This research underscores the potential of advanced analytical techniques to improve the precision of sensory evaluations in food quality assessment.
Several scientific research works aim to provide outcomes related to quality control (QC), technological processing, traceability, and authenticity in food products. In the case of meat products and their derivatives, exception is not applied. In Portugal, there exist three native pig breeds: the Bísaro, the Malhado de Alcobaça, and the Alentejano[1]. Bísaro represents a breed of autochthonous Portuguese pigs with Celtic origins and a part of Portugal's biological, economic, and cultural heritage[2]. It is typically produced in a semi-extensive system, with its dietary management relying on locally available agricultural resources[3]. In addition, the Bísaro breed is known for the quality of the meat and fat from these animals, used for the manufacture of various products of excellence and specific qualities that hold designations such as: Protected Geographical Indication (PGI) and Protected Designation of Origin (PDO). Despite the fact that several researches have been carried out into the reformulation of meat products, the traditional processes used to manufacture these meat products are still of the utmost importance. Most of them undergo traditional meat preservation methods, such as salting, drying, or smoking. Some examples are typical Mediterranean dry-cured meat processing characterized by dry salting, no smoking, and long drying process, while brine salting and smoking are used in continental parts of Europe[4]. The specific regional conditions for the application of these methods together with the typicality of raw material (genetic type, feed, rearing system, etc) makes it possible to obtain particularly diverse dry-cured products. Dry-cured products for overall acceptance depended to a large reaction that result from traditional production methods, which is mainly determined by flavor, taste and odor compounds[5]. In dry-cured products, the key attributes are markedly affected by the ripening process, complex chemical and biochemical changes in the main components of raw meat which contribute to their characteristic aroma and flavor[6]. On the other hand, the meat products resulting from pork also depends on factors intrinsic and extrinsic to the animal. However, quality of meat products from Iberian pigs strongly depends upon breed and rearing system, which includes a number of different lines, causing a great heterogeneity within the same breed. Indeed, the information provided to consumers on the impact on health or any other quality aspect can significantly influence their acceptance of meat products[7][8]. Moreover, consumer expectations regarding with the perception of safety associated with processed meat, animal welfare, processing techniques, and the preservation of traditional production methods[9], denote the importance of utilizing an autochthonous breed, such as the Bísaro, in order to add value to the product. These characteristics, in the classic way, are usually determined by mechanical, physico-chemical measurements and sensory analysis most of them are invasive, expensive, and time-consuming[10]. Consequently, the sensory evaluation (SE) of the meat and meat products is a necessary tool to check the acceptability by the consumers as well as characterize each product. However, there are inherent subjective issues (including assessor selection, attribute generation, and context effects, among others)[11] that generate greater variability associated with sensory evaluations due to their reliance on humans' resources. Although this SE is carried out by a panel of trained tasters, it is crucial to carefully select and train sensory marks, use analytic tests (discrimination and descriptive), establish clear evaluation protocols, and awareness of potential sources of variability are necessary to improve the reliability of sensory evaluations[12]. Descriptive analysis (DA) has been used to characterize and discriminate meat and meat products and is the most reliable sensory method for sensory characterization of these meat products. Although DA provides accurate and reliable results, the economic efforts and time needed to train evaluators is a major disadvantage[11]. With all these in mind, the use of alternative techniques, such as computed tomography (CT)[13], magnetic resonance imaging (MRI)[10], hyperspectral imaging (HIS) or near infrared (NIR) spectroscopy[14][15] must be considered in order to support and complement conventional sensory techniques more quickly. The alternative techniques, are increasingly gaining priority due to prompt, easy to use and minimal pre-processing requirements, making them suitable for rapid implementation in industrial meat applications[10]. In this sense, the results obtained are valuable for the companies and researchers in the sense that they are more objective, giving an idea of the quality of the product throughout the real-time analyses. This allows for improvements to be made in order to achieve the desired level of QC, without forget the organoleptic properties such as odor, flavor, and texture must be considered.
For decades, DA has been used for the characterization and discrimination of meat and meat products[16]. For that, the descriptor chosen are vital in these analyses because all specific attributes allow a complete characterization of the dry-cured product in study and describe it in more detail. Also, it was applied with a 9-point hedonic scale (used for the assessment of the linking of the product) because this hedonic test could provide information about different products, textures, composition, etc., which would help to better understand the taster’s answer[11]. In these sense, the descriptive values of dry-cured Bísaro loins determined are summarized in Table 1. Mean values together with the minimum, maximum and standard deviation for the 40 samples analyzed shown that the range of variation was wide enough to guarantee an adequate margin for calibration purposes.
Table 1.
Sensory attributes of the forty dry-cured Bísaro loins with min, max and mean represented (n=40).
Attributes |
Defenition |
Min |
Max |
Mean (±sd) |
Odor |
The presence of a typical odor of a dry-cured product [16] |
5.13 |
6.78 |
5.92(±0.38) |
Andros |
The presence of a metabolites of testosterone [45] |
1.11 |
2.38 |
1.50(±0.28) |
Skatole |
The presence of organic compounds who contributor to fecal odor [45] |
1.00 |
1.78 |
1.24(±0.19) |
Lean color |
The color of the part of muscle sample [45] |
2.88 |
6.11 |
4.05(±0.73) |
Fat color |
The color intensity and brightness of the fat [45] |
1.56 |
4.89 |
3.11(±0.85) |
Hardness |
The force necessary to penetrate the meat with the incisors [16] |
2.44 |
6.56 |
3.90(±1.23) |
Juiciness |
The amount of juice given off by the sample when chewed [16] |
3.44 |
6.11 |
5.10(±0.64) |
Chewiness |
The number of times the sample must be chewed before it can be swallowed [16] |
2.13 |
5.44 |
3.62(±0.85) |
Flavor intensity |
The intensity of overall flavor on the samples [16] |
5.11 |
6.44 |
5.89(±0.33) |
Flavor persistence |
The persistence of overral flavor on the mouthfeel [16] |
4.56 |
6.33 |
5.67(±0.42) |
sd– standard deviation; Andros- odor androsterone; Min- minimum; Max- maximum; 9- point scale with the extremes representing either the minimum (low intense sensation) or the maximum (high intense sensation)
.In a general way, it can be concluded that the data obtained from the evaluation of trained tasters give quantitative and objective information despite the wide range. In fact, with the AD test (hedonic evaluation) used, it was possible to obtain information on the magnitude of the liking or disliking of these sensory attributes studied. It was possible to verify that the intramuscular composition and fat deposition existing in the dry-cured Bísaro loin lines[17][18] resulted in the rapid release of fluid contained in them and justify the variation found in the texture parameters of our work. All of them showed relatively large range (6.56-2.44; 5.44-2.13; 6.11-3.44 for hardness, chewiness and juiciness, respectively) with an average value of 3.90, 3.62, and 5.10, respectively, a standard deviation of 1.23, 0.85 and 0.64, respectively. Therefore, the development of acceptable texture characteristics in this meat products is very important for the successful marketing of high-quality products, as this study aims to achieve. The texture properties include a complex process in which both ingredients and traditional processing steps have a vital role.The odor is also important attribute inter with flavor, because the ripening degree (proteolysis and lipolysis changes that occur in drying process) reflects the ripening odor compounds that become a significant part of meat flavor[19][20][21]. Therefore, the range values of odor could be partially due to the biochemical changes during the drying period, very high compared with Seong et al.[22] work. Regarding appearance, mean values for fat color and color were lower (3.11 and 4.05, respectively) to those described by Revilla et al.[23] and within those obtained by Seong et al.[22], with 4.75 value given by the tasters for the loins ripened for 60 days in their study. Other attributes evaluated like androsterone and skatole were included in flavor parameters in this work. Studies demonstrate that, due to the high levels of intrinsic factors such as age, sex and weight at slaughter, the meat from entire males presents undesirable odors, even when clarity a limit of 150 days of age and 100 kg of slaughter weight[24]. Our samples complied the weights close to 100 kg and used castrated animals so it was expected that these hormones would not have a high expression in our analysis. Autochthonous breeds such as Bísaro are not suited for high yields of lean meat production, but for processed products, as they have highly marbled meat that confers excellent flavor due to the variety of feed they typically consume[3]. Regarding the presence of fat, the feeding process can influence the duration of the drying process, specifically affecting the intensity and persistence of flavor[25]. Longer periods for drying positive affects the biochemical and microbiological development of flavor. In this way, it was expected high range values on flavor intensity and flavor persistence (6.44-5.11; 6.33-4.56, respectively), in accordance with the findings of (4.60 flavor intensity) on study of Seong et al. [53] for 60 days of ripening time on sensory characteristics of dry-cured loins. So, it is crucial to define the distinctive attributes that characterize these products. Sensory evaluation plays a significant role in this regard, as attributes such as odor, taste, color, texture and even the presence of visual fat are representative of the product. Thus, NIR is a key tool that must be applied to streamline and substantiate the entire process[9].
The use of Support Vector machine Regression (SVmR) has been shown to be the most suitable for data modelling, after a comparison was made between PLS and SVmR models to evaluate their performance in modelling meat characterization data[17]. In addition, when pre-treatments are applied to reduce and correct possible interferences the results are improved[9][17][26][27]. In this context, the regression models to predict the sensory attributes of the dry-cured loins were obtained using different kinds of SVR tested. Through a hybrid algorithm based on particle swarm optimization (PSO)[28] was possible to minimize the root mean square error (RMSECV) (used as the predictive evaluation criteria) obtained by 5-fold cross-validation for the parameters. The best results were achieved by applying MSC pre-treatment, MinMax normalization, ε-SVR, and the radial base kernel. Table 2 presents the best optimized ε-SVR for each sensory attribute.
Table 2
. Parameters and figures of merit of the best ε-SVR models obtained for each sensory attribute.
|
Calibration |
Prediction |
||||||
Attribute |
C |
ε |
γ |
RMSE |
R2 |
RMSE |
R2 |
RSD(%) |
Odor |
23.43 |
0.0161 |
0.0227 |
0.0155 |
0.9995 |
0.0549 |
0.9888 |
0.98 |
Andros |
18.11 |
0.0010 |
0.0258 |
0.0011 |
1.0000 |
0.0400 |
0.9892 |
2.87 |
Scatol |
87.79 |
0.0051 |
0.0221 |
0.0051 |
0.9998 |
0.0548 |
0.9616 |
4.47 |
Lean color |
39.03 |
0.0074 |
0.0151 |
0.0072 |
0.9998 |
0.0507 |
0.9853 |
1.85 |
Fat color |
100.0 |
0.0010 |
0.0446 |
0.0010 |
1.0000 |
0.0685 |
0.9878 |
2.26 |
Hardness |
100.0 |
0.0010 |
0.0257 |
0.0011 |
1.0000 |
0.0403 |
0.9955 |
1.03 |
Juiciness |
34.28 |
0.0522 |
0.0155 |
0.0499 |
0.9966 |
0.1031 |
0.9705 |
2.65 |
Chewiness |
40.46 |
0.0395 |
0.0090 |
0.0374 |
0.9966 |
0.0674 |
0.9800 |
1.81 |
Flavor intensity |
53.00 |
0.0252 |
0.0135 |
0.0240 |
0.9974 |
0.0554 |
0.9876 |
0.98 |
Flavor persistence |
51.00 |
0.0010 |
0.0124 |
0.0011 |
1.0000 |
0.0417 |
0.9907 |
0.80 |
* All models used MSC pre-treatment, MinMax normalization, and radial base kernel; MSC- multiplicative scatter correction; Andros- odor androsterone; C/ γ /ε PSO- particle swarm optimization parameters; RSD- percentual relative standard deviation; RMSE- root mean square error; R2- coefficient of determination.
For the ε-SVR with the radial base kernel, the parameters C, ε, and γ were optimized (Table 2) using the PSO algorithm to minimize the RMSECV for 5-fold cross-validation. The values of these parameters control the complexity of the regression model and, consequently, the prediction capabilities for new data sets. Low values of ε bring the models closer to the calibration data. However, this excessive adjustment may result in a loss of generalization to predict new data. The parameter γ is related to smoothness and C to the complexity of the regression model. The regression model is spikier and more complex for high values of γ and C[29][30]. Good SVR models were obtained for all sensory attributes with R2 close to the value 1 (0.9616 - 0.9955) and low values of RMSE (0.0400 – 0.1031) for the prediction set. Furthermore, the relative standard deviation (RSD) for the prediction set was less than 5 % for all sensory attributes, and the confidence interval (95 %) contains the ideal point (unit slope and zero intercept) [64] for all the models in Table 3. All PLS models had poor prediction capabilities, and the Durbin-Watson statistical test (DW) had non-significant probabilities (pDW > 0.05), indicating a lack of correlation between PLS residuals and lack of non-linearities in the multivariate signal[31]. However, applying a non-linear SVR model greatly improved the prediction capability of the meat sensory attributes using the NIR spectra because of several key advantages: its ability to capture intricate data patterns that PLS, due to its linear nature, might overlook; its diminished sensitivity to multicollinearity when compared to PLS; and its more effective handling of outliers through a margin-based approach.
This work generated acceptable predictive models of sensory parameters using advanced chemometric techniques. The NIR spectra exhibited characteristic peaks of physico-chemical characteristics related to major components such as proteins, lipids and water. Overall, the present study shows that NIR has potential as an analytical tool for real-time meat quality control. The study demonstrated that non-linear SVR models, particularly when applied to NIR spectra, significantly improved the prediction of sensory attributes in dry-cured Bísaro loins by offering a promising method for classifying individual animals in breeding programs (extensive production system) and applying this technique in situ at an industrial level to obtain product recognition characteristics of the Bísaro breed. This research's potential applications include refining quality assessment methodologies, guiding product development strategies, and fostering innovation in meat processing technologies. Implementing precise sensory attribute prediction can elevate product standards, meet consumer preferences, and drive advancements in the dry-cured loins industry.