Currently, sensory and microbiological analyses are often used to evaluate the freshness, spoilage or safety of meat and meat products. However, sensory analysis is costly and relies heavily on highly trained panelists. Furthermore, results of microbiological analyses are time consuming (conventional microbiology), costly, may require high-tech molecular tools, and are usually destructive to test products. For this reason, rapid analytical methods or tools are necessary to ensure meat quality. The iMeatSense project proposes a multidisciplinary and interdisciplinary scientific approach integrating food microbiology, analytical chemistry, existing rapid sensor technologies and data analysis based on chemometrics, machine learning and computational intelligence techniques that will ensure the quality of meat products.
Materials & methods
For the purposes of the project, widely used instrumental techniques were explored: (i) multispectral imaging (MSI), (ii) Fourier Transform InfraRed (FTIR) Spectrometry, (iii) electronic nose (e-nose), (iv) Gas Chromatography - Mass Spectrometry (GC-MS) and (v) High-Performance Liquid Chromatography (HPLC). Instrument and methodology specifications were described in Deliverable (D) 1.1. All experimental procedures were performed in the context of Work Package (WP) 2, as well as WP 6 (for model validation purposes) and included one or multiple of the abovementioned sensors. They can be classified in three major categories: The first category was a survey of a large number of minced beef, pork and mix (70-30% w/w beef and pork) samples, while the second included a series of experiments on spoilage where samples were stored in various constant or dynamic temperature conditions. Aerobic storage and Modified Air Packaging (MAP) conditions were also explored. Whereas the former experiments refer to meat quality in terms of bacteria associated to meat spoilage, the last category refers to cases of fraud detection, i.e. adulteration of minced beef with pork or horsemeat. Lastly, the experimental procedures performed included fillets and/or minced beef, pork and horsemeat.
Data mining/analysis and results
The resulting data were included in D2.1 and D2.2 where “microbial fingerprints” and a list of chemical indices were provided. In D2.3 initial analysis exploring the correlation between meat quality/ spoilage/ shelf life and the acquired data was presented.
In parallel, intelligent pre-processing tools for feature extraction and dimensionality reduction were developed in MATLAB software. These tools included an automated pre-processing scheme for multispectral images combining image segmentation, wavelet transform and Gray level co-occurrence matrix feature extraction (D3.1), a Principal Component Analysis variant (D3.2), fuzzy curves and surfaces for variable selection (D3.3), a fuzzy integral methodology for data fusion (D3.4) and a fuzzy logic approach to the problem of small number of samples (D3.5).
WP 4 was dedicated in the development of “Diagnostic” systems. Specifically, the commonly used multi-layer perceptron (MLP) and the radial basis function (RBF) neural networks (D4.1), hybrid-learning (combination of fuzzy and neural) techniques (D4.2) and statistical methods (D4.3) were applied. Lastly, in D4.4 an incremental hybrid learning scheme was proposed.
In the following WP, D5.1 was dedicated to kinetic modelling and prediction models based on metabolite data in meat, whereas in D5.2 a hybrid recurrent system was developed in MATLAB software. Lastly, a prototype web GUI for on-line intelligent decision utilizing multispectral images entitled “iQMS” (intelligent Quality Management System) was developed in D5.3 and is now available online.
Additionally, in WP 6 validation was performed studying the applicability of these methods in real life conditions (D6.1), as well as validating mathematical models (D6.2). For the latter case, two applications were presented: one for image segmentation and one for modelling using various popular data analysis methods.
Conclusions and project dissemination
The aforementioned WPs provided the project’s team members with various and diverse results. While the resulting data varied depending on data pre-processing, modelling method and sensor(s) used, it was evident that the proposed disciplinary approach based on sensor technology is capable of providing an alternative method for meat quality assessment. This is further validated by the successful dissemination of the project’s outcomes (WP 7), which included three publications in international scientific journals, 14 oral and 13 poster presentations in national, European and international conferences which can be found in the project’s official website “http://www.imeatsense.gr/” (D7.1). Lastly, results were transferred to university students, meat industry professionals and/or consumers through training seminars, press and specialized magazines (D7.2).
Back to Top