI) has recently emerged as a useful tool for quality analysis of consumer goods (e.g., food and pharmaceutical products). These products are typically packaged in polymeric film prior to distribution; however, HSI experiments are typically carried out on such samples ex-packaging (either prior to or after removal from packaging). This research examines the effects of polymer packaging films (polyvinyl chloride (PVC) and polyethylene terephthalate (PET)) on spectral and spatial features of HSI data in order to investigate the potential of HSI for quality evaluation of packaged goods. The effects of packaging film were studied for hyperspectral images of samples obtained in the visible–near-infrared (Vis-NIR, i.e., 450–950 nm) wavelength range, which is relevant to many food, agricultural, and pharmaceutical products. The dominant influence of the films tested in this wavelength range could be attributed to light scattering. Relative position of the light source, film, and detector were shown to be highly influential on the scattering effects observed. Detection of features on samples imaged through film was shown to be possible after some data preprocessing. This suggests that quality analysis of products packaged in polymer film is feasible using HSI. These findings would be useful in the development of quality monitoring tools for consumer products post-packaging using HSI.