Search

Hyperspectral imaging (HSI) is one of the most studied optical techniques to estimate the internal quality of fruits and vegetables. Absorbance and reflectance of the light radiation are specific to each biological tissue and are directly related to its chemical composition and physical characteristics. These properties are influenced by other extrinsic factors, such as the instrumentation or the light source, which can reduce the reproducibility of the experiments. Determining the actual depth of light penetration into tissue could help validate non-contact methods as accurate tools to assess quality properties based on optical properties. In the case of HSI systems, it is crucial to know how far the light penetrates at each wavelength. A non-destructive approach, based on the spatially resolved spectroscopic principle, was proposed to estimate the light penetration depth of a HSI system in a Vis-NIR configuration (in the range 450–1050 nm). This method was applied to measure the light penetration depth in persimmon fruit. The absorption (μ a) and scattering (μ' s) coefficients from Farrell's diffusion theory were estimated using the backscattered light measured at different distances from the incident point light at each wavelength in hyperspectral images of persimmon fruit. The actual light penetration depth was obtained by measuring the reflectance of cut pieces of persimmon fruit with different thicknesses. Linear regression was used to relate the depth of penetrability obtained by both protocols, the estimated or non-destructive protocol and the actual or destructive protocol, showing a high relationship (R2 > 0.8 and RPD>2.5) in the range 610–1050 nm. This confirms that this non-destructive approach proposed for estimating the light penetration depth of a Vis-NIR HSI system in persimmon fruit is accurate, so it could be used as a valuable method to evaluate other HSI systems for different fruits. • Light penetrability of a Vis-NIR HSI system in fruit was estimated non-destructively. • Absorption and reduced scattering were estimated using Farrell's diffusion theory. • A method was developed to measure the actual penetrability destructively. • High relationship was found between estimated and actual light penetrability. • Light penetration depth for persimmon fruit was limited to 2 mm. [ABSTRACT FROM AUTHOR]

Machine learning models were developed to predict the degree of rancidity of beef by a non-destructive method using a near infrared hyperspectral image acquisition system. The beef subject to the experiment was naturally oxidized during the 15-day cooling process. In a darkroom environment, hyperspectral data cubes were collected using a data acquisition device. Additionally, a technique was developed to selectively extract lean-meat spectra from hyperspectral data obtained from beef that was refrigerated for a variety of lengths of time. Thiobarbituric acid reactive substances (TBARS) experiment was performed in a traditional method to secure reference values for the rancidity level of the sample. Spectra were extracted through data selection and separated by training set and test set. PLSR, ANN, and 1D-CNN techniques were applied to model development. Variable Importance in Projection (VIP) score for the wavelength band was calculated, and the portion judged as valid was cut out to generate a reduced data set. Chemical maps were created for each developed model to visualize the performance of the model. As a result of the development, it was confirmed that the rancidity level of beef could be predicted through a model generated by hyperspectral data. • Proposal of non-destructive and real time method for predicting rancidity of beef using hyperspectral spectrum. • Suggestion of technical method for extracting only valid areas from hyperspectral data cubes. • ANN based predictive model showed the best results in both the full spectrum and selectively reduced spectrum. • Data reduction using significant wavelengths is expected to be efficient in application. [ABSTRACT FROM AUTHOR]

Abstract Hitherto, the rapid and nondestructive determination of the moisture content of tea leaves is still an unresolved issue because the upward facing surfaces of tea leaves lying on a conveyor belt are randomly chosen by the collapse of the leaves onto their front side or back side. To study the above issue, hyperspectral images of both the front side and back side of tea leaves on a conveyor belt were captured in the lab to simulate a practical production environment, and LS-SVR models with R v 2 values of 0.951 and 0.918 for the front side and back side, respectively, were established based on their characteristic spectral bands. To ensure that the spectrum of each pixel can be correctly imported into its corresponding model, a logistic regression classifier with a correct classification rate of 100% was designed to identify the front side and back side of the leaves. Finally, a distribution map of the moisture content of the tea leaves was generated successfully according to the following steps: (1) Extracting the average spectrum of each leaf; (2) Identifying which side of the leaf the spectrum belongs to; (3) Importing the adjusted spectrum of each pixel into its corresponding regression model; and (4) Generating a distribution map of the moisture content. This research creatively provides a scheme for detecting the moisture content of tea leaves. Highlights • Establish regression and classification models for different sides of tea leaves. • Achieve the visualization of moisture content of tea leaves on different sides. • Provide an automatic moisture content detection scheme for stacked tea leaves. [ABSTRACT FROM AUTHOR]

Abstract This study aimed at the assessment of pork tenderness using hyperspectral backscattering imaging technique. Backscattering images for 54 pork samples in the spectral range of 480–900 nm were characterized by 3-parameter Lorentzian distribution function (LD) and diffusion equation (DE), which represent a statistical model and a theoretical model respectively. Characteristic parameters spectra were obtained and employed to establish partial least squares regression (PLSR) models for the Warner-Bratzler Shear Force (WBSF) values of pork samples. By contrast, the results showed that LD function interpreted pork scattering profiles more accurately than DE model between 480 and 600 nm. The PLSR model based on two-LD parameter combinations (the asymptotic value a and the peak value b of the backscattering profile) gained the best prediction results, with correlation coefficient in the prediction set (R p) of 0.902, root mean square error of prediction (RMSEP) of 5.218 N, indicating that hyperspectral backscattering imaging technique coupled with LD model is preferable for rapid and nondestructive evaluation of pork tenderness. Highlights • Hyperspectral scatter imaging was investigated to predict pork tenderness. • Scattering profiles were characterized by statistical and theoretical models. • The fitting capabilities of two models for scattering profiles were compared. • Characteristic parameters were explored for predicting pork tenderness. [ABSTRACT FROM AUTHOR]

Abstract Hyperspectral imaging (HSI) can acquire data in two modes: imaging and spectroscopy, revealing the spatially-resolved spectral properties of materials. Traditional HSI processing in the close-range domain primarily focuses on the spectral information with minimal utilisation of the spatial information present in the data. The present work describes a methodology for utilising the spatial information present in HSI data to improve classification modelling over that achievable with spectral information alone. The methodology has been evaluated using near infrared (NIR) HSI data of sixteen green tea products from seven different countries. The methodology involves selecting and sharpening an image plane to enhance the textural details. The textural information is then extracted from the statistical properties of the grey level co-occurrence matrix (GLCM) of the sharpened image plane using a moving window operation. Finally, the textural properties are combined with the spectral information using one of the three different levels of data fusion, i.e. raw data level, feature level and decision level. Raw data-level fusion involved concatenating the spectral and textural data before performing the classification task. The feature-level fusion involved performing principal component analysis (PCA) on spectral and textural information and combining the PC scores obtained prior to performing classification. Decision-level fusion involved a majority voting scheme to enhance the final classification maps. All the classification tasks were performed using multi-class support vector machine (SVM) models. The results showed that combining the textural and spectral information during modelling resulted in improved classification of the sixteen green tea products compared to models built using spectral or textural information alone. Highlights • Green tea products were analysed by near infrared hyperspectral imaging. • Textural information was extracted from the grey level co-occurrence matrix. • Textural properties were fused with near-infrared spectral information. • Data fusion improved the classification accuracy for green tea products. [ABSTRACT FROM AUTHOR]

Abstract Hyperspectral imaging in the visible and near infrared spectral range (450–1664 nm) coupled with chemometrics was investigated for classification of brined and non-brined pork loins and prediction of brining salt concentration employed. Hyperspectral images of control, water immersed and brined (5, 10 or 15% salt (w/v)) raw and cooked pork loins from 16 animals were acquired. Partial least squares (PLS) discriminative analysis models were developed to classify brined pork samples and PLS regression models were developed for prediction of brining salt concentration employed. The ensemble Monte Carlo variable selection method (EMCVS) was used to improve the performance of the models developed. Partial least squares (PLS) discriminative analysis models developed correctly classified brined and non-brined samples, the best classification model for raw samples (Sen = 100%, Spec = 100%, G = 1.00) used the 957–1664 nm spectral range, and the best classification model for cooked samples (Sen = 100%, Spec = 100%, G = 1.00) used the 450–960 nm spectral range. The best brining salt concentration prediction models developed for raw (RMSE p 1.9%, R2 p 0.92) and cooked (RMSE p 2.6%, R2 p 0.83) samples used the 957–1664 nm spectral range. This study demonstrates the high potential of hyperspectral imaging as a process analytical tool to classify brined and non-brined pork loins and predict brining salt concentration employed. Highlights • Vis-NIR hyperspectral imaging is suitable for the assessment of brining of raw and cooked pork loins. • Chemometric models were developed to classify brined and non-brined pork samples and to predict brining salt concentration. • Spectral pre-treatments and variable selection improved performance of models developed. [ABSTRACT FROM AUTHOR]

Abstract Secondary structure of proteins is closely related to the overall quality of meat. The sequence of the changes in chemical bonds caused by secondary structure modification and rapid detection method of secondary structure fractions was investigated by near-infrared hyperspectral imaging combined with generalized (G2D-CS) and heterospectral (H2D-CS) two-dimensional correlation analysis in this study. Synchronous and asynchronous spectral results from G2D-CS analysis indicated that, as α-helix fraction decreasing, the spectral intensity of N H stretching changed before that of the C H and O H, and the spectral fluctuation of O H stretching occurred after the carbonyl related C H band and before those aliphatic and aromatic C H band. Feature wavebands identified by H2D-CS analysis obtained good results with R2 CV of 0.836 in predicting α-helix fractions. The results of this study are useful for interpreting the secondary structure modification process of proteins during frozen storage and monitoring the secondary structure fraction in a rapid way. Highlights • NIR HSI was employed by the first time to monitor α-helix fraction of actomyosin. • G2D-CS was used for understanding the process of secondary structure modification. • H2D-CS was applied for identifying α-helix related NIR wavebands. • PLSR model built by the selected wavebands showed good results. [ABSTRACT FROM AUTHOR]

Abstract Protein is an important nutrient for people diet, and protein content is one of the most vital properties in quality assessment of pork meat. The potential of line-scanning based hyperspectral imaging (HSI) has been proved to be able to assess the protein content of meat, however quality of images obtained from line-scanning HSI is affected by sample transformation conditions. A novel single shot (snapshot) HSI sensor was employed to evaluate the protein content of numerous processed pork meat by using back propagation - neural network (BP-NN) and partial least squares regression (PLSR) predictive models. The results of BP-NN were better than that of PLSR. The best spectral profile was selected from reflectance, absorbance, and Kubelka–Munk (K–M) spectra by comparing the performances of their BP-NN models. The BP-NN model combined with absorbance spectra showed the best performance for evaluating the protein content of various processed pork meats with determination coefficient of cross-validation set (R2 CV) = 0.8318, and root mean square error of cross-validation set (RMSECV) = 8.38 mg/g, respectively. Results indicated the feasibility of determining the protein content of pork meats by means of single shot HSI. Highlights • The single shot hyperspectral imaging (HSI) could capture an image in 1/150 s. • Protein content of pork decreased after wet salting – dehydrating. • Absorbance spectra provided better accuracy. • BP-NN model was used to predict protein contents of pork by snapshot HSI. • Protein contents distribution of processed pork were showed by snapshot HSI. [ABSTRACT FROM AUTHOR]

In this paper, the feasibility of using spectral profiles for modeling water activity a w in white quinoa grains (Chenopodium quinoa Willd.) is studied. For this purpose, five hundred samples of five white varieties were stabilized at different a w values using the isopiestic method. Next, hyperspectral images (HSIs) of ten grains for each combination (variety, a w value), covering the range of 400–1000 nm were acquired, and mean spectral for each grain extracted. Then, due to a linear relationship that the spectral profiles are shown, the modeling was performed with a w values over 0.741 using partial least square regression (PLSR). From total spectra, three hundred spectrum were selected and randomly divided into training and validation sets. The results shown coefficient of determination from 0.59 to 0.834 concluding than for a w over 0.741, HSI + PLSR show potential for a w prediction in white quinoa grains. [ABSTRACT FROM AUTHOR]

The feasibility of using spectral profiles for the estimation of the shelf life of the rocket leaves was evaluated using a multivariate accelerated shelf life testing (MASLT) approach. Spectral changes over time were modeled by using principal component analysis (PCA) and as variation to the conventional method, partial least squares (PLS) method. Kinetic charts were built fitting the first principle component (PC1) and the first latent variable (LV1) scores versus time. In both cases, the kinetics were described by a first order reaction, obtaining R 2 values of 0.73, 0.94 and 0.95 for samples stored at 5, 10 and 15 °C, respectively. The spectra of samples judged unacceptable were used for the calculation of the cut-off value, estimated to be 3.955, leading to shelf life estimations of 9.8, 4.3 and 3.1 days for PCA based MASLT at the three temperatures, respectively. For PLS based MASLT the shelf life was 9.4, 4.5 and 3.3 days for samples stored at the three respective temperatures. Conclusively, shelf-life was correctly estimated by conventional MASLT using PCA and also with the newly proposed technique using PLS. [ABSTRACT FROM AUTHOR]

Tea is the most consumed manufactured drink in the world. In recent years, various high end analytical techniques such as high-performance liquid chromatography have been used to analyse tea products. However, these techniques require complex sample preparation, are time consuming, expensive and require a skilled analyst to carry out the experiments. Therefore, to support rapid and non-destructive assessment of tea products, the use of near infrared (NIR) (950–1760 nm) hyperspectral imaging (HSI) for classification of six different commercial tea products (oolong, green, yellow, white, black and Pu-erh) is presented. To visualise the HSI data, linear (principal component analysis (PCA) and multidimensional scaling (MDS)) and non-linear (t-distributed stochastic neighbour embedding (t-SNE) and isometric mapping (ISOMAP)) data visualisation methods were compared. t-SNE provided separation of the six commercial tea products into three groups based on the extent of processing: minimally processed, oxidised and fermented. To perform the classification of different tea products, a multi-class error-correcting output code (ECOC) model containing support vector machine (SVM) binary learners was developed. The classification model was further used to predict classes for pixels in the HSI hypercube to obtain the classification maps. The SVM-ECOC model provided a classification accuracy of 97.41 ± 0.16% for the six commercial tea products. The methodology developed provides a means for rapid, non-destructive, in situ testing of tea products, which would be of considerable benefit for process monitoring, quality control, authenticity and adulteration detection. [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]

Hyperspectral imaging (1000–2500 nm) was used for rapid prediction of moisture and total lipid content in intact green coffee beans on a single bean basis. Arabica and Robusta samples from several growing locations were scanned using a “push-broom” system. Hypercubes were segmented to select single beans, and average spectra were measured for each bean. Partial Least Squares regression was used to build quantitative prediction models on single beans (n = 320–350). The models exhibited good performance and acceptable prediction errors of ∼0.28% for moisture and ∼0.89% for lipids. This study represents the first time that HSI-based quantitative prediction models have been developed for coffee, and specifically green coffee beans. In addition, this is the first attempt to build such models using single intact coffee beans. The composition variability between beans was studied, and fat and moisture distribution were visualized within individual coffee beans. This rapid, non-destructive approach could have important applications for research laboratories, breeding programmes, and for rapid screening for industry. [ABSTRACT FROM AUTHOR]

Near-infrared (NIR) hyperspectral imaging was used to evaluate soluble solids content (SSC) in ‘Fuji’ apples [ Malus sylvestris (L.) Mill. var. domestica (Borkh. Mansf.)]. Eighty ‘Fuji’ apples were analyzed by collecting four small block samples from each one (approximately 2.0 cm × 2.0 cm × 1.5 cm). Partial least squares (PLS) regression analysis was performed to determine the relation between SSC reference data and NIR spectral data measured from each sample. The cross-validation coefficient of determination (r 2 ) between predicted and measured SSC values is 0.89 with a root mean squared error of cross-validation (RMSECV) of 0.55%. Then, we successfully mapped SSC at a high spatial resolution (375 μm per pixel). In addition, the absorption and reduced scattering coefficients of the measured samples were determined based on a diffusion theory model. The absorption coefficients are positively correlated to the SSC values (chemical information), whereas water cored tissue content (physical information) causes a characteristic change in light scattering coefficients. The fitting results were validated by Monte Carlo simulation, and the light penetration depth in ‘Fuji’ apples was estimated to be around 0.33 cm at 1198 nm and 0.17 cm at 1450 nm, respectively. [ABSTRACT FROM AUTHOR]

The ability to quantify and qualify subtle differences between milk powders is very advantageous to industrial manufacturers. Hyperspectral imaging (HSI) combines the spatial attributes of image processing with the chemical diagnostic attributes of spectroscopy, and was evaluated to determine if it could be used to discriminate between milk powders produced in various factories, and of differing functional qualities, such as dispersibility. The results showed that HSI can achieve these aims when multivariate analysis techniques such as Principal Component Analysis (PCA) and Partial Least Squares (PLS) regression are applied. The PCA results showed that the most obvious differences were in the first and second principal components. Strategies to pre-process hyperspectral data, and to optimally automatically detect and remove artefacts in the images were also established. The PLS results showed that the information from HSI can be used to predict with reasonable accuracy the key functional property of dispersibility, and is the first step in a ‘real-time quality’ initiative to establish correlations between hyperspectral images and key quality attributes of milk powder either on, or at-line in close to real-time. [ABSTRACT FROM AUTHOR]

Total volatile basic nitrogen (TVB-N) content is an important index used to evaluate the freshness of pork. In this paper, a strategy for measurement of TVB-N content in pork through hyperspectral imaging (HSI) (400–1000 nm) was developed. Firstly, image textural features based on Gabor filter and spectral features were obtained from the hyperspectral image after determining the region of interest. Then, nine feature wavelengths were selected using partial least-squares projection algorithm. And, major components were obtained from the 2D principal component analysis (2DPCA). Finally, a calibration model was established based on major components using least-squares support vector machine to predict TVB-N values. The results of two methods for data fusion, which are 2DPCA and principal component analysis (PCA), are compared. The correlation coefficients of prediction ( R P ) and root-mean-square errors of prediction (RMSEP) obtained through 2DPCA were 0.955 and 1.86 mg/100 g respectively, which was superior to the results based on PCA ( R P = 0.944, RMSEP = 2.07 mg/100 g). Compared to PCA, the residual prediction deviations (RPD) based on 2DPCA was raised from 3.01 to 3.35. Results demonstrated that the proposed model based on 2DPCA exhibited potential for nondestructive detection of TVB-N content in pork. [ABSTRACT FROM AUTHOR]

This present work aims to provide new insights into the spectral variations associated with water and plasticizers in the protein matrix of casein based biopolymers by conducting time series hyperspectral chemical imaging (HCI). Near infrared (NIR) hyperspectral images (880–1720 nm) of sodium caseinate film were acquired at different periods during film forming process. Second derivative spectra revealed a red shift (from lower to higher wavelength) in the region of 1410–1492 nm during drying, indicating that hydrogen bond distribution converted from free toward associated hydroxyl groups. The introduction of plasticizers resulted in a strong hydrogen bonding matrix in the caseinate polymeric film as evidenced by analysis of second derivative spectra and curve fitting results. Moisture content (MC) of film samples were predicted with coefficient of determination in prediction ( R P r e d 2 ) of 0.96 and 0.97, and root mean square error of prediction (RMSEP) of 0.067 and 0.056 by conducting partial least squares (PLS) and support vector machine (SVM), respectively. SVM model was further used to visualize water distribution within the film samples over drying time and the time-evolution of water migration into a dried film. A novel time series spectral angle mapper (SAM) was also proposed to elucidate water penetration pattern in film material. Time series principal components analysis (PCA) and K-means analysis exhibited temporal variation of film samples during drying steps of film formation. The results clearly demonstrated the occurrence of strong hydrogen-bonded aggregates in the film samples, which was more likely due to the congregation of redundant plasticizers. The overall results confirmed that time series NIR HCI is a powerful and versatile tool to investigate water and plasticizer interactions in the protein-based polymeric film. [ABSTRACT FROM AUTHOR]

Fruits with thick rind have been reported to interfere with the measurement of internal quality of non-destructive near infrared spectroscopy. This review provides an overview of issues related to the use of near infrared spectroscopy for measuring internal and external quality attributes of horticultural produce with thick rinds. The use of other non-destructive techniques for assessing internal and external quality thick rind fruit, such as hyperspectral and multispectral imaging systems, X-ray micro-computed tomography, nuclear magnetic resonance and Raman spectroscopy are also discussed. A concise summary of research and potential commercial application for each of the techniques is highlighted. [ABSTRACT FROM AUTHOR]

Freshness of hens’ eggs is important for consumers and the food processing industry and the Haugh unit (HU) is a commonly used index for freshness. Measurement of HU is destructive and also assumes that the sample that is tested accurately reflects the batch of eggs being processed. This research tests the use reflectance near infrared hyperspectral imaging in the wavelength range of 900–1700 nm for nondestructive prediction of eggs freshness and compared these measurements to HU. To achieve this fresh eggs were stored at 25 °C and were measured after storage for 0, 4, 7, 10, 14, 18 and 21 days by hyperspectral imaging technique and compared to HU for each egg. Hyperspectral imaging technique combines between conventional imaging and NIR spectroscopy to achieve spatial and spectral information from eggs. The acquired near infrared hyperspectral imaging data from samples in the calibration set were analyzed in order to develop a calibration model for HU using partial least squares regression (PLSR) and then crossvalidated. The standard normal variate transformation (SNV) spectral pretreatment gave the optimum conditions for establishing the calibration model with a coefficient of determination (R2) of 0.91 and root mean square error of calibration (RMSEC) of 4.58. Distribution maps of HU were generated from the acquired calibration model by interpretation of predicted HU to different colors using image processing algorithms. Displayed colors of acquired image of eggs were different correspond to the freshness of the eggs based on HU. The results show that the near infrared hyperspectral imaging technique can be possible to use for presenting the images of egg related to HU in order to nondestructively evaluate hens’ eggs freshness. [ABSTRACT FROM AUTHOR]

The aim of this study was to develop an improved method for detecting internal intrusions (pits and their fragments) in cherries using hyperspectral imaging technique in transmittance mode. Hyperspectral transmission images of pitted and intact cherries of three popular cultivars: ‘Łutówka’, ‘Pandy 103’, and ‘Groniasta’, differing by soluble solid content (SSC), were acquired in the visible and near-infrared (VNIR) range (450–1000 nm). The Correlation-based Feature Selection (CFS) algorithm and 2nd derivative pre-treatments of the hyperspectral data were used to construct the supervised classification models. From all the studied classifiers, the best prediction accuracies for whole pit or pit fragment detection were obtained by the backpropagation neural network (BNN) model (94.6% of correctly classified instances for fresh samples and 83.3% for frozen samples). The accuracy of distinguishing between drilled and intact cherries was higher than 87% jointly for fresh and frozen cherries. These results showed that hyperspectral imaging in transmittance mode is an accurate and objective tool for pit detection in fresh and frozen cherries and can be applied in on-line sorting systems. [ABSTRACT FROM AUTHOR]

Visible–near-infrared (450–1040 nm) hyperspectral reflectance imaging was studied in order to assess the internal physicochemical properties and sensory perception of ‘Big Top’ and ‘Magique’ nectarines ( Prunus persica L. Batsch var. nucipersica ) (yellow and white-flesh cultivar, respectively) during ripening using the Ripening Index (RPI) and the Internal Quality Index (IQI). Hyperspectral images of the intact fruits were acquired during the ripeness under controlled conditions, and their physicochemical properties (flesh firmness, total soluble solids, titratable acidity and flesh colour) were analysed. IQI and RPI were used to relate the spectral information obtained from nectarines with the physicochemical properties and the sensory perception of their maturity using Partial Least Square (PLS) regression with proper variable selection. Optimal results were obtained with R 2 values higher than 0.87 for the two indices and the two cultivars. The ripeness of each fruit could be visualised by projecting the PLS models of the IQI on the pixels of the fruits in the images, showing great potential for further monitoring of the evolution of intact nectarine ripeness in industrial setups. [ABSTRACT FROM AUTHOR]

The food industry must maintain high quality and safety standards. These goals can be achieved by applying analytical procedures able to provide information about composition, structure, physicochemical properties, and sensory characteristics of foods. The conventional analytical techniques are often time-expensive and unsuitable to be used on line, and require sample preparation. Instead, the modern food industry requires efficient and non-invasive inspection technologies able to provide information about external and internal quality attributes of food. An example is given by hyperspectral imaging, which allows the obtainment of spatial, spectral, and multi-constituent information. This note provides an up-to-date review on the major applications of hyperspectral imaging to liquid and semi-liquid food products (oils, milk, yogurt, and eggs). [ABSTRACT FROM AUTHOR]

A new method for blood detection in whitefish fillets is presented. The method is based on diffuse reflectance hyperspectral imaging in the VIS/NIR range, and unmixing of measured absorbance spectra into known spectra for hemoglobin, water and muscle tissue. Scattering effects are modelled as a polynomial function of wavelength, and are included in the unmixing procedure. The unmixing method is based on a modified Beer-Lambert law and is performed using a non-negativity constrained least squares algorithm. Application of the method to images of samples with varying amounts of blood results in a very good fit between measured and modelled spectra. Reference measurements on homogenized cod muscle with controlled concentrations of added blood are used to model the relationship between the spectral unmixing results and hemoglobin concentration, enabling pixel-by-pixel estimation of hemoglobin concentration. The method is seen as a useful tool for automatic quality grading and processing of fillets in the whitefish industry. [ABSTRACT FROM AUTHOR]

Visible and near-infrared (VNIR) hyperspectral imaging (400–900 nm) was used to evaluate pH of fresh chicken breast fillets (pectoralis major muscle) from the bone (dorsal) side of individual fillets. After the principal component analysis (PCA), a band threshold method was applied to the first principal component (PC 1 ) score image in order to get the region of interest (ROI). Then, the average reflective spectrum of ROI of each image was acquired by inverse PCA transform. Eight pretreatment algorithms were evaluated for partial least squares regression (PLSR) models. The PLSR model with the pretreatment of multiplicative scatter correction followed by second derivative showed the best performance with coefficients of determination for validation ( R v 2 ) of 0.87, root mean square error for validation (RMSEv) of 0.16 and the ratio of percentage deviation (RPD) of 2.02. Optimal 20 wavelengths were selected using competitive adaptive reweighed sampling (CARS) method to develop a new multispectral PLSR model, leading to an enhanced result with R v 2 of 0.94, RMSEv of 0.06 and RPD of 3.55. To assess the performance of the prediction models, new ROIs where pH values were measured using a pH probe, were defined and corresponding mean spectra were used as an independent test set of the new multispectral PLSR model. Coefficients of determination for independent test set ( R p 2 ) and root mean square error for independent test set (RMSEp) were 0.73 and 0.29, respectively. The prediction image showing the spatial distribution of the predicted pH values was generated to analyze the spatial context of pH values as well as the overall pH level of each fillet. The results demonstrated that VNIR hyperspectral imaging could be used to predict spatial and global pH values of fresh chicken breast meat. [ABSTRACT FROM AUTHOR]

Herein, a hyperspectral imaging system in the 380–1030 nm range was used to rapidly determine the moisture content of scallops in different dehydration periods. Mean spectral values of scallops were extracted from hyperspectral images. Only eight optimal wavelengths were selected using the regression coefficient method. Spectra of full wavebands and selected wavelengths were used as independent variables for modeling. Partial least squares regression (PLSR) and least-squares support vector machines (LSSVM) were employed to establish multispectral calibration models to correlate spectral features with moisture content. The best results, with correlation coefficients of prediction (R P ), root mean square error of prediction (RMSEP), and residual predictive deviation (RPD) of 0.9673, 3.5584%, and 3.7150, respectively, were achieved using the optimal wavelength-based PLSR model. To visualize moisture content in scallops, a visualization map was generated using the selected wavelength-based PLSR model. These results highlight the potential of hyperspectral imaging for non-destructive prediction of moisture content in scallops. [ABSTRACT FROM AUTHOR]

Freezing, storage and thawing all significantly affect the quality of seafood products. In this article, we explore the potential for estimating these parameters using online optical spectroscopic measurements. Fillets of cod were vacuum packed and underwent different programs of freezing, thawing and storage. Hyperspectral imaging of the fillets was performed at each program stage. Both single and double-frozen samples are studied. Different freezing and thawing methods are also investigated. Multivariate analysis of the hyperspectral data show that freezing history can clearly be determined on samples in the frozen state and to some degree on samples in the thawed state. Analysis of the data shows a less clear distinction between methods of thawing. Measurement on samples after a year of storage still showed the ability to classify samples based on their freezing history. These results show that hyperspectral technology can be used to evaluate the freeze-thaw history and potentially estimate quality of cod products in a manner that meets the speed and non-invasive requirements of an industrial setting. [ABSTRACT FROM AUTHOR]

This study investigated the feasibility of hyperspectral imaging (HSI) to predict moisture content (MC) values in heated - dehydrated (H-D) and cool - air - dehydrated (C-A-D) pork samples using the calibrated models established based on fresh (F), frozen - thawed (F-T), dry - salting - dehydrated (D-S-D) and wet - salting - dehydrated (W-S-D) pork samples. The full spectra were extracted from the region of interests (ROIs) in the spectral range of 400–1000 nm and the textural variables were extracted by gray-level gradient co-occurrence matrix (GLGCM) method from the first two PC images accounting for 98.73% of the total variance. Moreover, the optimal wavelengths were selected by regression coefficients (RC). Partial least-squares regression (PLSR) predictive model was developed based on the above extracted data and their mutual combination, in which the changes can be correlated with MC - related attributes. The results demonstrated that the PLSR model based on the incorporation of the optimal wavelengths and the textures (OW-T) from F, F-T, D- S -D and W-S-D samples were the best to predict MC in H-D and C-A-D samples with R p 2 of 0.9489 and RMSEP of 1.4736. Moreover, the generated visualization maps provided a rapid way to screen the MC values unequally distributed in the whole pork samples after diverse processing conditions. Therefore, it is feasible and promising to improve the applicability of the existing MC predictive models for MC prediction in the pork samples by supplementing more samples under different treatments as another prediction sets. [ABSTRACT FROM AUTHOR]

This study was carried out to explore the potential of computer vision system (CVS) and two hyperspectral imaging (HSI) systems covering the visible and short-wave near infrared range (400–1000 nm) and the long-wave near infrared range (897–1753 nm), respectively, for differentiation of organic and conventional farm-raised salmon fillets in fresh and chill-stored conditions. Partial least squares-discriminant analysis (PLS-DA), support vector machine (SVM) and random forests (RF) classifiers were used to build classification models for recognition and authentication of the tested samples. The results suggested that hyperspectral discrimination performed much better than CVS. For the same validation set, the highest correct classification rate (CCR) equivalent to 98.2% was presented in two SVM models, one was built from full spectral variables in the 400–1000 nm and the other from four optimal wavelengths in the same spectral region. The best prediction for CVS was obtained by using PLS-DA with CCR of 83.6% for validation, while the most satisfying results for HSI in 897–1753 nm were achieved by applying SVM algorithm on full spectral region as well as 10 important wavelengths with both CCRs of 92.7% for validation. In a nutshell, hyperspectral imaging in the 400–1000 nm has presented the best predictive ability to differentiate between organic and conventional salmon fillets, while SVM classifier has been confirmed to be very powerful for multivariable analysis in our case. Overall, the outcome achieved from this research strongly suggested the capability of the hyperspectral imaging for objective and rapid categorization of the two salmon varieties under circumstances of fresh and chill-stored conditions. [ABSTRACT FROM AUTHOR]

Semi-hard cheese samples made from pasteurized 3.5% and 5% fat cow’s milk were partly enzyme treated with microbial transglutaminase (mTG). Cheese colour and yield was greatly affected by mTG at both fat levels. The cross-linking effect of mTG also led to improved protein content and hardness in cheese. As Texture Profile Analysis (TPA) could only distinguish according to enzyme treatment we studied hyperspectral imaging (HSI) to differentiate also according to fat level and ripening status. HSI was used on a weekly basis during 4–10 weeks period of ripening time. In case of uneven surface, using the spatial information and making proper normalization on the spectra of ROI pixels, the hyperspectral method can provide even better signal-to-noise ratio then the conventional spectrophotometric method. Statistical analysis with Partial Least Squares (PLS) regression and Monte Carlo Cross Validation (MCCV) led to determination of significant wavelengths both for the detection of enzyme treatment (1387 nm) and for fat content (1190, 1234 nm) independent of the actual age of examined semi-hard cheese. Hyperspectral measurement method allowed remote inspection of the product through transparent vacuum foil. The spectrum normalization allowed the elimination of effects originating from inhomogeneous illumination caused by cheese holes. [ABSTRACT FROM AUTHOR]

A rapid and effective near infrared (NIR; 900–1700 nm) hyperspectral imaging method to determine intramuscular fat (IMF) content of pork at the last 6 ribs of the round rib end was developed and tested. Pattern analysis techniques including Gabor filter, grey-level co-occurrence matrix (GLCM), and wide line detector (WLD) were applied to process hyperspectral images into data from which spectral and image features, including texture-spectral, texture and line features were extracted. Among all spectral and image features of the 6 ribs' images, the first derivative of Gabor filtered mean spectra showed the strongest correlation with IMF content, and was therefore selected as the optimal feature for prediction of IMF content at any given rib number. Multiple linear regression (MLR) was exploited to build prediction models between IMF content at rib end and at 6 different rib locations. Leave-one-out cross validation was used to test the robustness of established models. The correlation coefficients of calibration, cross validation and full validation ( R c , R cv , R fv , respectively) were used to assess model accuracy. All six MLR models showed a good performance ( R c ≥ 0.90, R cv ≥ 0.87, R fv ≥ 0.81), with that for IMF content at the 2nd/3rd last rib showing the best predictive ability ( R c = 0.96, R cv = 0.95, R fv = 0.83). The Gabor filter-MLR models were applied to each pixel in an image to successfully derive distribution maps and visualize IMF content distribution in the loins. Results demonstrated that the conventional determination of IMF content of pork at different ribs along a single Longissimus dorsi could be replaced by pattern analysis technique-processed hyperspectral images of rib ends. [ABSTRACT FROM AUTHOR]

This study aimed to develop simplified models for rapid and nondestructive monitoring TVB-N contents during drying of cured meats based on a hyperspectral imaging system (HSI) in the spectral range of 400–1000 nm. Multivariate calibration models were first developed using partial least-squares regression (PLSR) and least-squares support vector machines (LS-SVM) in the full spectral range. In order to simplify the calibration model, a set of 9 feature wavelengths was then selected using the regression coefficient and two simplified models using PLSR and multiple linear regression (MLR) were subsequently established. The best simplified model obtained was based on MLR, with an R 2 p of 0.861 and RMSEP of 4.73. Mapping of TVB-N contents was realized by transferring the quantitative model to each pixel in the image to display protein degradation in cured pork slices at different drying periods. The results provided a possibility of realizing a multispectral imaging technique for on-line monitoring the TVB-N contents of cured meats during drying processes. [ABSTRACT FROM AUTHOR]

A hyperspectral image analysis was used to characterize heat treatment in oat flour, performed by treating oat flour at 80, 100 and 130 °C for 30 min. Images from both original oat and treated flours were captured, and hyperspectral information was collected. Oat flours were used to obtain composite flours based on two different substitution levels (10 and 20%) of wheat flour. Composite breads were produced from the obtained flours. A battery of analyses was run to characterize them in terms of physical properties. The hyperspectral information of oat flours was analyzed by multivariate statistics and a pattern evolution-depending temperature was observed. Similarly, a set of the physical properties of breads was analyzed based on multivariate statistics, and a pattern of temperature-dependent evolution, in addition to the substitution level, was also recognized. Multivariate non linear regressions were done between both data sets to study their relationship and high values in the calibration and cross-validation results obtained. The changes undergone by oat flour during treatment were characterized with hyperspectral information, which could represent a non destructive monitoring tool to then regulate it until oat flours are obtained that confer composite bread adequate properties. [ABSTRACT FROM AUTHOR]

This study aimed to develop an alternative technique for rapid, accurate and non-invasive determination of gross energy density values of salmon fillets based on hyperspectral imaging (900–1700 nm). Spectral data were extracted from the hyperspectral images with outlier pixels removed. Good performances were achieved with r P of 0.906 and 0.909, and normalized RMSEP of 6.768% and 7.064% for partial least squares regression (PLSR) and epsilon-support vector regression (epsilon-SVR) analysis, respectively. The optimized stepwise-PLS model built with four wavelengths (931, 1001, 1135 and 1168 nm) yielded good results with r P of 0.908 and normalized RMSEP of 6.871%. The distribution maps for visualizing energy density indices in all portions of the salmon fillet were subsequently generated. The overall results confirmed the successful development of a rapid and non-destructive technique using hyperspectral imaging for determining energy value of salmon fillets, and the provision of an alternative method for caloric density measurement for the food industry. [ABSTRACT FROM AUTHOR]

The uniqueness and consistency of commercial food and beverage brands are critically important for their marketability. Thus, it is important to develop quality control tools and measures, so that both companies and consumers can monitor whether a given food product or beverage meets certain quality expectations and/or is consistent when purchased at different times or at different locations. In this study, we characterized the consistency (levels of extractable protein and reducing sugars) of 15 brands of roasted coffee beans, which were obtained from a supermarket at two dates about six months apart. Coffee brands varied markedly in extractable protein and reducing sugar contents between dates, and also within and among roasting classes (light, medium, medium-dark, and dark roasts). We acquired hyperspectral imaging data (selected bands out of 220 narrow spectral bands from 408 nm to 1008 nm) from ground samples of the roasted coffee beans, and reflectance-based classification of roasting classes was associated with fairly low accuracy. We provide evidence that the combination of hyperspectral imaging and a general quality indicator (such as extractable protein content) can be used to monitor brand consistency and quality control. We demonstrated that a non-destructive method, potentially real-time and automated, and quantitative method can be used to monitor the consistency of a highly complex beverage product. We believe the results from this study of brand consistency are not only of relevance to the coffee industry but to a wide range of commercial food and beverage brands. [ABSTRACT FROM AUTHOR]

Since 2008, the detection of the adulterant melamine (2,4,6-triamino-1,3,5-triazine) in food products has become the subject of research due to several food safety scares. Near-infrared (NIR) hyperspectral imaging offers great potential for food safety and quality research because it combines the features of vibrational spectroscopy and digital imaging. In this study, NIR hyperspectral imaging was investigated for quantitative evaluation of melamine particles in nonfat and whole milk powders. Melamine was mixed into milk powders in a concentration range of 0.02–1.00% (w/w). A NIR hyperspectral imaging system was used to acquire images (938–1654 nm) of melamine powder, whole milk powder, nonfat milk powder, and mixtures of melamine and each of the milk powders. Two optimal bands (1447 nm and 1466 nm) were selected by a linear correlation algorithm with pure milk and pure melamine. Band ratio (B 1447/1466 ) images coupled with a single threshold were used to create resultant images to visualize identification and distribution of the melamine adulterant particles in milk powders. The identification results were verified by spectral feature comparison between separated mean spectra of melamine pixels and milk pixels. Linear correlations (r) were found between the number of pixels identified as containing melamine and melamine concentration in nonfat milk and whole milk powders, which were 0.980 and 0.970 or higher, respectively. The study demonstrated that the combination of NIR hyperspectral imaging and simple band ratioing was promising for rapid quantitative analysis of melamine in milk powders. [ABSTRACT FROM AUTHOR]

A hyperspectral imaging system covering two spectral ranges (380–1030 nm and 874–1734 nm) was applied to evaluate strawberry ripeness. The spectral data were extracted from hyperspectral images of ripe, mid-ripe and unripe strawberries. The optimal wavelengths were obtained from spectra of 441.1–1013.97 and 941.46–1578.13 nm by loadings of principal component analysis (PCA). Pattern texture features (correlation, contrast, entropy and homogeneity) were extracted from the images at optimal wavelengths. Support vector machine (SVM) was used to build classification models on full spectral data, optimal wavelengths, texture features and the combined dataset of optimal wavelengths and texture features, respectively. SVM models using combined datasets performed best among all datasets. SVM models using datasets from hyperspectral images at 441.1–1013.97 nm performed better with classification accuracy over 85%. The overall results indicated that hyperspectral imaging could be used for strawberry ripeness evaluation, and data fusion combining spectral information and spatial information showed advantages in strawberry ripeness evaluation. [ABSTRACT FROM AUTHOR]

Total acid content (TAC) and moisture content (MC) are very important parameters during Solid-State Fermentation (SSF) processes. The feasibility of using hyperspectral imaging (HSI) technology for predicting TAC and MC in vinegar cultures during SSF processes was investigated. Prediction models were constructed using variables selected from spectral and spatial data from associated 3-D hyperspectral datacubes to predict the relative content of TAC and MC for each pixel in the hyperspectral image. Models were developed using genetic algorithm (GA) optimization combined with partial least squares regression (PLS) dependent on the spectral variables yielded good prediction results for both TAC and MC. The determination coefficients ( R p 2 ) for TAC and MC were 0.8565 and 0.8162, respectively. Finally, the distribution maps of TAC and MC for a vinegar culture sample were obtained. These distribution maps could be implemented to estimate the uniformity of fermentation products during SSF. [ABSTRACT FROM AUTHOR]

Lamb muscle discrimination is important for the meat industry due to the different pricing of each type of muscle. In this paper, we combine hyperspectral imaging, operating in the wavelength range 380–1028 nm, with several machine learning algorithms to deal automatically with the classification of lamb muscles. More specifically, we study the discrimination of four different lamb muscles, namely, Longissimus dorsi, Psoas major, Semimembranosus and Semitendinosus from thirty lambs of Churra Galega Mirandesa breed. The objective of the paper is to determine the best method for muscle classification. In the experimental study we report an analysis of the performance of seven classifiers. We study their behavior when they are applied over the original data as well as over the data pre-processed using Principal Component Analysis (PCA) to reduce the dimensionality of the problem. The seven classifiers used to differentiate the muscle types are two Artificial Neural Networks, namely the linear Least Mean Squares (LMS) classifier and the Multilayer Perceptron with Scaled Conjugate Gradient (MLP-SCG), two Support Vector Machines (SVM), namely the ν SVM and the SVM trained with Sequential Minimal Optimization (SMO), the Logistic Regression (LR), the Center Based Nearest Neighbor classifier and the Linear Discriminant Analysis. The best result, determined using a leave-one-animal-out scheme, is provided by the linear LMS classifier using the original data, since it correctly classifies 96.67% of the samples. The LR, the MLP-SCG using original data and the SVM trained with SMO on data preprocessed with PCA are also suitable techniques to tackle the lamb muscle classification problem. [ABSTRACT FROM AUTHOR]

The main objective of this study was to evaluate the potential of visible near-infrared (VNIR) hyperspectral imaging (400–1000 nm) and machine learning to detect adulteration in fresh minced beef with chicken. Minced beef samples were adulterated with minced chicken in the range 0–50% (w/w) at approximately 2% intervals. Hyperspectral images were acquired in the reflectance (R) mode and then transformed into absorbance (A) and Kubelka–Munck (KM) units. Partial least squares regression (PLSR) models were developed to relate the three spectral profiles with the adulteration levels of the tested samples. These models were then validated using different independent data sets, and obtained the coefficient of determination (R 2 p ) of 0.97, 0.97, and 0.96 with root mean square error in prediction (RMSEP) of 2.62, 2.45, and 3.18% (w/w) for R, A and KM spectra, respectively. To reduce the high dimensionality of the hyperspectral data, some important wavelengths were selected using stepwise regression. PLSR models were again created using these important wavelengths and the best model was then transferred in each pixel in the image to obtain prediction map. The results clearly ascertain that hyperspectral imaging coupled with machine learning can be used to detect, quantify and visualize the amount of chicken adulterant added to the minced beef. [ABSTRACT FROM AUTHOR]

A prototype on-line hyperspectral imaging system (λ = 400–1000 nm ) was developed and used to acquire images of exposed ribeye muscle on hanging beef carcasses (n = 274) at 2-day postmortem in a commercial beef packing plant. After image acquisition, a strip steak was cut from each carcass and vacuum packaged. After aging for 14 days, the steaks were cooked and Warner-Bratzler shear force values were collected as a measure of tenderness. Four different principal component analysis-based dimensionality reduction methods were implemented to reduce information redundancy in beef hyperspectral images. Textural features extracted from the 2-day hyperspectral images were modeled using Fisher's linear discriminant (FLD), support vector machines (SVM), and decision tree (DT) models to predict 14-day aged, cooked beef tenderness. Based on a true validation procedure using 101 samples, the FLD model yielded a tender certification accuracy of 86.7%. In addition, wavelengths corresponding to myoglobin and its derivatives (541, 577, and 635 nm ), beef aging (541, 577, 635, 756, and 980 nm ), protein (910 nm ), fat (928 nm ), and water (739, 756, and 988 nm ) were identified. [ABSTRACT FROM AUTHOR]

This study investigated the potential of using hyperspectral imaging technology (900–1700 nm) to determine the thiobarbituric acid (TBA) value and pH for evaluation of lipid oxidation in Atlantic salmon ( Salmo salar ) fillets during cold storage for 0, 3, 6, 9 and 12 days at 1 ± 1 °C. Good results were achieved for both parameters by using partial least square regression (PLSR) calibration models with full spectral region. Two simplified models were then built by using forward stepwise-multiple linear regression (MLR) variable selection method to select 18 and 10 most important wavelengths for TBA value and pH, respectively. The optimised stepwise-MLR model for TBA value yielded satisfactory results with correlation coefficient ( r C ) of 0.921 and root mean square error of calibration (RMSEC) of 1.840 μmol MDA/kg fish. This model was used to visualise the TBA value distributions during different storage days. Further improvements were achieved by applying particle analysis on the images to extract only the spectra from white stripes in salmon fillet. When in tandem with detrend pre-processing technique, the calibration model based on particle analysis demonstrated the best performance for TBA value prediction ( r C = 0.957 and RMSEC = 1.449 μmol MDA/kg). In addition, a novel chemometric strategy accomplished by the use of hyperspectrograms was also proposed in this work and satisfactory results were obtained. The overall results confirmed the capability of near-infrared hyperspectral imaging as a rapid and non-invasive technique to monitor lipid oxidation in salmon fillets. [ABSTRACT FROM AUTHOR]

This review covers the application of backscattering imaging as a non-invasive technique for monitoring the quality of agricultural and food products. The review enumerates and discusses the concepts and various applications of laser light backscattering imaging (LLBI), multispectral laser backscattering imaging (MBI) and hyperspectral laser backscattering imaging (HBI). All the methods make use of laser light which varies in spectrum from visible up to near-infrared to detect changes in the quality of fresh produce. Emphasis is placed on applications which demonstrate promising potential for agricultural and food applications under various conditions. A critical review of the limitations is also given. [ABSTRACT FROM AUTHOR]

The fungi or moldy peanuts have high possibility containing the potent carcinogen. And the risk of human ingesting toxic carcinogen from the moldy peanuts can be reduced if the moldy peanuts can be efficiently identified and separated from healthy ones before entering the food chain. The object of this study mainly focuses on how to identify the moldy peanuts by using Near-Infrared (NIR) hyperspectral images. NIR hyperspectral images were acquired at the wavelength range between 970 and 2570 nm. The method of Principle Component Analysis (PCA) was mainly used in the spectral dimension to select sensitive bands, and to project the spectral vector in the direction that is favorable to identify the moldy information. Meanwhile, the marker-controlled watershed algorithm was adopted to segment the images into kernel-scale objects in spatial dimensions. Finally, the results both from PCA and segmentation were combined to judge whether the peanut kernels were moldy or not via the thresholds. The results illustrated the proposed method could be better used to identify the moldy kernels with accuracy of 87.14% in learning image and accuracy of 98.73% in validation image. [ABSTRACT FROM AUTHOR]

In recent years, Independent Components Analysis (ICA) has proven itself to be a powerful signal-processing technique for solving the Blind-Source Separation (BSS) problems in different scientific domains. In the present work, an application of ICA for processing NIR hyperspectral images to detect traces of peanut in wheat flour is presented. Processing was performed without a priori knowledge of the chemical composition of the two food materials. The aim was to extract the source signals of the different chemical components from the initial data set and to use them in order to determine the distribution of peanut traces in the hyperspectral images. To determine the optimal number of independent component to be extracted, the Random ICA by blocks method was used. This method is based on the repeated calculation of several models using an increasing number of independent components after randomly segmenting the matrix data into two blocks and then calculating the correlations between the signals extracted from the two blocks. The extracted ICA signals were interpreted and their ability to classify peanut and wheat flour was studied. Finally, all the extracted ICs were used to construct a single synthetic signal that could be used directly with the hyperspectral images to enhance the contrast between the peanut and the wheat flours in a real multi-use industrial environment. Furthermore, feature extraction methods (connected components labelling algorithm followed by flood fill method to extract object contours) were applied in order to target the spatial location of the presence of peanut traces. A good visualization of the distributions of peanut traces was thus obtained. [ABSTRACT FROM AUTHOR]

The feasibility of detecting Aflatoxin B 1 (AFB 1 ) in single maize kernel inoculated with Aspergillus flavus conidia in the field, as well as its spatial distribution in the kernels, was assessed using near-infrared hyperspectral imaging (HSI) technique. Firstly, an image mask was applied to a pixel-based image mosaic to remove background and shading. Secondly, bad lines in spectra imaging caused by inherent defects of Mercury Cadmium Telluride (MCT) detector were removed through an interactive analysis based on principal component analysis (PCA). Then a PCA procedure was carried out again on the cleaned image, key wavelengths such as 1729 and 2344 nm were shown clearly from the loading line plot of the seventh principal component (PC 7 ). And the pixel of AFB 1 extracted from the 5-dimensional scatter plot space formed by five principal components (PCs) from PC 4 to PC 8 (especially PC 7 and PC 5 ) were taken as the input of the spectral angle mapper (SAM) classifier, accuracies of the three varieties of kernels reached 96.15%, 80%, and 82.61% respectively if kernels containing either high (⩾100 ppb) or low (<10 ppb) levels of aflatoxin. A slightly better test result could be got if the kernels placed with different germ orientation. Finally, the repeatability was verified using the fourth variety of kernels. [ABSTRACT FROM AUTHOR]

The objectives of this study were to investigate the feasibility of hyperspectral scattering imaging to predict the bacterial contamination in meat nondestructively, and propose an optimal approach for detecting low levels of total viable count (TVC) contamination in beef. Fresh beef samples were obtained from a commercial slaughtering plant, and stored at 4 °C for 0–12 days. The visible/near-infrared (VIS/NIR) hyperspectral images in the backscattering mode were acquired from 3–5 beef samples on each day of the experiment, in parallel with microbiological analysis to enumerate the TVC population. Lorentzian function was used to resolve the light scattering information within the hyperspectral image and consequently Lorentzian parameters, which represented different hyperspectral scattering characteristics were extracted. In this study, not only the individual Lorentzian parameters but also the parameter combinations were used to establish the multivariate statistical models for predicting beef TVC, based on the modeling methods of principal component regression (PCR), partial least squares regression (PLSR), and back propagation neural network (BPNN), respectively. The models established using individual Lorentzian parameters did not perform well in predicting low levels of TVC contamination in beef, and the best prediction result could only achieved with the correlation coefficient of prediction set ( R P ) and root mean squared error of prediction set (RMSEP) of 0.81 and 1.27 log CFU/g, respectively. Based on the parameter combinations, the best modeling results were achieved with R P and RMSEP of 0.86 and 0.93 log CFU/g, 0.87 and 0.79 log CFU/g, 0.90 and 0.88 log CFU/g by PCR, PLSR, and BPNN methods, respectively, which confirmed the superiority of the parameter combination method. The results of this study demonstrated for the first time that hyperspectral scattering imaging combined with Lorentzian function and the proposed parameter combination method could be used to detect low levels of bacterial contamination in beef nondestructively. [ABSTRACT FROM AUTHOR]

The investigation of visible and near infrared hyperspectral imaging (400–1000 nm) coupled with classifiers and spectral pre-processing techniques was conducted to discriminate fresh from cold-stored (4 °C for 7 days) and frozen-thawed (−20 °C and −40 °C for 30 days) grass carp fish fillets. Four classifiers with four spectral pretreatment methods were applied to establish the classification models. Compared with the original models established using the full wavelengths, the classification models with three classifiers of soft independent modeling of class analogy (SIMCA), least squares-support vector machine (LS-SVM) and probabilistic neural network (PNN) in tandem with the first derivative pretreatment showed the best classification performance and the highest correct classification rate (CCR) of 94.29%. In addition, in order to reduce the high dimensionality of hyperspectral images, seven optimal wavelengths were selected by successive projections algorithm (SPA) and used to simplify the classification models. The simplified model obtained by the LS-SVM classifier coupled with the first derivative pre-processing method also presented good prediction accuracy with the CCR of 91.43%. The results demonstrated that the integration of hyperspectral imaging and classifiers analysis had a great potential for on-line detection and was feasible to rapidly and non-invasively discriminate fresh and frozen-thawed fish fillets. [ABSTRACT FROM AUTHOR]

The study develops a nondestructive test based on hyperspectral imaging using a combination of existing analytical techniques to determine the internal quality of eggs, including freshness, bubble formation or scattered yolk. Successive projections algorithm (SPA) combined with support vector regression established a freshness detection model, which achieved a determination coefficient of 0.87, a root mean squared error of 4.01%, and the ratio of prediction to deviation of 2.80 in the validation set. In addition, eggs with internal bubbles and scattered yolk could be discriminated by support vector classification (SVC) model with identification accuracy of 90.0% and 96.3% respectively. Our findings suggest that hyperspectral imaging can be useful to non-destructively and rapidly assess egg internal quality. [ABSTRACT FROM AUTHOR]

A prototype on-line acousto-optic tunable filter (AOTF)-based hyperspectral image acquisition system ( λ = 450–900 nm) was developed for tenderness assessment of beef carcasses. Hyperspectral images of ribeye muscle on stationary hanging beef carcasses ( n = 338) at 2-day postmortem were acquired in commercial beef slaughter or packing plants. After image acquisition, a strip steak was cut from each carcass, vacuum packaged, aged for 14 days, cooked, and slice shear force tenderness scores were collected by an independent lab. Beef hyperspectral images were mosaicked together and principal component (PC) analysis was conducted to reduce the spectral dimension. Six different textural feature sets were extracted from the PC images and used in Fisher’s linear discriminant model to classify beef samples into two tenderness categories: tender and tough. The pooled feature model performed better than the other models with a tender certification accuracy of 92.9% and 87.8% in cross-validation and third-party true validation, respectively. Two additional metrics namely overall accuracy and a custom defined metric called accuracy index, were used to compare the tenderness prediction models. [ABSTRACT FROM AUTHOR]

Hyperspectral imaging technology is increasingly regarded as a powerful tool for the classification and spatial quantification of a wide range of agrofood product properties. Taking into account the difficulties involved in validating hyperspectral calibrations, the models constructed here proved moderately accurate for predicting NaCl content, while good accuracy was obtained for moisture content (Shenk and Westerhouse, 1996) in Iberian dry-cured ham slices. Using spatial information provided by chemical imaging, the distribution of both analytes was characterized by textural analysis as a function of muscle and commercial category. [ABSTRACT FROM AUTHOR]