Exploiting multispectral imaging for non-invasive contamination assessment and mapping of meat samples

Exploring the potential of hyperspectral imaging for microbial assessment of meat: A review

Food safety is always of paramount importance globally due to the devasting social and economic effects of foodborne disease outbreaks. There is a high consumption rate of meat worldwide, making it an essential protein source in the human diet, hence its microbial safety is of great importance. The food industry stakeholders are always in search of methods that ensure safe food whilst maintaining food quality and excellent sensory attributes. Currently, there are several methods used in microbial food analysis, however, these methods are often time-consuming and do not allow real-time analysis. Considering the recent technological breakthroughs in artificial intelligence and machine learning, it raises the question of whether these advancements could be leveraged within the meat industry to improve turnaround time for microbial assessments. Hyperspectral imaging (HSI) is a highly prospective technology worth exploring for microbial analysis. The rapid, non-destructive method has the potential to be integrated into food production systems and allows foodborne pathogen detection in food samples, thus saving time. Although there has been a substantial increase in research on the utilisation of HSI in food applications over the past years, its use in the microbial assessment of meat is not yet optimal. This review aims to provide a basic understanding of the visible-near infrared HSI system, recent applications in the microbial assessment of meat products, challenges, and possible future applications.

Spectroscopic Data for the Rapid Assessment of Microbiological Quality of Chicken Burgers

The rapid assessment of the microbiological quality of highly perishable food commodities is of great importance. Spectroscopic data coupled with machine learning methods have been investigated intensively in recent years, because of their rapid, non-destructive, eco-friendly qualities and their potential to be used on-, in- or at-line. In the present study, the microbiological quality of chicken burgers was evaluated using Fourier transform infrared (FTIR) spectroscopy and multispectral imaging (MSI) in tandem with machine learning algorithms. Six independent batches were purchased from a food industry and stored at 0, 4, and 8 °C. At regular time intervals (specifically every 24 h), duplicate samples were subjected to microbiological analysis, FTIR measurements, and MSI sampling. The samples (n = 274) acquired during the data collection were classified into three microbiological quality groups: “satisfactory”: 4−7 log CFU/g, “acceptable”: 7−8 log CFU/g, and “unacceptable”: >8 logCFU/g. Subsequently, classification models were trained and tested (external validation) with several machine learning approaches, namely partial least squares discriminant analysis (PLSDA), support vector machine (SVM), random forest (RF), logistic regression (LR), and ordinal logistic regression (OLR). Accuracy scores were attained for the external validation, exhibiting FTIR data values in the range of 79.41−89.71%, and, for the MSI data, in the range of 74.63−85.07%. The performance of the models showed merit in terms of the microbiological quality assessment of chicken burgers.

Microbiological Quality Assessment of Chicken Thigh Fillets Using Spectroscopic Sensors and Multivariate Data Analysis

Fourier transform infrared spectroscopy (FT-IR) and multispectral imaging (MSI) were evaluated for the prediction of the microbiological quality of poultry meat via regression and classification models. Chicken thigh fillets (n = 402) were subjected to spoilage experiments at eight isothermal and two dynamic temperature profiles. Samples were analyzed microbiologically (total viable counts (TVCs) and Pseudomonas spp.), while simultaneously MSI and FT-IR spectra were acquired. The organoleptic quality of the samples was also evaluated by a sensory panel, establishing a TVC spoilage threshold at 6.99 log CFU/cm². Partial least squares regression (PLS-R) models were employed in the assessment of TVCs and Pseudomonas spp. counts on chicken’s surface. Furthermore, classification models (linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), support vector machines (SVMs), and quadratic support vector machines (QSVMs)) were developed to discriminate the samples in two quality classes (fresh vs. spoiled). PLS-R models developed on MSI data predicted TVCs and Pseudomonas spp. counts satisfactorily, with root mean squared error (RMSE) values of 0.987 and 1.215 log CFU/cm², respectively. SVM model coupled to MSI data exhibited the highest performance with an overall accuracy of 94.4%, while in the case of FT-IR, improved classification was obtained with the QDA model (overall accuracy 71.4%). These results confirm the efficacy of MSI and FT-IR as rapid methods to assess the quality in poultry products.

Data Science in the Food Industry

Food safety is one of the main challenges of the agri-food industry that is expected to be addressed in the current environment of tremendous technological progress, where consumers' lifestyles and preferences are in a constant state of flux. Food chain transparency and trust are drivers for food integrity control and for improvements in efficiency and economic growth. Similarly, the circular economy has great potential to reduce wastage and improve the efficiency of operations in multi-stakeholder ecosystems. Throughout the food chain cycle, all food commodities are exposed to multiple hazards, resulting in a high likelihood of contamination. Such biological or chemical hazards may be naturally present at any stage of food production, whether accidentally introduced or fraudulently imposed, risking consumers' health and their faith in the food industry. Nowadays, a massive amount of data is generated, not only from the next generation of food safety monitoring systems and along the entire food chain (primary production included) but also from the Internet of things, media, and other devices. These data should be used for the benefit of society, and the scientific field of data science should be a vital player in helping to make this possible.

Rapid Microbial Quality Assessment of Chicken Liver Inoculated or Not With Salmonella Using FTIR Spectroscopy and Machine Learning

Chicken liver is a highly perishable meat product with a relatively short shelf-life and that can get easily contaminated with pathogenic microorganisms. This study was conducted to evaluate the behavior of spoilage microbiota and of inoculated Salmonella enterica on chicken liver. The feasibility of Fourier-transform infrared spectroscopy (FTIR) to assess chicken liver microbiological quality through the development of a machine learning workflow was also explored. Chicken liver samples [non-inoculated and inoculated with a four-strain cocktail of ca. 10³ colony-forming units (CFU)/g Salmonella] were stored aerobically under isothermal (0, 4, and 8°C) and dynamic temperature conditions. The samples were subjected to microbiological analysis with concomitant FTIR measurements. The developed FTIR spectral analysis workflow for the quantitative estimation of the different spoilage microbial groups consisted of robust data normalization, feature selection based on extra-trees algorithm and support vector machine (SVM) regression analysis. The performance of the developed models was evaluated in terms of the root mean square error (RMSE), the square of the correlation coefficient (R²), and the bias (B_f) and accuracy (A_f) factors. Spoilage was mainly driven by Pseudomonas spp., followed closely by Brochothrix thermosphacta, while lactic acid bacteria (LAB), Enterobacteriaceae, and yeast/molds remained at lower levels. Salmonella managed to survive at 0°C and dynamic conditions and increased by ca. 1.4 and 1.9 log CFU/g at 4 and 8°C, respectively, at the end of storage. The proposed models exhibited A_f and B_f between observed and predicted counts within the range of 1.071 to 1.145 and 0.995 to 1.029, respectively, while the R² and RMSE values ranged from 0.708 to 0.828 and 0.664 to 0.949 log CFU/g, respectively, depending on the microorganism and chicken liver samples. Overall, the results highlighted the ability of Salmonella not only to survive but also to grow at refrigeration temperatures and demonstrated the significant potential of FTIR technology in tandem with the proposed spectral analysis workflow for the estimation of total viable count, Pseudomonas spp., B. thermosphacta, LAB, Enterobacteriaceae, and Salmonella on chicken liver.

Spoilage of Chilled Fresh Meat Products during Storage: A Quantitative Analysis of Literature Data

A literature search was performed on spoilage of fresh meat products by combining keyword query, text mining and expert elicitation. From the 258 collected studies, a quantitative analysis was first performed to identify the methods which are the most used to evaluate spoilage beside the preservation strategies suggested. In a second step focusing on a subset of 24 publications providing quantitative data on spoilage occurrence time, associations between spoilage occurrence time of meat products and specific spoilage indicators were investigated. The analysis especially focused on factors well represented in the 24 publications, i.e., gas packaging (O2 and CO2) and storage temperature. Relationships between spoilage occurrence and several microbiological indicators were also sought. The results point out possible advantages of removing dioxygen in packaging to delay spoilage occurrence, whereas, in the presence of dioxygen, the carbon dioxide proportion in the gas mixtures was shown to influence spoilage occurrence. The collected data clearly reveal a potentially protective role of lactic acid bacteria. Besides, while a spoilage role could be attributed to Pseudomonas spp., the growth of mesophilic aerobic microbes, Brochothrix spp. and Enterobacteriaceae seemed independent of spoilage occurrence time.

A machine learning workflow for raw food spectroscopic classification in a future industry

Over the years, technology has changed the way we produce and have access to our food through the development of applications, robotics, data analysis, and processing techniques. The implementation of these approaches by the food industry ensure quality and affordability, reducing at the same time the costs of keeping the food fresh and increase productivity. A system, as the one presented herein, for raw food categorization is needed in future food industries to automate food classification according to type, the process of algorithm approaches that will be applied to every different food origin and also for serving disabled people. The purpose of this work was to develop a machine learning workflow based on supervised PLS regression and SVM classification, towards automated raw food categorization from FTIR. The system exhibited high efficiency in multi-class classification of 7 different types of raw food. The selected food samples, were diverse in terms of storage conditions (temperature, storage time and packaging), while the variability within each food was also taken into account by several different batches; leading in a classifier able to embed this variation towards increased robustness and efficiency, ready for real life applications targeting to the digital transformation of the food industry.

Implementation of Multispectral Imaging (MSI) for Microbiological Quality Assessment of Poultry Products

The aim of this study was to investigate on an industrial scale the potential of multispectral imaging (MSI) in the assessment of the quality of different poultry products. Therefore, samples of chicken breast fillets, thigh fillets, marinated souvlaki and burger were subjected to MSI analysis during production together with microbiological analysis for the enumeration of Total Viable Counts (TVC) and Pseudomonas spp. Partial Least Squares Regression (PLS-R) models were developed based on the spectral data acquired to predict the "time from slaughter" parameter for each product type. Results showed that PLS-R models could predict effectively the time from slaughter in all products, while the food matrix and variations within and between batches were identified as significant factors affecting the performance of the models. The chicken thigh model showed the lowest RMSE value (0.160) and an acceptable correlation coefficient (r = 0.859), followed by the chicken burger model where RMSE and r values were 0.285 and 0.778, respectively. Additionally, for the chicken breast fillet model the calculated r and RMSE values were 0.886 and 0.383 respectively, whereas for chicken marinated souvlaki, the respective values were 0.934 and 0.348. Further improvement of the provided models is recommended in order to develop efficient models estimating time from slaughter.

Preliminary investigation of the use of Raman spectroscopy to predict beef spoilage in different types of packaging

In this study, pH, meat color analysis, microbial counts and Raman spectroscopic data were obtained from beef steaks stored at 4 °C for up to 21 days using two different packaging methods: vacuum (VP) and modified atmosphere packaging (MAP). Models using partial least square regression (PLSR), indicated that Raman spectroscopy was able to predict total viable counts (TVC) and lactic acid bacteria (LAB) measured at 21d post mortem (TVC in VP: R²_cv = 0.99, RMSEP = 0.61; TVC in MAP: R²_cv = 0.90, RMSEP = 0.38; LAB in VP: R²_cv = 0.99, RMSEP = 0.54; LAB in MAP: R²_cv = 0.75, RMSEP = 0.60). The results of this study demonstrate that Raman spectroscopy may have potential for the rapid determination of meat spoilage.

Online Feature Selection for Robust Classification of the Microbiological Quality of Traditional Vanilla Cream by Means of Multispectral Imaging

The performance of an Unsupervised Online feature Selection (UOS) algorithm was investigated for the selection of training features of multispectral images acquired from a dairy product (vanilla cream) stored under isothermal conditions. The selected features were further used as input in a support vector machine (SVM) model with linear kernel for the determination of the microbiological quality of vanilla cream. Model training (n = 65) was based on two batches of cream samples provided directly by the manufacturer and stored at different isothermal conditions (4, 8, 12, and 15 °C), whereas model testing (n = 132) and validation (n = 48) were based on real life conditions by analyzing samples from different retail outlets as well as expired samples from the market. Qualitative analysis was performed for the discrimination of cream samples in two microbiological quality classes based on the values of total viable counts [TVC ≤ 2.0 log CFU/g (fresh samples) and TVC ≥ 6.0 log CFU/g (spoiled samples)]. Results exhibited good performance with an overall accuracy of classification for the two classes of 91.7% for model validation. Further on, the model was extended to include the samples in the TVC range 2-6 log CFU/g, using 1 log step to define the microbiological quality of classes in order to assess the potential of the model to estimate increasing microbial populations. Results demonstrated that high rates of correct classification could be obtained in the range of 2-5 log CFU/g, whereas the percentage of erroneous classification increased in the TVC class (5,6) that was close to the spoilage level of the product. Overall, the results of this study demonstrated that the UOS algorithm in tandem with spectral data acquired from multispectral imaging could be a promising method for real-time assessment of the microbiological quality of vanilla cream samples.

Lighting Deviation Correction for Integrating-Sphere Multispectral Imaging Systems

In an integrating sphere multispectral imaging system, measurement inconsistency can arise when acquiring the spectral reflectances of samples. This is because the lighting condition can be changed by the measured samples, due to the multiple light reflections inside the integrating sphere. Besides, owing to non-uniform light transmission of the lens and narrow-band filters, the measured reflectance is spatially dependent. To deal with these problems, we propose a correction method that consists of two stages. The first stage employs a white board to correct non-uniformity and a small white patch to correct lighting deviation, both under the assumption of ideal Lambertian reflection. The second stage uses a polynomial regression model to further remove the lighting inconsistency when measuring non-Lambertian samples. The method is evaluated on image data acquired in a real multispectral imaging system. Experimental results illustrate that our method eliminates the measurement inconsistency considerably. This consequently improves the spectral and colorimetric accuracy in color measurement, which is crucial to practical applications.

Estimation of Minced Pork Microbiological Spoilage through Fourier Transform Infrared and Visible Spectroscopy and Multispectral Vision Technology

Spectroscopic and imaging methods coupled with multivariate data analysis have been increasingly studied for the assessment of food quality. The objective of this work was the estimation of microbiological quality of minced pork using non-invasive spectroscopy-based sensors. For this purpose, minced pork patties were stored aerobically at different isothermal (4, 8, and 12 °C) and dynamic temperature conditions, and at regular time intervals duplicate samples were subjected to (i) microbiological analyses, (ii) Fourier transform infrared (FTIR) and visible (VIS) spectroscopy measurements, and (iii) multispectral image (MSI) acquisition. Partial-least squares regression models were trained and externally validated using the microbiological/spectral data collected at the isothermal and dynamic temperature storage conditions, respectively. The root mean squared error (RMSE, log CFU/g) for the prediction of the test (external validation) dataset for the FTIR, MSI, and VIS models was 0.915, 1.173, and 1.034, respectively, while the corresponding values of the coefficient of determination (R²) were 0.834, 0.727, and 0.788. Overall, all three tested sensors exhibited a considerable potential for the prediction of the microbiological quality of minced pork.

Evaluation of Fourier transform infrared spectroscopy and multispectral imaging as means of estimating the microbiological spoilage of farmed sea bream

The objective of the present study was the evaluation of Fourier transform infrared (FTIR) spectroscopy and multispectral imaging (MSI), in tandem with multivariate data analysis, as means of estimating the microbiological quality of sea bream. Farmed whole ungutted fish were stored aerobically at 0, 4 and 8 °C. At regular time intervals, fish samples (i.e. cut portions) were analysed microbiologically, while FTIR and MSI measurements also were acquired at both the skin and flesh sides of the samples. Partial least squares regression (PLSR) models were calibrated to provide quantitative estimations of the microbiological status of fish based on spectral data, in a temperature-independent manner. The PLSR model based on the FTIR data of fish skin exhibited good performance when externally validated, with the coefficient of determination (R²) and the root mean square error (RMSE) being 0.727 and 0.717, respectively. Hence, FTIR spectroscopy appears to be promising for the rapid and non-invasive monitoring of the microbiological spoilage of whole sea bream. Contrarily, the MSI models' performance was unsatisfactory, delimitating their potential exploitation in whole fish quality assessment. Model optimization results concerning fish flesh indicated that MSI may be propitious in skinned fish products, with its definite competence warranting further investigation.

Investigation of pausing fermentation of salamis with multispectral imaging for optimal sensory evaluations

The fermentation process of salamis involves several parameters influencing taste, texture, and color of the salami. One significant parameter is the fermentation time. It is difficult to conduct sensory evaluations to assess the effect of time without introducing variation between observation days. It may be possible to overcome this by stalling or pausing the fermentation by deep-chilling the salamis. This study investigates the difference of non- and deep-chilled salamis with the use of a multispectral imaging system. The statistical investigation, based on image features relating to size, visual texture, and color of the sausages over time, showed that it may be possible to stall the fermentation process. It was shown that a statistical difference in the two kinds of samples is present. For the size feature the difference could be quantified into a number of days. However, for the important color feature only a statistical difference was observed, whereas the visual difference expressed in terms of ΔE_ab^⁎ was barely present.

Optimized lighting method of applying shaped-function signal for increasing the dynamic range of LED-multispectral imaging system

This paper proposes an optimized lighting method of applying a shaped-function signal for increasing the dynamic range of light emitting diode (LED)-multispectral imaging system. The optimized lighting method is based on the linear response zone of the analog-to-digital conversion (ADC) and the spectral response of the camera. The auxiliary light at a higher sensitivity-camera area is introduced to increase the A/D quantization levels that are within the linear response zone of ADC and improve the signal-to-noise ratio. The active light is modulated by the shaped-function signal to improve the gray-scale resolution of the image. And the auxiliary light is modulated by the constant intensity signal, which is easy to acquire the images under the active light irradiation. The least square method is employed to precisely extract the desired images. One wavelength in multispectral imaging based on LED illumination was taken as an example. It has been proven by experiments that the gray-scale resolution and the accuracy of information of the images acquired by the proposed method were both significantly improved. The optimum method opens up avenues for the hyperspectral imaging of biological tissue.

Rapid detection of frozen-then-thawed minced beef using multispectral imaging and Fourier transform infrared spectroscopy

In recent years, fraud detection has become a major priority for food authorities, as fraudulent practices can have various economic and safety consequences. This work explores ways of identifying frozen-then-thawed minced beef labeled as fresh in a rapid, large-scale and cost-effective way. For this reason, freshly-ground beef was purchased from seven separate shops at different times, divided in fifteen portions and placed in Petri dishes. Multi-spectral images and FTIR spectra of the first five were immediately acquired while the remaining were frozen (-20°C) and stored for 7 and 32days (5 samples for each time interval). Samples were thawed and subsequently subjected to similar data acquisition. In total, 105 multispectral images and FTIR spectra were collected which were further analyzed using partial least-squares discriminant analysis and support vector machines. Two meat batches (30 samples) were reserved for independent validation and the remaining five batches were divided in training and test set (75 samples). Results showed 100% overall correct classification for test and external validation MSI data, while FTIR data yielded 93.3 and 96.7% overall correct classification for FTIR test set and external validation set respectively.