Hyperspectral imaging and machine learning in food microbiology: Developments and challenges in detection of bacterial, fungal, and viral contaminants.

Hyperspectral imaging (HSI) is a robust and nondestructive method that can detect foreign particles such as microbial, chemical, and physical contamination in food. This review summarizes the work done in the last two decades in this field with a highlight on challenges, risks, and research gaps. Considering the challenges of using HSI on complex matrices like food (e.g., the confounding and masking effects of background signals), application of machine learning and modeling approaches that have been successful in achieving better accuracy as well as increasing the detection limit have also been discussed here. Foodborne microbial contaminants such as bacteria, fungi, viruses, yeast, and protozoa are of interest and concern to food manufacturers due to the potential risk of either food poisoning or food spoilage. Detection of these contaminants using fast and efficient methods would not only prevent outbreaks and recalls but will also increase consumer acceptance and demand for shelf-stable food products. The conventional culture-based methods for microbial detection are time and labor-intensive, whereas hyperspectral imaging (HSI) is robust, nondestructive with minimum sample preparation, and has gained significant attention due to its rapid approach to detection of microbial contaminants. This review is a comprehensive summary of the detection of bacterial, viral, and fungal contaminants in food with detailed emphasis on the specific modeling and datamining approaches used to overcome the specific challenges associated with background and data complexity.

Anisotropic light propagation in bovine muscle tissue depends on the initial fiber orientation, muscle type and wavelength.

The effects of fiber orientation on vis/NIR light propagation were studied in three bovine muscles: biceps brachii, brachialis and soleus. Broadband light was focused onto the sample and the diffuse reflectance spot was captured using a hyperspectral camera (470-1620 nm), after which rhombuses were fitted to equi-intensity points. In samples with fibers running parallel to the measurement surface, the rhombus' major axis was oriented perpendicular to the fiber direction close to the point of illumination. However, at larger distances from the illumination spot, the major axis orientation aligned with the fiber direction. This phenomenon was found to be muscle dependent. Furthermore, the rhombus orientation was highly dependent on the sample positioning underneath the camera, especially when the muscle fibers ran parallel to the measurement surface. The bias parameter, indicating the deviation from a circular shape, was higher for samples with the fibers running parallel to the measurement surface. Moreover, clear effects of wavelength and distance from the illumination point on this parameter were observed. These results show the importance of fiber orientation when considering optical techniques for measurements on anisotropic, fibrous tissues. Moreover, the prediction of muscle fiber orientation seemed feasible, which can be of interest to the meat industry.

Towards the non-invasive assessment of staling in bovine hides with hyperspectral imaging.

Microbial spoilage or staling of bovine hides during storage leads to poor leather quality and increased chemical consumption during processing. Conventional microbiological examinations of hide samples which require time-consuming microbe culture cannot be employed as a practical staling detection approach for leather production. Hyperspectral imaging (HSI), featuring fast data acquisition and implementation flexibility has been considered ideal for in-line detection of microbial contamination in Agri- food products. In this study, a linescan hyperspectral imaging system working in a spectral range of 550 nm to 1700 nm was utilized as a rapid and non-destructive technique for predicting the aerobic plate counts (APC) on raw hide samples during storage. Fresh bovine hide samples were stored at 4 °C and 20 °C for 3 days. Every day, hyperspectral images were acquired on both sides for each sample. The APCs were determined simultaneously by conventional microbiological plating method. Leather quality was evaluated by microscopic inspection of grain surfaces, which indicate the acceptable threshold of microbe load on hide samples for leather processing. Partial least squares regression (PLSR) was applied to fit the spectral information extracted from the samples to the logarithmic values of APC to develop microbe load prediction models. All models showed good prediction accuracy, yielding a Rcv2 in the range of 0.74-0.92 and standard error of cross validation (SECV) in the range of 0.61-0.76 %. The prediction capability of the HSI was explored using the model developed with SNV + smoothened pre-processing to spatially predict plate count in the samples. Models established in this study successfully predicted the staling states characterised by bacterial loads on hide samples with low prediction errors. Models, visually, showed the differences in microbial load across the storage time and temperatures. Results illustrate that HSI can be potentially implemented as a non-invasive tool to predict microbe loads in bovine hides before leather processing, so that real-time grading of hides based on staling states can be achieved. This will reduce the cost of leather production and waste management and pave the way for allocating material supply for different production purposes.

Internal reflectance cell fluorescence measurement combined with multi-way analysis to detect fluorescence signatures of undiluted honeys and a fusion of fluorescence and NIR to enhance predictability.

Honey is a complex food matrix that contains diverse polyphenolic compounds. Some phenolics exhibit fluorescence signatures which can be used to evaluate honey quality, and authenticity and to determine botanical origin. Manuka honey contains two unique fluorescence markers: Leptosperin (MM1) and LepteridineTM (MM2) that are derived from Leptospermum scoparium nectar. Fluorescence measurement of supersaturated solutions such as undiluted honeys can be challenged by complex inner filter effects. The current study shows the ability of internal reflectance cell fluorescence measurement and multi-way analysis to detect fluorophores in undiluted honeys. This study scanned honeys from different geographic districts generating excitation emission matrices (250-400/300-600 nm), and by near infrared (NIR) hyperspectral camera (547-1701 nm). PARAFAC and tri-PLS could track two fluorescence markers: MM1 (R2 = 0.82 & RMSEP = 138.65) and MM2 (R2 = 0.82 & RMSEP = 2.75) from undiluted honey fluorescence data with > 80 % accuracy. Classification of mono-floral, multi-floral and non-manuka honeys achieved 90 % overall accuracy. Fusion of fluorescence data at ƛex 270 & 330 nm and NIR hyperspectral data combined with multi-block PLS analysis enhances predictability of fluorescence markers further. The study revealed the potential of internal reflectance cell fluorescence measurement combined with chemometrics and data fusion for rapid evaluation of honey quality and botanical origin.

Support vector regression for prediction of stable isotopes and trace elements using hyperspectral imaging on coffee for origin verification.

The potential of using rapid and non-destructive near-infrared - hyperspectral imaging (HSI-NIR) for the prediction of an integrated stable isotope and multi-element dataset was explored for the first time with the help of support vector regression. Speciality green coffee beans sourced from three continents, eight countries, and 22 regions were analysed using a push-broom HSI-NIR (700-1700 nm), together with five isotope ratios (δ13C, δ15N, δ18O, δ2H, and δ34S) and 41 trace elements. Support vector regression with the radial basis function kernel was conducted using X as the HSI-NIR data and Y as the geochemistry markers. Model performance was evaluated using root mean squared error, coefficient of determination, and mean absolute error. Three isotope ratios (δ18O, δ2H, and δ34S) and eight elements (Zn, Mn, Ni, Mo, Cs, Co, Cd, and La) had an R2predicted 0.70 - 0.99 across all origin scales (continent, country, region). All five isotope ratios were well predicted at the country and regional levels. The wavelength regions contributing the most towards each prediction model were highlighted, including a discussion of the correlations across all geochemical parameters. This study demonstrates the feasibility of using HSI-NIR as a rapid and non-destructive method to estimate traditional geochemistry parameters, some of which are origin-discriminating variables related to altitude, temperature, and rainfall differences across origins.

Hyperspectral imaging through vacuum packaging for monitoring cheese biochemical transformation caused by Clostridium metabolism.

This study developed a novel method for monitoring cheese contamination with Clostridium spores non-invasively using hyperspectral imaging (HSI). The ability of HSI to quantify Clostridium metabolites was investigated with control cheese and cheese manufactured with milk contaminated with Clostridium tyrobutyricum, Clostridium butyricum and Clostridium sporogenes. Microbial count, HSI and SPME-GC-MS data were obtained over 10 weeks of storage. The developed method using HSI successfully quantified butyric acid (R2 = 0.91, RPD = 3.38) a major compound of Clostridium metabolism in cheese. This study creates a new venue to monitor the spatial and temporal development of late blowing defect (LBD) in cheese using fast and non-invasive measurement.

A critical review of social media research in sensory-consumer science.

The collection and analysis of digital data from social media is a rapidly growing methodology in sensory-consumer science, with a wide range of applications for research studying consumer attitudes, preferences, and sensory responses to food. The aim of this review article was to critically evaluate the potential of social media research in sensory-consumer science with a focus on advantages and disadvantages. This review began with an exploration into different sources of social media data and the process by which data from social media is collected, cleaned, and analyzed through natural language processing for sensory-consumer research. It then investigated in detail the differences between social media-based and conventional methodologies, in terms of context, sources of bias, the size of data sets, measurement differences, and ethics. Findings showed participant biases are more difficult to control using social media approaches, and precision is inferior to conventional methods. However, findings also showed social media methodologies may have other advantages including an increased ability to investigate trends over time and easier access to cross-cultural or global insights. Greater research in this space will identify when social media can best function as an alternative to conventional methods, and/or provide valuable complementary information.

Comparative evaluation of miniaturized and conventional NIR spectrophotometer for estimation of fatty acids in cheeses.

The miniaturization of near-infrared spectrometers has been growing rapidly. Several designs are now available, but there is a lack of understanding of how spectral data from these designs are affected by complex matrices and what are the limitations when compared to established systems. This study compares a popular miniaturized NIR device based on Hadamard-transform spectrometer (named miniaturized NIR) with a system based on dispersive spectrometer (named handheld-NIR) to assess: 1) their predictive performance; 2) the effect of a complex matrix on the performance, and 3) ability to discriminate multiples compounds in that matrix. The devices were challenged with a wide range of cheese types (n = 36) from different species (cow, goat, ewe and buffalo), brands (n = 30), countries of origin (n = 9) and with a broad range of cheese matrices (soft, fresh, semi-hard, hard and aged) to predict fat composition. Spectra were collected non-invasively with no sample preparation. Three wavelength ranges from handheld NIR were compared to miniaturized NIR based on two modelling approaches were used: a linear (Partial Least Square - PLS) and a non-linear (Support Vector Machine - SVM). The important wavelengths for each model were identified and used to assess the ability of the spectral data to differentiate among fatty acids. The highest prediction performance was observed for saturated fatty acids (C4.0, C14.0, C15.0 C16.0, total SCF and total SFA) with the RPDEXT-VAL for the external validation dataset presenting values higher than 3 and the coefficient of determination for the external validation dataset (R2EXT-VAL) higher than 0.89, mostly for SVM models. The sum of fatty acids also shows good prediction performance with RPDEXT-VAL higher than 3 and R2EXT-VAL higher than 0.89. Models with RPDEXT-VAL between 2 and 3 includes: C6.0; C17.0; C18.0; C10.1; C16.1; C17.1; iso.C15.0; iso.C.16; iso.C17; C18.1.c11; C18.1.c9; anteiso C17; total MUFA; and total BCFA. The cheese matrix affected the linearity between spectral data and fatty acids concentration requiring a more complex model (SVM), but this effect was not enhanced by the instrument type. It was shown that the spectral information allows discrimination among fatty acids and this ability was not affected by the type of instrument. These findings demonstrated that the miniaturized NIR can be directly applied to a cheese matrix to monitor fatty acid composition with results equivalent to an optical-based design.

Hyperspectral imaging and deep learning for quantification of Clostridium sporogenes spores in food products using 1D- convolutional neural networks and random forest model.

Clostridium sporogenes spores are used as surrogates for Clostridium botulinum, to verify thermal exposure and lethality in sterilization regimes by food industries. Conventional methods to detect spores are time-consuming and labour intensive. The objectives of this study were to evaluate the feasibility of using hyperspectral imaging (HSI) and deep learning approaches, firstly to identify dead and live forms of C. sporogenes spores and secondly, to estimate the concentration of spores on culture media plates and ready-to-eat mashed potato (food matrix). C. sporogenes spores were inoculated by either spread plating or drop plating on sheep blood agar (SBA) and tryptic soy agar (TSA) plates and by spread plating on the surface of mashed potato. Reflectance in the spectral range of 547-1701 nm from the region of interest was used for principal component analysis (PCA). PCA was successful in distinguishing dead and live spores and different levels of inoculum (102 to 106 CFU/ml) on both TSA and SBA plates, however, was not efficient on the mashed potato (food matrix). Hence, deep learning classification frameworks namely 1D- convolutional neural networks (CNN) and random forest (RF) model were used. CNN model outperformed the RF model and the accuracy for quantification of spores was improved by 4% and 8% in the presence and absence, respectively of dead spores. The screening system used in this study was a combination of HSI and deep learning modelling, which resulted in an overall accuracy of 90-94% when the dead/inactivated spores were present and absent, respectively. The only discrepancy detected was during the prediction of samples with low inoculum levels (<102 CFU/ml). In summary, it was evident that HSI in combination with a deep learning approach showed immense potential as a tool to detect and quantify spores on nutrient media as well as on specific food matrix (mashed potato). However, the presence of dead spores in any sample is postulated to affect the accuracy and would need replicates and confirmatory assays.

Preparation of drip samples from leg of lamb with extended shelf life for nuclear magnetic resonance metabolomics studies.

Metabolomics has been used for the analysis of meat samples for different applications. Using drip as a proxy for meat could offer an easy and non-invasive way of sampling meat, yielding a homogenous liquid sample easy to prepare for metabolomics analysis. There is currently no standard method for the preparation of drip samples for quantitative metabolomics. The aim of this study was to evaluate six different sample preparation methods for quantitative Nuclear Magnetic Resonance (NMR) metabolomics analysis of drip from a lamb leg with extended shelf life: centrifugation, ultrafiltration, and solvent precipitation using four different solvents or solvent mixtures. The six methods were evaluated based on protein removal efficiency, ability to quantify metabolites, metabolite concentrations, reproducibility, speed and relative cost. Three methods (ultrafiltration, solvent precipitation with either acetonitrile/acetone/methanol or chloroform/methanol) resulted in excellent protein removal, high concentrations of metabolites and high reproducibility and are therefore recommended for preparation of extended shelf life lamb leg drip samples for NMR metabolomics.

Identification of Cold Spots Using Non-Destructive Hyperspectral Imaging Technology in Model Food Processed by Coaxially Induced Microwave Pasteurization and Sterilization.

The model food in this study known as mashed potato consisted of ribose (1.0%) and lysine (0.5%) to induce browning via Maillard reaction products. Mashed potato was processed by Coaxially Induced Microwave Pasteurization and Sterilization (CiMPAS) regime to generate an F0 of 6-8 min and analysis of the post-processed food was done in two ways, which included by measuring the color changes and using hyperspectral data acquisition. For visualizing the spectra of each tray in comparison with the control sample (raw mashed-potato), the mean spectrum (i.e., mean of region of interest) of each tray, as well as the control sample, was extracted and then fed to the fitted principal component analysis model and the results coincided with those post hoc analysis of the average reflectance values. Despite the presence of a visual difference in browning, the Lightness (L) values were not significantly (p < 0.05) different to detect a cold spot among a range of 12 processed samples. At the same time, hyperspectral imaging could identify the colder trays among the 12 samples from one batch of microwave sterilization.

Identification of Weeds Based on Hyperspectral Imaging and Machine Learning.

Weeds can be major environmental and economic burdens in New Zealand. Traditional methods of weed control including manual and chemical approaches can be time consuming and costly. Some chemical herbicides may have negative environmental and human health impacts. One of the proposed important steps for providing alternatives to these traditional approaches is the automated identification and mapping of weeds. We used hyperspectral imaging data and machine learning to explore the possibility of fast, accurate and automated discrimination of weeds in pastures where ryegrass and clovers are the sown species. Hyperspectral images from two grasses (Setaria pumila [yellow bristle grass] and Stipa arundinacea [wind grass]) and two broad leaf weed species (Ranunculus acris [giant buttercup] and Cirsium arvense [Californian thistle]) were acquired and pre-processed using the standard normal variate method. We trained three classification models, namely partial least squares-discriminant analysis, support vector machine, and Multilayer Perceptron (MLP) using whole plant averaged (Av) spectra and superpixels (Sp) averaged spectra from each weed sample. All three classification models showed repeatable identification of four weeds using both Av and Sp spectra with a range of overall accuracy of 70-100%. However, MLP based on the Sp method produced the most reliable and robust prediction result (89.1% accuracy). Four significant spectral regions were found as highly informative for characterizing the four weed species and could form the basis for a rapid and efficient methodology for identifying weeds in ryegrass/clover pastures.

Investigation of metabolites associated with confinement odour in chilled vacuum-packed lamb by proton nuclear magnetic resonance (1H NMR) spectroscopy.

This study aimed to determine the profiles of water-soluble metabolites in lamb drip and meat by Nuclear Magnetic Resonance (NMR) spectroscopy, in order to better understand the confinement odour (CO) phenomenon in lamb meat on a molecular level. Thirty-five lamb legs were obtained from two New Zealand meat processing plants and stored for 11 to 13 weeks at temperatures ranging from -1.5 °C to +4.0 °C. A sensorial test classified meat samples as having CO, no odour (NO) or persistent odour (PO). Sixty-three and sixty-two metabolites were identified and quantified in drip and meat samples, respectively. Partial least squares canonical analysis (PLS-CA) showed that CO was correlated with meat and drip metabolites tyramine, formate, alanine, carnosine, urea, proline, aspartate, glutathione and nicotinate. CO was also positively associated with appearance and bloom, but not directly associated with pH, size of the bacterial population or with processing plant. Metabolites associated with CO/PO are substrates or products of glucose fermentation and amino acid catabolism.

Characterization of chicken muscle disorders through metabolomics, pathway analysis, and water relaxometry: a pilot study.

Metabolite profiles of chicken breast extracts and water mobility in breasts were studied using proton nuclear magnetic resonance (1H-NMR) spectroscopy and time-domain NMR (TD-NMR) relaxometry, respectively, using normal breast (NB), and wooden breast (WB) and white striping (WS) myopathies in broilers. One thousand eight hundred sixty broilers were raised to commercial standards, receiving the same diets that were formulated as per the different growth stages. At 49 D of age, 200 animals were slaughtered following routine commercial procedures, and at 4 h postmortem, the whole breast (pectoralis major muscle) was removed and visually inspected by an experienced meat inspector who selected NB (without myopathies) and samples with the presence of WS and WB myopathies. Fifteen breasts (5 each of NB, WS, and WB) were analyzed through TD-NMR relaxometry, and samples of approximately 20 g were taken from each breast and frozen at -80°C for metabolite profiling through 1H-NMR spectroscopy. Multivariate statistical analysis was used to evaluate the effect on water relaxometry and metabolite profile in accordance with the presence and type of myopathy in the breast. 1H-NMR data showed that the metabolite profiles in WS and WB breasts were different from each other and from NB. This pilot study shows that myopathies appear to be related to hypoxia, connective tissue deposition, lower mitochondrial function, and greater oxidative stress compared with NB. The longitudinal and transverse relaxation time of the breasts determined by TD-NMR relaxometry was shorter for NB than that for WS and WB, indicating greater water mobility in breasts affected by myopathies. 1H-NMR spectroscopy can be used to differentiate the metabolism of WS, WB, and NB, and TD-NMR has the potential to be a fast, simple, and noninvasive method to distinguish NB from WB and WS. As a practical application, the metabolomic profile as per the occurrence of breast myopathies may be used for a better understanding of these issues, which opens a gap to mitigate the incidence and severity of WS and WB. In addition, the present study brings an opportunity for the development of a new and objective tool to classify the incidence of breast myopathies through TD-NMR relaxometry.

Comprehensive Evaluation of Parameters Affecting One-Step Method for Quantitative Analysis of Fatty Acids in Meat.

Despite various direct transmethylation methods having been published and applied to analysis of meat fatty acid (FA) composition, there are still conflicting ideas about the best method for overcoming all the difficulties posed by analysis of complex mixtures of FA in meat. This study performed a systematic investigation of factors affecting a one-step method for quantitative analysis of fatty acids in freeze-dried animal tissue. Approximately 280 reactions, selected using factorial design, were performed to investigate the effect of temperature, reaction time, acid concentration, solvent volume, sample weight and sample moisture. The reaction yield for different types of fatty acids, including saturated, unsaturated (cis, trans and conjugated) and long-chain polyunsaturated fatty acids was determined. The optimised condition for one-step transmethylation was attained with four millilitres 5% sulfuric acid in methanol (as acid catalyst), four millilitres toluene (as co-solvent), 300 mg of freeze-dried meat and incubation at 70 °C for 2 h, with interim mixing by inversion at 30, 60 and 90 min for 15 s. The optimised condition was applied to meat samples from different species, covering a broad range of fat content and offers a simplified and reliable method for analysis of fatty acids from meat samples.

Mapping the Spatiotemporal Distribution of Acid and Moisture in Food Structures during Gastric Juice Diffusion Using Hyperspectral Imaging.

This study investigated the feasibility of using hyperspectral imaging (HSI) to characterize the diffusion of acid and water within food structures during gastric digestion. Two different sweet potatoes (steamed and fried) and egg white gel (pH5 and pH9 EWGs) structures were exposed to in vitro gastric digestion before scanning by HSI. Afterward, the moisture or acid present in the digested sample was analyzed for calibration purposes. Calibration models were subsequently built using partial least-squares (PLS). The PLS models indicated that the full-wavelength spectral range (550-1700 nm) had a good ability to predict the spatial distribution of acid (Rcal2 > 0.82) and moisture (Rcal2 > 0.88). The spatiotemporal distributions of moisture and acid were mapped across the digested food, and they were shown to depend on the food composition and structure. The kinetic data revealed that the acid and moisture uptakes are governed by Fickian diffusion or by both diffusion and erosion-controlled mechanisms.

Chemometrics and hyperspectral imaging applied to assessment of chemical, textural and structural characteristics of meat.

Spectroscopy in the visible near-infrared spectral (Vis-NIRS) range combined with imaging techniques (hyperspectral imaging, HSI) allows assessment of chemical composition, texture, and meat structure. The use of HSI in the meat and food industry has observed a significant growth in the last decade, yet its use for assessment of meat it is not optimal yet. The application of HSI for assessment of meat is reviewed with focus on its ability to capture meat unique chemical and structural characteristics. While HSI is widely used for assessment of chemical composition, a limited number of evidences on its ability to handle the effect of different sources of variation on the assessment is found. The use of spatially resolved spectroscopy has been able to detect structural information related to animal background, muscle type, rigor process and ageing. Similarly the use of texture features seem to capture unique characteristics of meat.

Characterization of volatile metabolites associated with confinement odour during the shelf-life of vacuum packed lamb meat under different storage conditions.

Confinement odour was investigated. Volatiles were extracted directly from the pack, using solid phase microextraction and analysed by gas chromatography-mass spectrometry. Sensory evaluation and microbiological analysis of the meat surface were also performed. Commercial samples of vacuum packed lamb legs (n=85), from two meat processing plants, were kept for 7weeks at -1.5°C then at different regimes of temperature (-1.5 to +4°C) until 11, 12 or 13weeks. Persistent odour was observed in 66% of samples, confinement odour in 24% and no odour in 11%. Volatiles associated with confinement odour (3-methyl-butanal, 3-hydroxy-2-butanone and sulphur dioxide) corresponded with end/sub products of glucose fermentation and catabolism of amino acids by bacteria (all bacteria naturally found in meat and do not represent a risk to health). Confinement odour could indicate a stage at which the environment for bacteria growth is becoming favourable for the production of volatiles with strong odours that are noticed by the consumer.

Effect of Dietary Complex Lipids on the Biosynthesis of Piglet Brain Gangliosides.

Gangliosides, found in mammalian milk, are known for their roles in brain development of the newborn. However, the mechanism involved in the impact of dietary gangliosides on brain metabolism is not fully understood. The impact of diets containing complex lipids rich in milk-derived ganglioside GD3 on the biosynthesis of gangliosides (assessed from the incorporation of deuterium) in the frontal lobe of a piglet model is reported. Higher levels of incorporation of deuterium was observed in the GM1 and GD1a containing stearic acid in samples from piglets fed milk containing 18.2 µg/mL of GD3 compared to that in those fed milk containing 25 µg/mL of GD3. This could suggest that the gangliosides from the diet may be used as a precursor for de novo biosynthesis of brain gangliosides or lead to the reduction of de novo biosynthesis of these gangliosides. This effect was more pronounced in the left compared to that in the right brain hemisphere.

Early on-line classification of beef carcasses based on ultimate pH by near infrared spectroscopy.

Prediction of ultimate pH (measured 48 h post mortem; pH(u)) in beef from Visible-near infrared (VIS-NIR) spectra collected 20 to 40 min post mortem was assessed. Spectra were collected from carcasses (cows: n = 86, bulls: n = 170, steers: n = 363, and heifers: n = 38) in a commercial hot boning abattoir under routine conditions. Partial Least Squares (PLS) models showed limited accuracy with RMSE for validation equal to 0.26, 0.20 and 0.36 for the All-animals, Non-bulls and Bulls models, respectively. The pH(u)-PLS-predicted values were used to segregate carcasses as normal (pH(u)<5.8) or high (pH(u) ≥ 5.8) showing better performance, by correctly classifying at least 90% of high pH(u) carcasses. The Non-bulls model was equivalent to the current technology used in the abattoir to classify carcasses based on pH(u). Thus near infrared spectroscopy (NIRS) could be used for on-line classification of beef carcasses based on pH(u).