Previous Article in Journal
Fabricating Zein-OSA Starch Complexes as Multifunctional Carriers for Carrot Oil
Previous Article in Special Issue
Unraveling the Relationship Between Lipids and Volatile Organic Compounds in Longissimus Dorsi of Chuanbai Rex and New Zealand White Rabbits
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Foods
Volume 15
Issue 3
10.3390/foods15030436
foods-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

The Influence of the Site of Recording and Benchtop and Portable NIRS Equipment on Predicting the Sensory Properties of Iberian Ham

by
Isabel Revilla
1,*,
Ana María Vivar-Quintana
1,
Iván Martínez-Martín
1,
Pedro Hernández-Ramos
2,
Miriam Hernández-Jiménez
3,
Justyna Grabska
4,
Krzysztof B. Beć
4 and
Christian W. Huck
4
1
Area of Food Technology, Polytechnical Superior School of Zamora, Universidad de Salamanca, Avenida Requejo 33, 49022 Zamora, Spain
2
Area of Graphic Expression in Engineering, Polytechnical Superior School of Zamora, Universidad de Salamanca, Avenida Requejo 33, 49022 Zamora, Spain
3
Science and Arts Faculty, Universidad Católica de Avila, Calle Canteros s/n, 05005 Avila, Spain
4
Institute of Analytical Chemistry and Radiochemistry, Leopold-Franzens University of Innsbruck, 6020 Innsbruck, Austria
*
Author to whom correspondence should be addressed.
Foods 2026, 15(3), 436; https://doi.org/10.3390/foods15030436 (registering DOI)
Submission received: 5 December 2025 / Revised: 16 January 2026 / Accepted: 22 January 2026 / Published: 24 January 2026
(This article belongs to the Special Issue Nutritional Manipulation, Composition Analysis, and Sensory Evaluation of Meat)
Browse Figures
Versions Notes

Abstract

Iberian ham is a Spanish product that is highly prized for its unique sensory characteristics. The possibility of predicting these characteristics using non-invasive methods is therefore of great interest. In this context, this study compares the performance of two benchtop and four portable devices in addition to the influence of recording factors (lean meat, fat, and whole slices) for the prediction of sensory parameters. A total of 28 descriptors were determined by a trained panel for 60 samples (100% Iberian and Iberian), and the sensory profile was predicted by using a feedforward multilayer perceptron artificial neural network. A significant effect of the breed on the sensory profile and the spectral characteristics was observed, especially in the case of fat. On the other hand, the NIRFlex N-500 (desktop) and MicroPhazir (portable) devices gave the best performances, with 27 parameters predicted for whole slices and fat, respectively (R2 > 0.5). In addition, it was possible to predict 25 parameters by using the MicroNIR portable device at all recording sites, which demonstrated that portable devices are suitable for this analysis. The results indicate that the number and type of sensory parameters predicted depend largely on the recording site and that measurements taken in different areas provide complementary information.

Graphical Abstract

1. Introduction

Iberian ham is a pork product that is not only highly appreciated by Spanish consumers but is also one of the most recognized Spanish products in the world; it is marketed in Europe, Japan, and the United States [1]. One of the reasons why it is in such high demand is the sensory quality of this product, which depends on several factors, such as genetics, breeding systems, and especially the feed provided [2], with 100% acorn-fed Iberian ham being the most prized for its aroma, texture, and flavor [3]. Iberian ham is characterized by the intense red color of its lean meat with a large amount of white intramuscular fat (marbling). In addition, its low fiber content, high juiciness, and soft, fluid fat as texture attributes, together with a high intensity of flavor owing to its cured taste with hints of sweetness, are the main characteristics that influence the acceptability of cured ham [4]. There is, therefore, a complete set of sensory attributes that sensory panels analyze when performing quantitative descriptive analysis (QDA), the most widely used descriptive method, to evaluate quality and to differentiate cured Iberian hams [5,6,7,8,9]. However, the implementation of QDA involves the selection, training, and qualification of evaluators in accordance with recognized standards [10,11,12], and also the validation of the panel to ensure its correct functioning over time [6,13]. Consequently, this method is tedious and time-consuming, in addition to requiring the destruction of the sample and a panel of assessors undergoing continuous training. However, the data obtained are not only reliable but also provide a significant amount of complementary information, which is easy to interpret [14].
As a result of all this, it is of great interest to replace the sensory analysis by non-destructive techniques such as near-infrared spectroscopy, which has advantages such as being quick and non-contaminant, and also offers the possibility of implementation on-line/in-line or at-line [15]. NIR spectroscopy has been used to predict sensory properties in different matrices, as reviewed by Chapman et al. [16]. With regard to meat and meat products, much more attention has been paid to fresh meat than to meat products [17], although some recent studies point to the feasibility of predicting the sensory properties of dry-cured products [18,19,20], including the texture and color or defects of cured ham [21,22] or even the complete QDA profile [8]. Portable devices are available that differ in technical solutions regarding the light source, the detectors, and the wavelength selectors. The light source of handheld NIR spectrometers is therefore based on tungsten-halogen lamps in analogy to benchtop instruments, but light-emitting diodes (LEDs) are also used (SCiO, Consumer Physics, Tel Aviv, Israel) due to their robustness, low price, and low power consumption. As far as detectors are concerned, array detectors (MicroNIR1700, VIAVI Solutions Inc., Milpitas, CA, USA) and single-detector devices (Enterprise Sensor NIR, TellSpec, Toronto, ON, Canada) can be found in portable NIR equipment. Indium gallium arsenide (InGaAs) is the preferred material for manufacturing detectors and has replaced lead salt detectors PbSe and PbS (Foss NIRSystem 5000, Foss NIRSystems, Silver Spring, MD, USA), although complementary metal–oxide–semiconductor (CMOS) technology has been gaining ground steadily, as this requires less power consumption (SCiO Consumer Physics, Tel Aviv, Israel). As for wavelength selectors, the equipment can currently be divided into the following: Linear-variable filter instruments with array detectors (trinamiX GmbH, Ludwigshafen, Germany; VIAVI Solutions Inc., CA, USA), MEMS-based FT-NIR instruments (Si-Ware Systems, Cairo, Egypt; Hamamatsu Photonics, Hamamatsu City, Japan), MEMS Hadamard masks (microPHAZIR, Thermo Fisher Scientific, Waltham, MA, USA), micro-mirror devices (DMD™) (Texas Instruments, Dallas, TX, USA; Innospectra Corp., Xinzhu, Taiwan, China), MEMS Fabry–Perot interferometers (Nirone Sensor S, Spectral Engines, Helsinki, Finland), NIR grating micro-spectrometers, (Insion GmbH, Obersulm, Germany; OrO Photonics, Xinzhu, Taiwan, China), and NIR scanners with 16 solar cell detectors (Senorics GmbH, Dresden, Germany) [23,24].
This diversity of technological solutions leads to different operational wavelength regions, spectral resolutions, and signal-to-noise ratios [23], which may affect the analytical performance or applicability to a specific matrix. However, in recent years, studies have shown that similar results can be obtained with benchtop and handheld devices for both quantitative and qualitative applications [25,26,27]. The performance of the handheld devices against the benchtop instruments should therefore be tested for each application, as they can be considered a reference for evaluating the analytical performance of portable spectrometers.
In this context, the objective of this study was to investigate the feasibility of predicting the sensory parameters of Iberian ham using different benchtop devices—Foss NIRSystem 5000 (Foss NIRSystems, Silver Spring, MD, USA) and BüchiNIRFlex N-500 (Büchi, Flawil, Switzerland)—and portable equipment—microPHAZIR (Thermo Fischer Scientific, USA); MicroNirProEs (VIAVI solutions, USA); SCiO (Consumer Physics, Israel); and Enterprise Sensor NIR (TellSpec, UK)—as they encompass a wide range of technological solutions, including different light sources, wavelength selectors, detectors, resolutions, and spectral intervals. On the other hand, the influence of recording factors (fat, lean meat, or whole slices) in order to determine the most appropriate methodology for determining the sensory profile easily and accurately was also studied. Achieving this objective will allow the use of NIRS handheld devices for sensory analysis to be implemented routinely in the meat sector.

2. Materials and Methods

2.1. Materials

A total of 60 samples of Iberian ham were selected, with 24 samples from the 100% Iberian acorn-fed category and 36 samples from the Iberian (Iberian x Duroc) acorn-fed category. All animals were raised in the same geographical area according to certificate RD 10/2014 [28], as previously described by Hernández-Jiménez et al. [27]. One hundred grams of 1 mm-thick ham slices (including Biceps femoris, Semimembranosus, and Semitendinosus muscles) were analyzed when taken perpendicularly to the bone at the same depth from the front of the hams.

2.2. Sensory Analysis

The sensory analysis of the samples was carried out according to the methodology described previously by Hernández-Ramos et al. [8] using a sensory panel of 10 members who had been trained in quantitative descriptive analysis (QDA) of cured ham. The assessors gave their written consent to participate before the start of the test. Ethical approval was obtained from the Ethics Committee of the University of Salamanca (registration number 1298). Prior to the analysis, the panel received training during six sessions using reference products and dummy samples. At the end of each session, the director provided feedback on performance using statistical descriptors. After the training phase, repeatability and reproducibility were calculated [29] by analyzing the same sample three times in three different sessions in order to evaluate the accuracy of the panel.
A total of 28 parameters were analyzed, including the visual, flavor, and texture properties of the samples (Supplementary Material, Table S1). A structured scoring scale was used, ranging from 1, low intensity, or the presence of the parameter evaluated, to 9, high intensity, or the presence of the parameter evaluated. The 60 samples, tempered to 20 ± 2 °C and coded, were presented and tasted individually during 15 sessions (4 samples per session).

2.3. Near-Infrared Spectroscopy Analysis

Prior to recording, the ham packages were tempered for two hours until the samples reached a temperature of 20 ± 2 °C. Four slices were separated from each package and placed on top of each other, facing in the same direction.
All samples were analyzed using two types of benchtop equipment, Foss NIRSystem 5000 (Foss NIRSystems, Silver Spring, MD, USA) and BüchiNIRFlex N-500 (Büchi, Flawil, Switzerland), and four portable devices: MicroNIR 1700 ES (VIAVI, Milpitas, CA, USA); Enterprise Sensor (TellSpec, Toronto, ON, Canada); SCiO Sensor (Consumer Physics, Tel Aviv, Israel); and microPHAZIR (Thermo Fisher Scientific, Waltham, MA, USA). The main features and technical parameters of the devices, in addition to the spectral range, recording interval, and the measurement of references, are described in the Supplementary Material (Table S2).
The NIR spectra were recorded by placing the device window directly on the surface of the portion. The Foss NIRSystem 5000 was equipped with a 5 cm × 5 cm quartz window so that the spectrum area included both lean muscle mass and fat. The spectra were recorded at three different points, and the average of these was considered to be the spectrum for the whole slice. In the case of the BüchiNIRFlex N-500 500 (Büchi, Flawil, Switzerland)microPHAZIR (Thermo Fisher Scientific, Waltham, MA, USA), MicroNIR 1700 ES (VIAVI, Milpitas, CA, USA), Enterprise Sensor (TellSpec, Toronto, ON, Canada), and SCiO (Consumer Physics, Tel Aviv, Israel) devices, six spectra of lean muscle and four spectra of fat were recorded. Subsequently, the mean of all replicates in each area was calculated to obtain a mean spectrum for fat and a mean spectrum for lean meat, and the mean spectrum for the entire slice was calculated as the mean of all spectra. Each sample was recorded by all devices in the same session.

2.4. Artificial Neural Network Calibration

The calibration curves for all sensory parameters were calculated by using a feedforward multilayer perceptron (MLP) artificial neural network (ANN), employing a Levenberg–Marquardt backpropagation training algorithm in accordance with the protocol previously described by Hernández-Ramos et al. [8].
The ANN selected consisted of an input layer that had as many neurons as NIR absorbance values. The ANN developed for the Foss NIRSystem 5000, therefore, had 451 neurons; for the BüchiNIRFlex N-500, it had 1501 neurons; for the MicroNIR 1700, it had 125 neurons; for the Enterprise Sensor, it had 256 neurons; and for the SCiO Sensor, it had 331 neurons, and for the microPHAZIR, it had 100 neurons. All ANNs had a hidden layer with a number of neurons ranging from 1 to 15. Finally, the output layer had only one neuron, which gave the predicted value for each parameter. The selected functions were the hyperbolic tangent sigmoid for the hidden layer and the pure linear transfer function for the output layer. For each scenario (device, recording area, and sensory parameter) and number of neurons in the hidden layer, the ANN was trained 100 times to obtain the best architecture using a known seed value as the initial weight and a bias matrix to allow the subsequent reproducibility of the data [30]. The pairs of input (absorbance value) and output (sensory score) data were randomly divided for all ANNs into three sets called the training set, validation set, and test set to represent 70%, 15%, and 15% of the data, respectively, i.e., a total of 42 samples in the training set and 9 samples in the validation and test sets. A total of 672,000 architectures were calculated (28 parameters, 5 devices, 3 recording sites, 100 training times, and between 1 and 15 neurons, plus 28 parameters, 1 device, 1 recording site, 100 training times, and between 1 and 15 neurons), and the best architecture was selected based on the highest R2 values and the lowest RMSE values. The RPD (ratio performance deviation) was calculated for the best final ANN architecture in order to assess the model performance. The Deep Learning Toolbox of MatLab (MathWorks®) version R2018 was the software used for all tests.

3. Results

3.1. Sensory Characteristics of the Samples Analyzed

The sensory characteristics or attributes of the samples analyzed by the sensory panel are shown in Table 1. Statistical analysis of the samples showed significant differences between the 100% Iberian and Iberian samples in four out of the six visual appearance parameters and seven out of the eight texture parameters, while significant differences were only observed in one of the parameters included in the flavor profile, rancid aroma. It is worth noting that, as described in the Materials and Methods, both groups of samples were processed in the same factory, following exactly the same protocol, and were cured in the same facilities for the same period of time. Therefore, the significant differences observed were mainly related to breed, as both groups were raised in the same geographical area and under the same management system, the montanera, in which the animals feed on grass and acorns during the fattening period.
The purebred Iberian samples were characterized by significantly less marbling, less color homogeneity, and fewer white spots, while color intensity was higher. These results are fully in agreement with those previously published by Hernández-Ramos et al. [8], who analyzed 100% Iberian and Iberian commercial hams. They attributed the increased presence of veining and exudate to the higher intramuscular fat content of the Iberian breed, as also reported by Mayoral et al. [31]. In terms of color, the 100% Iberian breed showed greater color intensity and less chromatic homogeneity, i.e., the presence of several shades in the same slice, which is consistent with the higher L* and a* values and higher myoglobin content of pure Iberian hams described by Fuentes et al. [32]. The samples analyzed were characterized by the presence of white spots, which are tyrosine and phenylalanine crystals that appear as a result of proteolysis when hams are aged for long periods [33]. Previous results have shown that the incidence of tyrosine crystals depends on the salinity, drying level, temperature, and the combined effect [34]. However, the results of both these samples and commercial samples [8] point to a significant influence of the breed on this phenomenon. In this regard, Carrapiso et al. [35] described that significant differences in bitterness, which are also related to the proteolysis process [36], were only related to crossbreeding.
Flavor parameters were not significantly affected by the breed, which is consistent with previous results showing no major significant differences between 100% and 50% Iberian pigs [8,35] attributed to genetic similarity. It should be noted that some attributes, such as saltiness, sourness, rancidity, or an atypical flavor, showed almost identical values, which shows that these parameters were more correlated with processing conditions. On the other hand, the odor intensity, cured aroma, rancidity, flavor intensity, and fat flavor intensity, although they showed no statistically significant differences, were slightly higher in 50% Iberian ham, which is probably due to the higher fat content. It is well-known that lipolysis, fatty acid oxidation, and Maillard reactions constitute the main processes contributing to the generation of the volatile responsible for the odor and flavor of cured ham [37,38,39]. In addition, proteolysis is also involved in the generation of volatile compounds [39] and the formed products formed either contribute directly to the taste [40] or are precursors of odorants [41]. Differences in proteolysis related to crossbreeding could, therefore, also affect the flavor profile.
Finally, texture attributes were affected by genetics in the same way, as previously reported by Hernández-Ramos et al. [8], which demonstrates the clear effect of crossbreeding on the texture of Iberian ham. The 50% Iberian ham samples were significantly less tough and fibrous and had lower chewiness and gumminess values, although these were not statistically significant. These samples also showed significantly higher levels of juiciness and fatness. Therefore, these results confirm the correlations between the marbling attribute and the texture parameters previously described by Hernández-Ramos et al. [8], which, as mentioned above, were attributed to the higher intramuscular fat content of the 50% Iberian samples, which increases juiciness and decreases hardness and fibrousness [42]. In addition, previous studies found a close relationship between the fat content and the sensory measurements of fat [35,43]. Furthermore, proteolytic changes in myofibrillar and sarcoplasmic proteins are crucial for the development of texture and sensory characteristics [44]. This process is affected by many factors, such as breeding conditions, the characteristics of the raw product, and the maturation process [45], but also by pig genetics [46] and lipid oxidation [47], as observed in this study.

3.2. Spectral Characteristics of the Samples Analyzed

Previous studies have shown that, although distinctive characteristics and differences can be observed between the spectra measured by benchtop and miniaturized spectrometers [48], the shapes of the absorption lines obtained with different equipment show similarities [23,49]. Therefore, Figure 1 shows the average spectra of 100% Iberian and 50% Iberian hams recorded for the lean meat, the fatty part of the slices, and the spectra of the whole slices, calculated as the average value of the spectra obtained for the fat and lean meat [27] by the NIRFlex N-500 equipment, which has the widest spectral range.
Differences owing to the breed of pig are more easily observed in the fat spectra, as in lean meat, they are less clear, and in whole slices, they are barely perceptible. In the first case, the 50% Iberian samples showed higher absorbance in the ranges between 1000 and 1600 nm and between 1800 and 2000 nm, which could be related to the higher fat content of the 50% Iberian samples, while in the 2100–2500 nm range, the 100% Iberian samples showed slightly higher absorbance. However, the 100% Iberian samples showed higher absorbance for all wavelengths when the spectra were recorded in lean meat.
The maximum absorbance peaks observed for the three recording sites at 1200 nm correspond to the second overtone of the C-H stretching vibration [50], and the peaks observed at 1730 and 1762 correspond to the first C-H stretching overtone. The lack of clear differences observed for these peaks corresponding to fatty acid composition could be due to the high number of shared absorbing molecular groups present in these compounds [51,52]. The combination tones of the C-H stretching and bending modes appear at 2310 and 2350 nm [53]. The NIR region between 2000 and 2300 is where the distinction between saturated and unsaturated compounds can most clearly be observed, as the specificity of the combined bands of unsaturated acids is manifested in this region [54]. The maximum peaks observed at 1450 and 1950 nm would be related to the -OH groups corresponding to the third, second, and first overtones of the stretching mode [55]. However, within this range, it is also possible to find N-H absorption bands around 1500, 1950, and 2450 nm, but most of them are rather difficult to discern, owing to their proximity to the absorption bands of fat and water, as the bands are very broad [53].
As regards the correlation between spectral characteristics and sensory parameters, the work of Hernández-Ramos et al. [8] showed a correlation between absorbance in the vicinity of 1500 nm and texture parameters such as hardness (1502, 1518, and 1530), fibrousness (1518 and 1528), and heterogeneity (1510, 1534), but also for 1720 and fibrousness, with all these parameters being significantly higher in the 100% Iberian samples. These wavelengths are related to the SH-SH bond (1736 nm) or to the absorption of amines, amides, urea, or proteins, and the 100% Iberian samples showed slightly higher absorbance when the spectra were taken from lean meat. On the other hand, the 50% Iberian samples were characterized by a greater sensation of fat, which showed a correlation with 1162 and 1720 nm, which, in turn, are correlated with C=O and CO-Oil bonds, but also with 1488 and 1520 nm, which correspond to amides, amines, and urea. The 50% Iberian samples showed significantly higher fatness values and slightly higher absorbance at 1500 and 1700 nm in the spectra of fat and the whole slices.

3.3. Prediction of Sensory Parameters

As previously reported, sensory characteristics are complex traits that require non-linear statistical approaches [56], such as artificial neural networks, which have proven to be a very suitable tool for meat products [8,19], as they provide better R2 than multiple regression tools [18]. Taking into account the results of previous studies on Iberian ham, a feedforward MLP ANN device with the Levenberg–Marquardt backpropagation training algorithm was used to predict the sensory parameters. In addition, 100 training times and a maximum number of neurons in the hidden layer of 15 were tested. Previous studies revealed that the greater the number of training sessions, the lower the number of neurons in the hidden layer required to obtain the highest R2. Therefore, for such a high number of training sessions, it would not be necessary to test a larger number of neurons. The best ANN architecture was selected based on the highest R2 values and the lowest RMSE values (training, validation, test, and total) (Table 2, Table 3 and Table 4). R2Train indicates the way in which the network is able to predict results using the data with which it has been trained, R2Valid gives information about the network’s overfitting, R2Test measures the performance of the network and its ability to predict unknown data, and R2Total measures the result of analyzing the data from all samples obtained through the network. Therefore, in this study, the R2Test was prioritized, as it is more useful for evaluating the performance of the network once it is operational, and subsequently the R2Total, R2Train, and R2Valid.
The joint analysis of the results obtained for all parameters, devices, and recording sites revealed that R2Train showed very high values, almost always above 0.8 and very often above 0.9. This is because during the training process, the errors between the data predicted and observed decrease until they reach a stable minimum. The network is then able to predict the training data with great accuracy. To calculate R2Valid, the network is tested with data that do not belong to the training set. After several passes through the training set data, the network is tested with data from the validation set, which was not used during training. As long as the validation errors continue to decrease, the network is learning. If, on the contrary, and as generally observed in this study, the errors increase during the validation process, this means that the network is overfitting. Therefore, the low values obtained in general for R2Valid showed that the networks were learning but had problems in generalizing their performance to other data, i.e., they were overfitting. As for the R2Test, this parameter tended to have high values, as the ANN architecture was selected in accordance with it. Since R2Total is the result of all the data passing through the network obtained, in general, satisfactory values were observed. It should also be noted that for some ANN architectures, apparently contradictory results were obtained, in which, although very good values were obtained during training, the validation values were very low, with the remainder of the values within a very high range. This was due to the small number of samples (60), which were divided into three data sets at random so that the validation and test sets included nine samples (15% of the samples). The results were therefore strongly affected by the type of samples included in each set for that ANN.

3.3.1. Prediction of Sensory Parameters Using Spectra Recorded in Lean Meat

Six spectra were recorded at different points of the lean meat part of the slices by the BüchiNIRFlex N-500, MicroNIR 1700 ES, Enterprise Sensor, and microPHAZIRand SCiO devices, and the average spectrum for each sample was calculated and used for ANN analysis. The correlation coefficients (R2) and root mean square error (RMSE) for each device and sensory parameter are shown in Table 2. The results of the appearance, flavor, and texture parameters are presented separately.
The following levels of interpretations were used as a reference: R2 > 0.9 was considered excellent, R2 > 0.8 as good, R2 > 0.7 as acceptable, R2 > 0.6 as moderate, and R2 > 0.5 as fair. In addition to this, according to Williams [57], a calibration with an RSQ of <0.5 is considered unsuccessful, while an RSQ of between 0.50 and 0.65 indicates the possibility of distinguishing between high and low values. Therefore, for this study, a minimum RSQ value of >0.5 was chosen, as it points to the possibility of a successful calibration and can also be considered as an initial reference for comparing the feasibility of the prediction of the different devices used. On the other hand, an RSQ of 0.5–0.6 can be considered prospective. In this sense, high values for some attributes, such as the juiciness, softness, cured flavor, flavor intensity, and sweetness, are valued as positive, while high fibrousness, mold odor, and saltiness are valued as negative by consumers [4,58]. Therefore, high or low values for these parameters can be correlated with the acceptability of the cured ham, bearing in mind that this correlation is not always linear.
As regards appearance parameters (Table 2 (a)), all devices showed an R2Total of >0.5 for all parameters, except microPHAZIR for color intensity. However, the number of parameters with an R2total of >0.7 was lower and depended on the device. It was thus possible to predict four out of six parameters with NIRFlex N-500 (veined, fat color, color homogeneity, and color intensity), with the former two showing an R2total of >0.8. For MicroNIR, veined, color homogeneity, and color intensity showed an R2Total of >0.7, with the same color homogeneity for microPHAZIR, and only veined for Enterprise. Therefore, veined and color homogeneity were the parameters that reached the highest R2Total values for this recording site, while the exudate showed the lowest values. It should be noted that the color homogeneity and exudate tended to have a higher number of neurons in the hidden layer, while the Enterprise device tended to show fewer neurons.
A total of 14 flavor parameters were analyzed (Table 2 (b)), and the results showed that 10 parameters for SCiO and 13 parameters for NIRFlex N-500 (all except aftertaste) were predicted with an R2Total of >0.5. As with the appearance, the number of flavor parameters that showed an R2Total of >0.7 was much lower, ranging from six for NIRFlex N-500 and microPHAZIR, including rancidity or sweetness with an R2Total of >0.8, to four for MicroNIR (sweetness, R2Total of >0.8), Enterprise (aftertaste, R2total of >0.8), and SCiO (pig aroma, R2total of >0.8). Flavor intensity was the best-predicted parameter, as all devices showed an R2Total of >0.7, followed by pig aroma, while rancidity showed the worst results, as only NIRFlex N-500 and microPHAZIR showed an R2Total of >0.5. For odor intensity, atypical aroma, and cured flavor, none of the devices showed an R2Total of >0.7, although almost all of these showed an R2Total of >0.5. In this case, in general, the cured aroma and atypical aroma showed a higher number of neurons in the hidden layer, while Enterprise and MicroNIIR tended to have a lower number of neurons.
Finally, eight texture parameters were also calibrated (Table 2 (c)). All devices were able to predict seven out of eight texture parameters with an R2Total of >0.5, except for microPHAZIR, which only showed this R2Total value for four parameters. For this group of sensory attributes, NIRFlex N-500, MicroNIR, and SCiO showed an R2Total of >0.7 for four parameters and an R2total of >0.8 for three parameters in the case of NIRFlex N-500 or two parameters for MicroNIR and SCiO (hardness, juiciness, and fibrousness in general). Hardness, juiciness, fatness, and fibrousness were predicted by all the devices with an R2Total of >0.5, with fibrousness and juiciness having the highest R2Total coefficients, while chewiness and homogeneity were predicted by all the devices except microPHAZIR. On the other hand, the parameter with the poorest prediction was chewing residue, as only NIRFlex N-500 showed an R2Total of >0.5. This attribute also showed the highest number of neurons in the hidden layer. Overall, the number of parameters with an R2Total of >0.5 was higher for appearance attributes, although the texture parameters showed slightly higher R2Total values for this recording site.
Similar results were observed when RPD was considered (Table S3). NIRFlex N500, therefore, showed the highest number of parameters with high RPD values, with the Enterprise device having, in general, the lowest values for this parameter, followed by the SCiO device.

3.3.2. Prediction of Sensory Parameters Using Spectra Recorded on Fat

As described above, for the spectral recording of fat, four spectra were taken from the fatty areas within the slices, and the average value of each sample was calculated. The results of calibration with ANN using sensory scores determined by the evaluators as reference analysis are shown in Table 3. For this recording site, it can be observed that all the parameters (except exudate for NIRFlex N-500, which showed an R2total of 0.565) were predicted with an R2Total of >0.6. MicroPHAZIR showed an R2Total of >0.8 for three parameters (veined, the fat color, and white dots), while NIRFlex N-500 obtained these results for only two (color intensity and white dots), and Enterprise for only one, exudate (Table 3 (a)). In this case, the best-predicted parameters were, in general, white dots with an R2total of >0.7 for most devices, followed by the fat color, while the exudate showed a higher R2total than those observed when the calibration was made on lean meat, as the R2totals were higher than 0.6 for most of the devices.
As for the odor and flavor parameters (Table 3 (b)), the NIRFlex N-500, microPHAZIR, and Enterprise devices were able to predict 13 out of 14 parameters with an R2total of >0.5, while MicroNIR showed this R2total value for 12 parameters, and SCiO only for 8 parameters. It is worth mentioning the best performance of NIRFlex N-500 for this set of parameters, as it showed an R2total of >0.7 for six parameters (the pig aroma, rancid aroma, atypical aroma, fat flavor intensity, sweetness, and aftertaste) and an R2total of >0.8 for two. The remainder of the devices showed an R2total of >0.7 between four parameters and one parameter, with the latter being the SCiO equipment, which showed the poorest results among the devices used. The parameters that were predicted by a higher number of devices with the highest R2total values were the atypical aroma, followed by the pig aroma, aftertaste, and fat flavor intensity. On the other hand, the lowest R2totals were obtained for rancid and cured aromas.
As far as texture parameters are concerned (Table 3 (c)), the most relevant result was that all devices predicted hardness with an R2Total of >0.8, reaching values of >0.9 for NIRFlex N-500 and microPHAZIR. These high R2Total values were the result of very high values also for R2Train, R2Valid, and R2Test, which reveals the feasibility of predicting this sensory parameter by NIRS spectroscopy. Juiciness, fibrousness, and fatness also attained high R2Total values of between 0.7 and 0.8 for all the devices except for SCiO (fatness). In contrast, gumminess showed very low R2total values, with an R2Total of >0.5 only observed for NIRFlex N-500 and microPHAZIR.
The best prediction performance was shown by NIRFlex N-500, as all the texture parameters were predicted with an R2total of >0.6, with six of them with an R2total of >0.7 and two of them with an R2Total of >0.8, closely followed by microPHAZIR, which also predicted all the parameters with an R2Total of >0.6. On the other hand, SCiO was only able to predict four out of eight values with an R2total of >0.6 (six parameters with an R2total of >0.5), and only hardness showed an R2Total of >0.8.
These results demonstrated the poorer performance of the SCiO device when the spectra were recorded in the fatty areas of the ham slices, and that, in general, texture parameters were better predicted than appearance and flavor parameters for this recording site. These results were also observed when RPD was considered to assess the model performance (Table S3). Furthermore, no clear trend was observed in terms of the number of neurons in the hidden layer; neither of the parameters with a lower R2Total showed a higher number, nor was there any device that tended to show a lower (or higher) number of neurons.

3.3.3. Prediction of Sensory Parameters Using the Whole Slice Spectra

Finally, since fat and lean meat are consumed together, the spectrum of the whole slice was calculated for each sample as the average of the ten spectra recorded for the slices, except in the case of the Foss 5000. The calibration results, correlation coefficients (R2), and the root mean square error (RMSE) are shown in Table 4.
As was observed for the lean meat and fat recording sites, the number of calibrated parameters depended largely on the device. NIRFlex N-500, MicroNIR, and Enterprise therefore showed an R2Total of >0.5 for all the appearance parameters (Table 4 (a)), while Foss 5000 and microPHAZIR showed this value for five out of six parameters, and SCiO showed this value only for three parameters. The number of parameters showing an R2Total of >0.7 ranged from five (microPHAZIR) to one (SCiO and Foss 5000). However, it should be noted that NIRFlex N-500 showed an R2Total of >0.8 for three parameters, which was the highest value among the devices used. The best performance of this equipment was corroborated by the higher RPD values observed for this set of sensory characteristics. The parameters that showed the highest R2Totals for all the devices tested were veined, followed by color intensity, which was predicted by all of them, and color homogeneity. On the other hand, the exudate showed the lowest R2total values and the highest number of neurons in the hidden layer.
In terms of flavor attributes (Table 4 (b)), NIRFlex N-500 again showed the highest number of predicted parameters (13), followed by microPHAZIR and MicroNIR (10), while SCiO showed the lowest performance. In fact, NIRFlex N-500 showed an R2total of >0.7 for eight parameters, while the other devices showed a maximum of four parameters with this value. The best-predicted parameters were the saltiness, followed by the aftertaste and odor intensity, which were predicted by all the devices, while the poorest results were observed for rancidity, as only NIRFlex N-500 showed an R2total of >0.5.
The results of texture parameter prediction (Table 4 (c)) revealed that NIRFlex N-500 showed an R2total of >0.6 for the eight parameters analyzed and an R2total of >0.8 for five of them. On the other hand, MicroNIR and SCiO had an R2total of >0.5 for all the parameters. However, MicroNIR showed an R2total of >0.7 for six parameters, while SCiO had this value for only four parameters. For this set of parameters, the Enterprise equipment showed the lowest prediction capacity, as only five parameters had an R2total of >0.5. Fibrousness was the best-predicted parameter, followed by juiciness and heterogeneity, while gumminess and chewing residue displayed the lowest R2total values. In this case, no clear trend regarding the number of neurons in the hidden layer was observed. Therefore, according to the RPD value, for this recording site, the best performance was observed for NIRFlex 500, while Foss 5000 did not show high RPD values. The high RPD values of the MicroPHAZIR device, followed by the MicroNIR device, are noteworthy.
The R2 values were generally lower than those previously reported for ham when whole slices were tested [8,21], and they were also lower than the R2 obtained for other meat products, such as dry-cured beef or loin [19,20], owing to the lower number and variability of the samples analyzed in this study and the lack of previous homogenization. For all recording sites and devices, in general, the highest R2 values were observed for texture parameters and the lowest R2 values were found for flavor parameters, which is in agreement with the results reported by Revilla et al. [19] and Curto et al. [59], while portable devices showed similar R2 values for both appearance and texture parameters, as reported by Vasconcelos et al. [20].
Table 5 summarizes the total number of parameters predicted by each device used, depending on the fixed R2Total (>0.5, >0.6, >0.7, or >0.8), grouped by appearance, flavor, or texture attributes.
It can be observed that the best recording site depended strongly on the device. The NIRFlex N-500 benchtop device gave the best performance among all the devices used, which coincides with the previous results [27,49,60,61] for all recordings, with particularly high values for the entire slice registration methodology, with 27 of the 28 parameters predicted with an R2Total of >0.5 for all recording sites and 18 parameters showing an R2total of >0.7. These results mean that this benchtop equipment could be considered as a reference. However, the FOSS 5000 benchtop device did not perform as well as the previous one, with 20 parameters showing an R2Total of >0.5 and only two parameters showing R2Total values of >0.8, i.e., both microPHAZIR and MicroNIR achieved better results.
As for portable devices, when the spectra were recorded for whole slices, the MicroNIR performed very well, with 25 parameters showing an R2total of >0.5 and 13 parameters reaching an R2total of >0.7. When the spectra were recorded for fat, microPHAZIR and Enterprise showed the highest number of parameters predicted, with 27 and 26 parameters, respectively, predicted with an R2total of >0.5 and 13 parameters predicted with an R2total of >0.7 in both cases. On the other hand, MicroNIR and SCiO demonstrated better performance when spectra were recorded for lean meat with 25 and 23 parameters, with an R2Total of >0.5 and 11 and 10 parameters with an R2Total of >0.7, respectively. These results differ, in general, from those previously reported for fatty acids, which were better predicted for lean meat except in the case of SCIO, which showed better performance for fat [27].
The results, therefore, suggest that, in general, the parameters predicted with the highest R2total depended on the recording area, except for MicroNIR, which showed very consistent results for all sites, which indicates that measurements taken in different areas provide complementary information, as previously reported by Hernández-Jiménez et al. [24]. Accordingly, NIRFlex N-500 was able to predict a total of 16 different sensory parameters with an R2Total of >0.8 (veined, fat color, color intensity, color homogeneity, white dots, rancidity, typical aroma, cured flavor, cured aroma, saltiness, hardness, juiciness, fatness, fibrousness, chewiness, and heterogeneity). Among the portable devices, microPHAZIR showed a total of nine parameters (veined, color homogeneity, fat color, sweetness, aftertaste, cured aroma, hardness, juiciness, and fibrousness), Enterprise showed eight parameters (veined, exudate, aftertaste, cured flavor, saltiness, hardness, juiciness, and fibrousness), MicroNIR1700 showed a total of seven parameters (veined, color homogeneity, fat color, sweetness, sourness, hardness, and juiciness), and SCiO showed a total of four different attributes (pig aroma, atypical aroma, saltiness, and hardness) when using the information from all the recording possibilities. It is worth highlighting that some of these characteristics, such as hardness, fibrousness, juiciness, color homogeneity, flavor intensity, sweetness, and fat color, are decisive in consumer acceptance of Iberian ham [4,58].

4. Conclusions

These results allowed us to conclude that, on the one hand, when using a portable device, it is essential to optimize the recording area, but, on the other hand, the results obtained at different recording sites are complementary. Among the sensory parameters analyzed in this study, veins, color homogeneity, color intensity, hardness, juiciness, fatness, and fibrousness showed statistically significant differences owing to the breed, which makes them particularly relevant to the characterization of Iberian ham. These parameters were predicted by all devices and at all recording sites. The best results were obtained with the NIRFlex N-500, with between 26 and 27 out of the 28 parameters predicted. Among all the portable devices used, MicroNIR and microPHAZIR gave the best results, as they were able to predict the aforementioned parameters with high R2 values in addition to the parameters of flavor intensity, sweetness, and fat color, which are decisive for consumer acceptance of Iberian ham. These results point to the suitability of portable devices for predicting the sensory properties of Iberian ham. However, a larger set of samples with greater variability must be analyzed to improve the performance of the ANN, especially the R2 values of the validation set. This would allow a broader application of the methods developed.

Supplementary Materials

The following supporting information can be downloaded at https://www.mdpi.com/article/10.3390/foods15030436/s1: Table S1. Sensory parameters evaluated by the assessors and the definition of sensory parameters; Table S2. Technical characteristics of the devices used in the study; Table S3: Ratio performance deviation (RPD) values for the best ANN architecture.

Author Contributions

I.R.: conceptualization, project management, fundraising, and original draft-writing. A.M.V.-Q.: fundraising, supervision, and writing—revising and editing. I.M.-M.: formal analysis and research. P.H.-R.: software and data curation. M.H.-J.: formal analysis, research, visualization, and original draft. J.G.: validation and review and editing. K.B.B.: methodology, validation, and review and editing. C.W.H.: resources, supervision, and review and editing. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by Diputación de Salamanca (Spain) under grant number 18VEUH 463AC06.

Institutional Review Board Statement

The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of the University of Salamanca (protocol code 1298 of 22 January 2025) for studies involving humans.

Informed Consent Statement

Informed consent was obtained from all subjects involved in the study.

Data Availability Statement

The data presented in this study are openly available in the Gredos repository: http://hdl.handle.net/10366/168068 (accessed on 21 January 2026).

Acknowledgments

The authors are grateful to “Carrasco Ibéricos” Guijuelo (Salamanca) for their collaboration, and Hernández-Jiménez M. is grateful to the Own Program III: Grants for Pre-doctoral Contracts of the University of Salamanca, co-funded by Banco Santander and Grant COST Action: CA19145.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Lorido, L.; Hort, J.; Estévez, M.; Ventanas, S. Reporting the Sensory Properties of Dry-Cured Ham Using a New Language: Time Intensity (TI) and Temporal Dominance of Sensations (TDS). Meat Sci. 2016, 121, 166–174. [Google Scholar] [CrossRef]
  2. Tejerina, D.; García-Torres, S.; Cabeza de Vaca, M.; Vázquez, F.M.; Cava, R. Effect of Production System on Physical–Chemical, Antioxidant and Fatty Acids Composition of Longissimus Dorsi and Serratus Ventralis Muscles from Iberian Pig. Food Chem. 2012, 133, 293–299. [Google Scholar] [CrossRef]
  3. Díaz-Caro, C.; García-Torres, S.; Elghannam, A.; Tejerina, D.; Mesias, F.J.; Ortiz, A. Is Production System a Relevant Attribute in Consumers’ Food Preferences? The Case of Iberian Dry-Cured Ham in Spain. Meat Sci. 2019, 158, 107908. [Google Scholar] [CrossRef]
  4. Ruiz, J.; García, C.; Muriel, E.; Andrés, A.I.; Ventanas, J. Influence of Sensory Characteristics on the Acceptability of Dry-Cured Ham. Meat Sci. 2002, 61, 347–354. [Google Scholar] [CrossRef] [PubMed]
  5. Caballero, D.; Antequera, T.; Caro, A.; Duran, M.L.; Perez-Palacios, T. Data Mining on MRI-Computational Texture Features to Predict Sensory Characteristics in Ham. Food Bioprocess Technol 2016, 9, 699–708. [Google Scholar] [CrossRef]
  6. González-Casado, A.; Jiménez-Carvelo, A.M.; Cuadros-Rodríguez, L. Sensory Quality Control of Dry-Cured Ham: A Comprehensive Methodology for Sensory Panel Qualification and Method Validation. Meat Sci. 2019, 149, 149–155. [Google Scholar] [CrossRef]
  7. Guàrdia, M.D.; Aguiar, A.P.S.; Claret, A.; Arnau, J.; Guerrero, L. Sensory Characterization of Dry-Cured Ham Using Free-Choice Profiling. Food Qual. Prefer. 2010, 21, 148–155. [Google Scholar] [CrossRef]
  8. Hernández-Ramos, P.; Vivar-Quintana, A.M.; Revilla, I.; González-Martín, M.I.; Hernández-Jiménez, M.; Martínez-Martín, I. Prediction of Sensory Parameters of Cured Ham: A Study of the Viability of the Use of NIR Spectroscopy and Artificial Neural Networks. Sensors 2020, 20, 5624. [Google Scholar] [CrossRef] [PubMed]
  9. Pérez-Palacios, T.; Ruiz, J.; Martín, D.; Barat, J.M.; Antequera, T. Pre-Cure Freezing Effect on Physicochemical, Texture and Sensory Characteristics of Iberian Ham. Food Sci. Technol. Int. 2011, 17, 127–133. [Google Scholar] [CrossRef] [PubMed]
  10. ISO 5492:2008; Sensory Analysis. Vocabulary. International Organization for Standardization: Geneva, Switzerland, 2008.
  11. ISO 6658:2005; Sensory Analysis. Methodology. General Guidance. International Organization for Standardization: Geneva, Switzerland, 2005.
  12. ISO 8586:2014; Sensory Analysis. General Guidelines for the Selection, Training and Monitoring of Selected Assessors and Expert Sensory Assessors. International Organization for Standardization: Geneva, Switzerland, 2014.
  13. Dos Santos Navarro da Silva, R.d.C.; Minim, V.P.R.; Silva, A.N.d.; Peternelli, L.A.; Minim, L.A. Optimized Descriptive Profile: How Many Judges Are Necessary? Food Qual. Prefer. 2014, 36, 3–11. [Google Scholar] [CrossRef]
  14. De Pilli, T.; Alessandrino, O.; Baiano, A. Quantitative Descriptive Analysis as a Strategic Tool in Research Activities Relating to Innovative Meat Tenderisation Technologies. Heliyon 2024, 10, e32618. [Google Scholar] [CrossRef]
  15. Chapman, J.; Elbourne, A.; Truong, V.K.; Cozzolino, D. Shining Light into Meat—A Review on the Recent Advances in In Vivo and Carcass Applications of near Infrared Spectroscopy. Int. J. Food Sci. Technol. 2020, 55, 935–941. [Google Scholar] [CrossRef]
  16. Chapman, J.; Elbourne, A.; Truong, V.K.; Newman, L.; Gangadoo, S.; Rajapaksha Pathirannahalage, P.; Cheeseman, S.; Cozzolino, D. Sensomics—From Conventional to Functional NIR Spectroscopy—Shining Light over the Aroma and Taste of Foods. Trends Food Sci. Technol. 2019, 91, 274–281. [Google Scholar] [CrossRef]
  17. Prieto, N.; Roehe, R.; Lavín, P.; Batten, G.; Andrés, S. Application of near Infrared Reflectance Spectroscopy to Predict Meat and Meat Products Quality: A Review. Meat Sci. 2009, 83, 175–186. [Google Scholar] [CrossRef] [PubMed]
  18. Hernández-Jiménez, M.; Hernández-Ramos, P.; Martínez-Martín, I.; Vivar-Quintana, A.M.; González-Martín, I.; Revilla, I. Comparison of Artificial Neural Networks and Multiple Regression Tools Applied to near Infrared Spectroscopy for Predicting Sensory Properties of Products from Quality Labels. Microchem. J. 2020, 159, 105459. [Google Scholar] [CrossRef]
  19. Revilla, I.; Vivar-Quintana, A.M.; González-Martín, M.I.; Hernández-Jiménez, M.; Martínez-Martín, I.; Hernández-Ramos, P. NIR Spectroscopy for Discriminating and Predicting the Sensory Profile of Dry-Cured Beef “Cecina”. Sensors 2020, 20, 6892. [Google Scholar] [CrossRef]
  20. Vasconcelos, L.; Dias, L.G.; Leite, A.; Ferreira, I.; Pereira, E.; Bona, E.; Mateo, J.; Rodrigues, S.; Teixeira, A. Can Near-Infrared Spectroscopy Replace a Panel of Tasters in Sensory Analysis of Dry-Cured Bísaro Loin? Foods 2023, 12, 4335. [Google Scholar] [CrossRef]
  21. García-Rey, R.M.; García-Olmo, J.; De Pedro, E.; Quiles-Zafra, R.; Luque de Castro, M.D. Prediction of Texture and Colour of Dry-Cured Ham by Visible and near Infrared Spectroscopy Using a Fiber Optic Probe. Meat Sci. 2005, 70, 357–363. [Google Scholar] [CrossRef] [PubMed]
  22. Ortiz, M.C.; Sarabia, L.; García-Rey, R.; Luque de Castro, M.D. Sensitivity and Specificity of PLS-Class Modelling for Five Sensory Characteristics of Dry-Cured Ham Using Visible and near Infrared Spectroscopy. Anal. Chim. Acta 2006, 558, 125–131. [Google Scholar] [CrossRef]
  23. Beć, K.B.; Grabska, J.; Huck, C.W. Principles and Applications of Miniaturized Near-Infrared (NIR) Spectrometers. Chem. A Eur. J. 2021, 27, 1514–1532. [Google Scholar] [CrossRef] [PubMed]
  24. Yan, H.; Neves, M.D.G.; Noda, I.; Guedes, G.M.; Ferreira, A.C.S.; Pfeifer, F.; Chen, X.; Siesler, H.W.; Yan, H.; Neves, M.D.G.; et al. Handheld Near-Infrared Spectroscopy: State-of-the-Art Instrumentation and Applications in Material Identification, Food Authentication, and Environmental Investigations. Chemosensors 2023, 11, 272. [Google Scholar] [CrossRef]
  25. de Lima, G.F.; Andrade, S.A.C.; da Silva, V.H.; Honorato, F.A. Multivariate Classification of UHT Milk as to the Presence of Lactose Using Benchtop and Portable NIR Spectrometers. Food Anal. Methods 2018, 11, 2699–2706. [Google Scholar] [CrossRef]
  26. Malegori, C.; Nascimento Marques, E.J.; de Freitas, S.T.; Pimentel, M.F.; Pasquini, C.; Casiraghi, E. Comparing the Analytical Performances of Micro-NIR and FT-NIR Spectrometers in the Evaluation of Acerola Fruit Quality, Using PLS and SVM Regression Algorithms. Talanta 2017, 165, 112–116. [Google Scholar] [CrossRef]
  27. Hernández-Jiménez, M.; Revilla, I.; Vivar-Quintana, A.M.; Grabska, J.; Beć, K.B.; Huck, C.W. Performance of Benchtop and Portable Spectroscopy Equipment for Discriminating Iberian Ham According to Breed. Curr. Res. Food Sci. 2024, 8, 100675. [Google Scholar] [CrossRef] [PubMed]
  28. Real Decreto 4/2014. de 10 de Enero. Por El Que Se Aprueba La Norma de Calidad Para La Carne. El Jamón. La Paleta y La Caña de Lomo Ibérico. 2014. Available online: https://www.boe.es/eli/es/rd/2014/01/10/4/con (accessed on 21 January 2026).
  29. Pérez Elortondo, F.J.; Ojeda, M.; Albisu, M.; Salmerón, J.; Etayo, I.; Molina, M. Food Quality Certification: An Approach for the Development of Accredited Sensory Evaluation Methods. Food Qual. Prefer. 2007, 18, 425–439. [Google Scholar] [CrossRef]
  30. Pillonel, L.; Badertscher, R.; Casey, M.; Meyer, J.; Rossmann, A.; Schlichtherle-Cerny, H.; Tabacchi, R.; Bosset, J.O. Geographic Origin of European Emmental Cheese: Characterisation and Descriptive Statistics. Int. Dairy J. 2005, 15, 547–556. [Google Scholar] [CrossRef]
  31. Mayoral, A.I.; Dorado, M.; Guillén, M.T.; Robina, A.; Vivo, J.M.; Vázquez, C.; Ruiz, J. Development of Meat and Carcass Quality Characteristics in Iberian Pigs Reared Outdoors. Meat Sci. 1999, 52, 315–324. [Google Scholar] [CrossRef]
  32. Fuentes, V.; Estévez, M.; Grèbol, N.; Ventanas, J.; Ventanas, S. Application of Time–Intensity Method to Assess the Sensory Properties of Iberian Dry-Cured Ham: Effect of Fat Content and High-Pressure Treatment. Eur. Food Res. Technol. 2014, 238, 397–408. [Google Scholar] [CrossRef]
  33. Arnau, J.; Guerrero, L.; Hortós, M.; García-Regueiro, J.A. The Composition of White Film and White Crystals Found in Dry-Cured Hams. J. Sci. Food Agric. 1996, 70, 449–452. [Google Scholar] [CrossRef]
  34. Coll-Brasas, E.; Gou, P.; Arnau, J.; Olmos, A.; Fulladosa, E. Processing Parameters Involved in the Development of Texture and Tyrosine Precipitates in Dry-Cured Ham: Modelisation of Texture Development. Meat Sci. 2021, 172, 108362. [Google Scholar] [CrossRef]
  35. Carrapiso, A.I.; Bonilla, F.; García, C. Effect of Crossbreeding and Rearing System on Sensory Characteristics of Iberian Ham. Meat Sci. 2003, 65, 623–629. [Google Scholar] [CrossRef]
  36. Ruiz, J.; García, C.; del Carmen Díaz, M.; Cava, R.; Florencio Tejeda, J.; Ventanas, J. Dry Cured Iberian Ham Non-Volatile Components as Affected by the Length of the Curing Process. Food Res. Int. 1999, 32, 643–651. [Google Scholar] [CrossRef]
  37. Bermúdez, R.; Franco, D.; Carballo, J.; Lorenzo, J.M. Influence of Type of Muscle on Volatile Compounds throughout the Manufacture of Celta Dry-Cured Ham. Food Sci. Technol. Int. 2015, 21, 581–592. [Google Scholar] [CrossRef]
  38. Fulladosa, E.; Garriga, M.; Martín, B.; Guàrdia, M.D.; García-Regueiro, J.A.; Arnau, J. Volatile Profile and Microbiological Characterization of Hollow Defect in Dry-Cured Ham. Meat Sci. 2010, 86, 801–807. [Google Scholar] [CrossRef]
  39. Pérez-Santaescolástica, C.; Carballo, J.; Fulladosa, E.; Garcia-Perez, J.V.; Benedito, J.; Lorenzo, J.M. Effect of Proteolysis Index Level on Instrumental Adhesiveness, Free Amino Acids Content and Volatile Compounds Profile of Dry-Cured Ham. Food Res. Int. 2018, 107, 559–566. [Google Scholar] [CrossRef] [PubMed]
  40. Jurado, Á.; García, C.; Timón, M.L.; Carrapiso, A.I. Effect of Ripening Time and Rearing System on Amino Acid-Related Flavour Compounds of Iberian Ham. Meat Sci. 2007, 75, 585–594. [Google Scholar] [CrossRef] [PubMed]
  41. Hidalgo, F.J.; Zamora, R. Strecker-Type Degradation Produced by the Lipid Oxidation Products 4,5-Epoxy-2-Alkenals. J. Agric. Food Chem. 2004, 52, 7126–7131. [Google Scholar] [CrossRef]
  42. Ruiz-Carrascal, J.; Ventanas, J.; Cava, R.; Andrés, A.I.; García, C. Texture and Appearance of Dry Cured Ham as Affected by Fat Content and Fatty Acid Composition. Food Res. Int. 2000, 33, 91–95. [Google Scholar] [CrossRef]
  43. Fuentes, V.; Ventanas, J.; Morcuende, D.; Ventanas, S. Effect of Intramuscular Fat Content and Serving Temperature on Temporal Sensory Perception of Sliced and Vacuum Packaged Dry-Cured Ham. Meat Sci. 2013, 93, 621–629. [Google Scholar] [CrossRef] [PubMed]
  44. Bermúdez, R.; Franco, D.; Carballo, J.; Lorenzo, J.M. Physicochemical Changes during Manufacture and Final Sensory Characteristics of Dry-Cured Celta Ham. Effect of Muscle Type. Food Control 2014, 43, 263–269. [Google Scholar] [CrossRef]
  45. Ruiz-Ramírez, J.; Arnau, J.; Serra, X.; Gou, P. Effect of pH24, NaCl Content and Proteolysis Index on the Relationship between Water Content and Texture Parameters in Biceps Femoris and Semimembranosus Muscles in Dry-Cured Ham. Meat Sci. 2006, 72, 185–194. [Google Scholar] [CrossRef]
  46. Škrlep, M.; Čandek-Potokar, M.; Santé-Lhoutellier, V.; Gou, P. Dry-Cured Ham Kraški Pršut Seasoning Losses as Affected by PRKAG3 and CAST Polymorphisms. Ital. J. Anim. Sci. 2011, 10, e6. [Google Scholar] [CrossRef]
  47. Harkouss, R.; Astruc, T.; Lebert, A.; Gatellier, P.; Loison, O.; Safa, H.; Portanguen, S.; Parafita, E.; Mirade, P.-S. Quantitative Study of the Relationships among Proteolysis, Lipid Oxidation, Structure and Texture throughout the Dry-Cured Ham Process. Food Chem. 2015, 166, 522–530. [Google Scholar] [CrossRef]
  48. Bec, K.B.; Grabska, J.; Pfeifer, F.; Siesler, H.W.; Huck, C.W. Rapid On-Site Analysis of Soil Microplastics Using Miniaturized NIR Spectrometers: Key Aspect of Instrumental Variation. J. Hazard. Mater. 2024, 480, 135967. [Google Scholar] [CrossRef] [PubMed]
  49. Hernández-Jiménez, M.; Revilla, I.; Vivar-Quintana, A.M.; Grabska, J.; Beć, K.B.; Huck, C.W. Comparison of Miniaturized and Benchtop NIR Spectrophotometers for Quantifying the Fatty Acid Profile of Iberian Ham. Appl. Sci. 2024, 14, 10680. [Google Scholar] [CrossRef]
  50. Dixit, Y.; Casado-Gavalda, M.P.; Cama-Moncunill, R.; Cama-Moncunill, X.; Cullen, P.J.; Sullivan, C. Prediction of Beef Fat Content Simultaneously under Static and Motion Conditions Using near Infrared Spectroscopy. J. Near Infrared Spectrosc. JNIRS 2016, 24, 353–361. [Google Scholar] [CrossRef]
  51. Prieto, N.; López-Campos, Ó.; Aalhus, J.L.; Dugan, M.E.R.; Juárez, M.; Uttaro, B. Use of near Infrared Spectroscopy for Estimating Meat Chemical Composition, Quality Traits and Fatty Acid Content from Cattle Fed Sunflower or Flaxseed. Meat Sci. 2014, 98, 279–288. [Google Scholar] [CrossRef]
  52. Sierra, V.; Aldai, N.; Castro, P.; Osoro, K.; Coto-Montes, A.; Oliván, M. Prediction of the Fatty Acid Composition of Beef by near Infrared Transmittance Spectroscopy. Meat Sci. 2008, 78, 248–255. [Google Scholar] [CrossRef]
  53. Morsy, N.; Sun, D.-W. Robust Linear and Non-Linear Models of NIR Spectroscopy for Detection and Quantification of Adulterants in Fresh and Frozen-Thawed Minced Beef. Meat Sci. 2013, 93, 292–302. [Google Scholar] [CrossRef]
  54. Grabska, J.; Beć, K.B.; Ishigaki, M.; Huck, C.W.; Ozaki, Y. NIR Spectra Simulations by Anharmonic DFT-Saturated and Unsaturated Long-Chain Fatty Acids. J. Phys. Chem. B 2018, 122, 6931–6944. [Google Scholar] [CrossRef] [PubMed]
  55. Cozzolino, D.; Murray, I. Identification of Animal Meat Muscles by Visible and near Infrared Reflectance Spectroscopy. LWT Food Sci. Technol. 2004, 37, 447–452. [Google Scholar] [CrossRef]
  56. Prevolnik, M.; Andronikov, D.; Žlender, B.; Font-i-Furnols, M.; Novič, M.; Škorjanc, D.; Čandek-Potokar, M. Classification of Dry-Cured Hams According to the Maturation Time Using near Infrared Spectra and Artificial Neural Networks. Meat Sci. 2014, 96, 14–20. [Google Scholar] [CrossRef]
  57. Williams, P. Near-Infrared Technology: Getting the Best out of Light: A Short Course in the Practical Implementation of Near-Infrared Spectroscopy for the User; Sun Press: New York, NY, USA, 2004. [Google Scholar]
  58. Resano-Ezcaray, H.; Sanjuán-López, A.I.; Albisu-Aguado, L.M. Combining Stated and Revealed Preferences on Typical Food Products: The Case of Dry-Cured Ham in Spain. J. Agric. Econ. 2010, 61, 480–498. [Google Scholar] [CrossRef]
  59. Curto, B.; Moreno, V.; García-Esteban, J.A.; Blanco, F.J.; González, I.; Vivar, A.; Revilla, I. Accurate Prediction of Sensory Attributes of Cheese Using Near-Infrared Spectroscopy Based on Artificial Neural Network. Sensors 2020, 20, 3566. [Google Scholar] [CrossRef] [PubMed]
  60. Garrido-Varo, A.; Riccioli, C.; Fearn, T.; De-Pedro Sanz, E.; Pérez-Marín, D.C. Multivariate Predictive Models for the Prediction of Fatty Acids in the EU High Added-Value “Acorn Iberian Pig Ham” Using a Miniature near-Infrared Spectroscopy Instrument. In Sensing for Agriculture and Food Quality and Safety XI; SPIE: Bellingham, WA, USA, 2019; Volume 11016. [Google Scholar] [CrossRef]
  61. González-Martín, M.I.; Escuredo, O.; Hernández-Jiménez, M.; Revilla, I.; Vivar-Quintana, A.M.A.; Martínez-Martín, I.; Hernández-Ramos, P. Prediction of Stable Isotopes and Fatty Acids in Subcutaneous Fat of Iberian Pigs by Means of NIR: A Comparison between Benchtop and Portable Systems. Talanta 2021, 224, 121817. [Google Scholar] [CrossRef] [PubMed]
Foods 15 00436 g001
Figure 1. Mean spectra for 100% Iberian and 50% Iberian hams recorded using the BüchiNIRFlex N-500 device in the three areas studied (lean meat, fat, and whole slices).
Figure 1. Mean spectra for 100% Iberian and 50% Iberian hams recorded using the BüchiNIRFlex N-500 device in the three areas studied (lean meat, fat, and whole slices).
Foods 15 00436 g001
Table 1. Mean value and statistical descriptors of each sensory attribute for the two genetic types.
Table 1. Mean value and statistical descriptors of each sensory attribute for the two genetic types.
100% IberianIberianSEMMinMaxCVp-Value
AppearanceVeined4.966.110.102.757.500.140.000
Fat color7.046.860.545.567.710.040.130
Color homogeneity5.595.980.644.676.880.060.002
Color intensity7.336.960.065.898.310.040.002
Exudate5.035.080.074.136.130.070.704
White dots2.873.400.102.005.250.170.005
FlavorOdor intensity6.676.750.046.007.330.030.308
Cured aroma6.756.780.035.947.280.030.740
Pig aroma 1.291.270.151.001.630.060.490
Rancid aroma1.521.620.031.222.170.080.050
Atypical aroma1.371.400.021.132.000.080.401
Flavor intensity6.776.830.046.177.380.030.417
Fat flavor intensity4.835.080.073.886.440.070.058
Cured flavor 6.676.660.045.887.250.030.903
Saltiness4.294.230.053.505.250.060.587
Sweetness2.482.480.031.883.000.070.909
Sourness1.401.420.031.112.130.090.819
Rancidity 1.611.660.031.112.250.100.516
Aftertaste5.956.060.055.256.750.040.274
Atypical flavor1.461.450.041.002.130.130.882
TextureHardness4.544.030.073.286.250.090.001
Juiciness5.726.030.064.756.710.050.009
Fatness4.614.920.063.576.140.070.019
Fibrousness3.503.170.072.435.000.100.020
Chewiness3.243.100.052.434.380.080.204
Gumminess2.372.440.041.753.130.090.430
Heterogeneity3.803.410.072.434.750.100.005
Chewing Residue2.732.720.042.223.500.080.832
SEM: standard error mean. Min: minimum value. Max: maximum value. CV: coefficient of variation.
Table 2. (a). Correlation coefficients (R2) and root mean square error (RMSE) values for the best ANN obtained from the NIR spectra of lean meat: Appearance parameters; (b). Correlation coefficients (R2) and root mean square error (RMSE) values for the best ANN obtained from the NIR spectra of lean meat: Flavor attributes; (c). Correlation coefficients (R2) and root mean square error (RMSE) values for the best ANN obtained from the NIR spectra of lean meat: Texture attributes.
Table 2. (a). Correlation coefficients (R2) and root mean square error (RMSE) values for the best ANN obtained from the NIR spectra of lean meat: Appearance parameters; (b). Correlation coefficients (R2) and root mean square error (RMSE) values for the best ANN obtained from the NIR spectra of lean meat: Flavor attributes; (c). Correlation coefficients (R2) and root mean square error (RMSE) values for the best ANN obtained from the NIR spectra of lean meat: Texture attributes.
(a)
VeinedFat ColorColor Homogeneity
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-50050.9560.6130.7300.81980.9970.7640.6620.855130.9990.5530.5050.756
0.2200.7440.8270.4690.0250.1740.3850.1650.0140.4180.5020.253
microPHAZIR90.9260.5200.6420.80280.9780.9140.5160.666111.0000.5350.7010.801
0.3460.7960.6860.5000.0580.5180.3520.2470.0080.5020.2730.221
MicroNIR 170080.8860.8870.7200.829130.9770.8080.5580.657150.9890.7240.5870.891
0.4090.3710.6430.4470.0620.3760.8200.3530.0520.2790.2970.164
Enterprise10.9800.5000.3930.81960.8140.3380.6500.647150.8620.7250.5230.634
0.2350.6140.8950.4640.1900.3510.3810.2560.1800.4360.6300.333
SCiO110.9680.7300.7590.78360.6340.6000.7710.548120.7020.4270.4990.577
0.1960.6451.3240.5940.2530.4040.2920.2860.2860.4000.3610.318
Color intensityExudateWhite dots
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-50060.8680.3350.3710.77540.5930.5610.5610.545100.7900.3100.4320.603
0.2010.2610.2790.2240.3200.2950.4560.3410.3350.7220.7750.497
microPHAZIR60.9080.5330.4590.612100.6740.4360.5170.458140.6780.1390.5850.558
0.1690.2640.6130.2950.3320.3850.5690.3850.4410.7790.5680.525
MicroNIR 170090.8230.5110.8570.707120.8110.2220.3390.525130.9990.5320.6530.699
0.1970.5100.1980.2680.2260.4980.6180.3610.0250.5501.0920.474
Enterprise30.7910.3500.4560.618150.8440.2130.4000.57110.9820.0510.4330.690
0.2320.4050.4050.2950.1990.6400.3970.3360.2950.9270.6020.494
SCiO50.6220.3700.4760.509140.7800.4880.3660.545140.9640.9160.5510.758
0.2840.3320.5340.3400.2380.4940.6660.3780.1260.5420.7470.373
(b)
Odor IntensityCured aromaPig aroma
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-500100.8490.0020.6880.559120.8670.6160.4840.65570.9330.4880.5500.655
0.1130.3640.2040.1870.0900.1990.3400.1700.0360.1300.0930.069
microPHAZIR80.8210.1000.5020.614100.9060.7540.5610.725130.9530.2070.6380.750
0.1180.2860.2520.1780.0840.1520.2980.1470.0310.1170.0670.058
MicroNIR 1700 0.9740.0180.5310.53240.9950.5020.3060.57280.8010.4130.4790.700
0.0450.3480.4360.2190.0280.2530.4440.1990.0560.0760.0890.065
Enterprise10.9030.4100.3930.573140.7080.7650.5650.657101.0000.2640.7450.778
0.0890.2260.3700.1840.1310.1550.2210.1520.0000.1320.0980.064
SCiO110.8110.2430.6130.583140.8920.5730.6210.728130.9700.4600.7360.859
0.1130.3330.2470.1860.0940.1630.2690.1450.0220.0780.0730.045
Rancid aromaAtypical aromaFlavor intensity
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-50070.9880.4620.8590.79510.8490.5910.6650.58210.9600.5940.4370.722
0.0190.2150.1060.0940.0560.1760.1840.1100.0860.2080.2920.156
microPHAZIR150.9170.4920.3370.550140.9980.0970.4030.610121.0000.1050.4110.727
0.0600.1750.2560.1300.0090.1210.2620.1120.0000.3180.2420.155
MicroNIR 170090.8040.1810.4440.394120.8760.5280.4090.673130.9840.4730.5370.751
0.0920.2120.3380.1730.0630.1400.1740.1010.0390.2360.2950.150
Enterprise50.7810.5070.4110.598140.8990.5440.3950.55470.9830.6910.6250.730
0.0990.1850.1520.1240.0600.1000.2480.1150.0350.2350.3150.155
SCiO50.5120.4750.4710.472110.5430.2250.4540.473120.8160.4250.4000.715
0.1500.1040.1450.1430.1270.1270.1100.1250.1350.2170.1860.158
Fat flavor intensityCured flavorSaltinessSweetness
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-500141.0000.1760.4690.65190.6870.1620.8030.62790.9950.4870.6260.74180.8250.2760.7830.676
0.0000.8030.5980.3880.1740.1780.2540.1890.0270.4010.3400.2050.1050.2780.1480.150
microPHAZIR70.8910.4680.6530.788150.8640.2930.1730.48270.9940.4240.3010.62611.0000.3630.7680.884
0.1720.3450.3780.2450.1060.4380.3470.2340.0280.2940.5010.2260.0060.1770.1490.090
MicroNIR 1700100.6640.6150.6170.614150.9990.4450.4260.64490.7730.7220.7780.73860.9290.8140.7350.833
0.3160.3550.3310.3240.0080.3440.4970.2340.1590.2130.2380.1820.0680.1020.2290.112
Enterprise50.9980.3850.3260.66320.9590.1950.4850.68870.7830.4980.7380.72110.8870.1970.5180.615
0.0290.5570.7390.3590.0600.3510.2470.1740.1780.2050.2730.1990.1140.2830.1480.156
SCiO60.7120.2880.8600.64450.7200.2120.7600.57210.8390.4150.6420.64040.8330.6830.6030.701
0.3010.4470.2440.3200.1710.3120.1810.2000.1460.3120.3940.2300.0960.2470.1610.140
SournessRancidityAftertasteAtypical flavor
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-50050.9440.2760.4860.70190.9750.3760.6910.82890.7540.2990.7250.35920.9620.4510.6860.727
0.0530.2210.1570.1140.0430.1950.1600.1040.1850.3650.5510.2990.0520.2930.2160.148
microPHAZIR100.9300.3630.3970.72530.7530.1730.6550.568130.6990.0190.5480.44340.7930.2640.7630.649
0.0570.1350.2050.1060.1600.2730.1870.1850.2190.4830.2740.2820.1520.1770.3220.191
MicroNIR 170050.9110.4180.5140.586120.8110.2060.5820.174110.7310.2970.8160.50270.8800.2110.6820.582
0.0610.1590.2910.1380.1050.4380.4150.2500.1930.2910.4670.2670.1130.2450.3500.191
Enterprise40.7530.0740.6440.52580.7320.0050.6900.35711.0000.5930.9020.863100.7420.2530.7280.360
0.1100.2230.1610.1410.1320.3280.2980.2040.0010.2640.2700.1460.1430.2520.4300.227
SCiO100.8960.1720.2720.495120.8690.2620.8170.38470.6980.2810.7290.566140.5370.2190.5890.464
0.0680.1680.3350.1560.0850.1990.4370.1990.2030.3870.2160.2410.1950.2640.1560.202
(c)
HardnessJuicinessFatnessFibrousness
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-50070.8360.8410.8780.843110.9800.6880.8060.885130.9340.2230.7770.578151.0000.5320.8400.862
0.2040.3020.2180.2240.0640.2300.2690.1470.1170.7720.4230.3550.0000.4970.2490.215
microPHAZIR60.7620.5670.8420.68620.8380.2650.7960.729110.7550.6070.8680.536111.0000.0260.7480.617
0.2570.6030.2640.3340.1660.3630.3080.2310.2710.3040.5840.3410.0000.7470.4590.339
MicroNIR 170021.0000.6990.8890.897 0.9180.7910.8340.85130.6870.4720.8040.65880.8040.5600.7460.714
0.0920.3040.4120.2130.1200.2740.2480.1750.2900.3940.2790.3060.2370.2810.5200.303
Enterprise130.9650.3660.9030.622140.7220.5360.6380.64190.5400.5540.7120.54980.9500.6310.8020.847
0.1090.6190.8260.4100.2700.3710.3440.2990.4080.5740.6190.4730.1260.2600.4090.215
SCiO40.8320.8000.9640.81350.6680.8570.8870.702140.8820.6380.6470.763100.9670.2330.8290.840
0.1950.3950.3470.2610.2670.1740.1180.2380.2170.4380.3060.2750.1060.3640.3730.220
ChewinessGumminessHeterogeneityChewing residue
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-500131.0000.2430.7790.64640.7890.6370.6940.28370.9130.6390.8560.76290.5220.3230.6130.504
0.0080.7080.3130.3000.1430.1620.6780.2950.1430.4880.3250.2570.2250.2610.1950.226
microPHAZIR100.8600.3120.7820.34840.8730.2370.6140.41811.0000.3220.7730.426120.6500.0590.6080.359
0.1610.5600.5000.3210.1210.3500.4620.2460.0000.5460.9650.4290.1950.2740.4030.250
MicroNIR 170020.7940.8170.7540.78440.8260.5610.7330.67050.5760.6170.6490.570140.5100.2900.6390.438
0.1710.2520.1700.1850.1310.2200.3910.2050.3040.4540.3730.3410.2370.2410.2560.241
Enterprise80.7350.3430.6760.672120.8030.5060.6140.67810.7440.3690.7140.661110.9980.3000.6670.335
0.1940.2560.3080.2250.1270.2840.2400.1790.2730.3970.3540.3070.0130.3400.7300.312
SCiO140.7040.6810.6410.66280.7730.1160.7220.55040.6700.6140.7640.643120.7410.3390.7900.358
0.1920.3650.2710.2380.1460.2870.3690.2190.2770.3830.3950.3150.1790.1900.5140.260
N: number of neurons in the hidden layer.
Table 3. (a). Correlation coefficients (R2) and root mean square error (RMSE) values for the best ANN obtained from the NIR spectra of fat: Appearance parameters; (b). Correlation coefficients (R2) and root mean square error (RMSE) values for the best ANN obtained from the NIR spectra of fat: Flavor attributes; (c). Correlation coefficients and root mean square error (RMSE) values for the best ANN architecture obtained from the NIR spectra of fat: Texture attributes.
Table 3. (a). Correlation coefficients (R2) and root mean square error (RMSE) values for the best ANN obtained from the NIR spectra of fat: Appearance parameters; (b). Correlation coefficients (R2) and root mean square error (RMSE) values for the best ANN obtained from the NIR spectra of fat: Flavor attributes; (c). Correlation coefficients and root mean square error (RMSE) values for the best ANN architecture obtained from the NIR spectra of fat: Texture attributes.
(a)
VeinedFat ColorColor Homogeneity
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-50070.8510.3950.6780.74560.7040.6350.6660.65450.8520.4740.4980.729
0.4270.8760.6790.5590.2140.4010.2210.2520.1990.3440.2580.377
microPHAZIR130.9990.5660.7400.83131.0000.7050.6410.84331.0000.5240.6780.758
0.0370.9320.7690.4690.0020.3570.2640.1720.0090.3200.2770.639
MicroNIR 170050.7380.3140.8000.660110.8540.6550.5400.71630.8510.6940.4660.706
0.5080.9080.8060.6340.1690.2040.4010.2250.1940.4300.2820.412
Enterprise50.8250.4890.7430.73850.9220.7000.6090.795140.9670.4750.4390.691
0.4510.8620.6430.5620.1150.2580.3690.1990.0970.5410.3270.614
SCiO140.8500.1010.5820.639120.8320.2820.4130.553150.8330.2880.7100.673
0.4431.0950.9440.6710.1760.3070.5340.2800.2030.4300.2880.417
Color intensityExudateWhite dots
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-50090.9920.3550.6100.815110.9700.2350.4930.565130.9170.7780.5480.822
0.0450.3820.4210.2230.0940.6850.5790.3560.2390.3330.5590.322
microPHAZIR70.9190.3390.4820.66280.8490.4730.3510.646151.0000.6260.4580.848
0.1370.5220.4190.2840.2150.4540.4430.3040.0000.5400.5500.299
MicroNIR 1700110.5880.6750.6240.601100.8430.2920.6520.65210.8830.6280.7810.703
0.3150.2800.2470.3010.2180.4580.5140.3230.3720.5221.0150.540
Enterprise20.7500.6660.5780.694130.9900.8380.3050.82660.9830.6340.7150.707
0.2350.3550.2920.2650.0470.2630.4810.2160.1040.6121.1360.508
SCiO80.9070.6170.2020.558100.9440.2420.7230.693130.8080.6510.4190.684
0.1430.3970.6670.3230.1170.6090.5820.3410.3280.3960.7500.428
(b)
Odor IntensityCured aromaPig aroma
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-50060.8570.3660.7290.59640.7810.7230.4720.67660.8690.1400.8110.734
0.1120.3610.2380.1920.1060.1840.2460.1480.0490.1000.1210.073
microPHAZIR100.8780.1670.5680.74380.9360.6830.1130.59910.9830.1950.3760.505
0.1130.2130.1740.1420.0640.2640.2960.1630.0220.1110.1910.088
MicroNIR 1700110.8000.4030.4490.609110.6490.2190.2710.460110.8770.5370.5470.730
0.1340.1580.3460.1850.1730.2210.2250.1890.0370.0900.1180.066
Enterprise130.7340.4730.3720.544110.9490.3440.2160.470100.9670.6280.4300.793
0.1760.1410.2760.1900.0600.3460.3690.2020.0240.0870.0950.054
SCiO50.5780.3570.4540.506140.4800.3890.1710.302140.7200.3100.3350.617
0.1950.2320.2070.2030.2010.2670.2330.2170.0660.0870.0880.073
Rancid aromaAtypical aromaFlavor intensity
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-500150.9900.0720.7080.33111.0000.4450.6520.863150.9740.5880.3440.602
0.0210.2400.4200.1880.0000.1350.0990.0650.0460.2320.5150.222
microPHAZIR40.6320.4200.7630.15690.8690.3970.4860.65430.8470.6500.6350.774
0.1130.2140.3300.1790.0570.1760.1480.1010.1120.2030.1870.142
MicroNIR 1700110.6270.5910.4030.568120.6380.6030.6820.60950.7000.3020.3490.442
0.1340.0970.1390.1300.1140.0840.0880.1060.1630.2550.4720.249
Enterprise80.9190.0280.4990.612130.9250.5890.6510.764140.8030.3090.5180.632
0.0500.2250.1860.1210.0520.1740.1400.0970.1320.3130.2000.181
SCiO80.6620.8030.5480.575150.9400.6870.3200.811140.7360.6520.3770.402
0.1220.0890.1820.1290.0440.0880.1440.0750.1600.1900.4700.237
Fat flavor intensityCured flavorSaltinessSweetness
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-50070.9520.3490.6260.74360.9510.7150.6470.82680.7790.5500.5810.680120.9500.3150.7190.761
0.1100.4680.4420.2660.0640.1900.2230.1260.1720.2620.2480.2000.0590.2510.1880.131
microPHAZIR90.7020.0470.6520.608120.7940.5460.7970.692150.9690.8130.2730.60090.9680.2280.5080.597
0.3380.4490.3760.3630.1260.2700.2250.1720.0600.2940.4950.2280.0480.4010.1730.174
MicroNIR 1700100.9800.6490.8400.774150.7290.1200.7800.637140.8970.4450.4970.610150.9470.7180.8260.858
0.0720.4400.5950.2930.1650.2130.2260.1830.1070.4290.3090.2240.0560.1700.1480.099
Enterprise140.9940.3930.3690.67970.7590.3830.4480.654120.7290.2440.6250.61580.6990.4990.8840.660
0.0410.4630.6570.3130.1470.2280.2570.1810.1990.3340.1440.2180.1480.2000.1880.164
SCiO110.7170.1710.4530.51390.7820.4600.3660.502120.5560.1530.6390.45950.5780.5410.2770.468
0.2800.4420.7360.4070.1290.2440.3960.2100.2360.3360.2650.2580.1760.1180.2680.186
SournessRancidityAftertasteAtypical flavor
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-50050.8410.7250.5000.68120.6580.6160.6640.571150.9420.3960.6780.79290.9990.4770.3290.511
0.0860.1060.2560.1290.1490.1690.1970.1600.0910.2860.2530.1660.0100.2640.4640.207
microPHAZIR40.8400.4600.3450.677110.7020.3170.6130.60370.9840.7000.8280.85420.9390.0950.7910.738
0.0860.1210.1920.1140.1550.1760.1690.1600.0480.1780.3370.1530.0780.2910.1690.146
MicroNIR 1700120.9580.5210.6070.84070.9260.1260.5660.67470.7850.3690.2420.59790.7420.4520.4160.624
0.0430.1380.1360.0830.0680.2990.1700.1450.1620.2900.3810.2300.1430.2420.2150.173
Enterprise111.0000.5980.4960.574130.9630.1280.4710.67080.8840.5190.5360.680110.9650.6020.3680.729
0.0000.1070.3460.1400.0510.3060.2090.1500.1290.3330.3100.2070.0490.2300.2900.149
SCiO50.7810.2770.1910.533120.5760.0170.4510.450150.6650.2190.3080.52810.6930.2720.3570.341
0.0980.1680.2270.1360.1640.2430.1880.1820.2090.3750.2680.2500.1510.2160.4290.225
(c)
HardnessJuicinessFatnessFibrousness
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-50090.9960.8440.7580.92680.9030.6820.5110.700130.8530.6540.7090.82150.8640.7510.5560.781
0.0380.2190.3360.1590.1270.3600.4460.2460.2140.2410.1910.2150.2040.3130.3610.252
microPHAZIR60.9630.6200.8980.913100.9880.6380.8110.797110.9940.0280.7570.742150.9960.6650.7340.794
0.1210.3190.1370.1680.0450.3360.4040.2070.0400.5220.4090.2590.0400.3880.6710.302
MicroNIR 1700140.8730.7630.8820.821150.8870.5130.5260.717110.8320.5140.6100.69570.8320.2520.7970.719
0.1900.4350.2410.2500.1440.3320.3910.2320.1890.4130.4160.2760.2160.4530.3270.282
Enterprise60.9680.7560.7270.851140.9570.8020.7070.85040.9980.5920.6350.737130.9980.5960.4950.804
0.1190.3460.4700.2470.0840.2270.3310.1710.0210.5320.5140.2870.0230.5100.3150.233
SCiO20.8640.8620.7600.82920.6530.2070.8020.57090.5900.2230.2000.459140.8570.8360.4300.671
0.1700.3610.3300.2370.2440.4530.2370.2840.3400.3970.4440.3660.2080.4000.6730.350
ChewinessGumminessHeterogeneityChewing residue
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-500100.9990.7160.6750.74860.8820.8170.6540.75750.8090.1700.7190.68230.9380.3260.3250.614
0.0090.3470.4460.2190.0920.1520.3250.1590.2360.4530.3540.2980.0810.2970.4430.217
microPHAZIR100.9920.2140.5330.63140.7450.4390.6030.673130.8310.5020.5130.675150.9970.2730.6350.715
0.0340.5000.4030.2500.1700.2560.2160.1920.2240.5510.3910.3220.0210.3060.3010.167
MicroNIR 1700110.7610.1590.6290.64830.5890.4250.4110.47840.8010.5210.7370.717100.8640.5110.3130.522
0.1970.3210.2770.2320.2040.2010.3860.2400.2190.4700.2670.2780.1050.3560.4720.245
Enterprise90.9630.4090.4670.70280.7760.3080.4590.470110.9490.2180.6960.66580.9660.3210.4140.715
0.0770.2660.4930.2260.1660.2780.4110.2370.1180.5650.6130.3380.0640.3240.2760.173
SCiO110.6910.5230.4430.60390.4930.3890.3490.449150.8170.1930.7030.649150.7200.4820.5740.590
0.2120.3280.3010.2470.2320.2280.2540.2350.2260.5960.2360.3120.1600.2590.2900.202
N: number of neurons in the hidden layer.
Table 4. (a). Correlation coefficients (R2) and root mean square error (RMSE) values for the best ANN obtained from the NIR spectra of the whole slice: Appearance parameters; (b). Correlation coefficients (R2) and root mean square error (RMSE) values for the best ANN obtained from the NIR spectra of the whole slice: Flavor attributes; (c). Correlation coefficients and root mean square error (RMSE) values for the best ANN architecture obtained from the NIR spectra of fat: Texture attributes.
Table 4. (a). Correlation coefficients (R2) and root mean square error (RMSE) values for the best ANN obtained from the NIR spectra of the whole slice: Appearance parameters; (b). Correlation coefficients (R2) and root mean square error (RMSE) values for the best ANN obtained from the NIR spectra of the whole slice: Flavor attributes; (c). Correlation coefficients and root mean square error (RMSE) values for the best ANN architecture obtained from the NIR spectra of fat: Texture attributes.
(a)
VeinedFat ColorColor Homogeneity
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-500120.9420.6030.8470.835130.9990.0540.9470.800151.0000.5430.7460.834
0.2670.8950.6360.4810.0120.4510.1980.1910.0030.4670.3470.225
Foss 5000 *50.6950.6620.7510.63240.7080.2840.7330.574110.9260.1930.6780.749
0.5180.8311.0240.6700.2350.4590.1790.2740.1450.5200.2910.261
microPHAZIR130.8820.4340.8210.79760.7940.4580.7940.72030.9450.6500.7090.801
0.4520.7370.5830.5250.2050.3400.1680.2260.1120.4020.3570.228
MicroNIR 170011.0000.3720.7080.82590.9100.3870.6970.80290.9090.3660.6320.667
0.0001.0360.5430.4530.1400.3340.1640.1860.1540.4320.5440.298
Enterprise110.9180.3940.8740.80820.7310.5870.7340.653101.0000.6150.6350.722
0.3160.8090.6070.4720.2300.4500.3360.2900.0040.4590.5930.290
SCiO30.7300.7160.8440.709130.8160.8300.7380.39220.4010.2230.5830.373
0.5570.5900.6880.5840.2020.1520.7470.3400.4120.3810.3070.393
Color intensityExudateWhite dots
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-500120.8410.6390.5960.73180.8190.7620.6360.51930.7060.1440.7900.628
0.2080.3130.4640.2780.2300.3100.7790.3770.4230.7110.2910.462
Foss 5000 *50.7500.5390.7880.63480.8180.0960.6490.63660.6250.2570.7350.423
0.2750.3610.3420.3000.2160.5440.3710.3130.5650.8700.4580.607
microPHAZIR70.8680.7840.5660.75880.9170.8460.5890.40710.9740.4510.7010.735
0.1700.4360.2660.2430.1500.1590.9730.4020.1600.8010.6820.429
MicroNIR 1700130.9090.6960.7340.676120.7900.3040.7800.650110.9570.1580.7400.711
0.1360.2840.5790.2740.2660.4560.3170.3090.1700.8400.7340.455
Enterprise10.9170.5020.7990.782150.7200.3640.6630.511140.6870.0170.7960.537
0.2380.3350.2820.2610.2910.7070.3530.3910.4540.7820.5090.525
SCiO30.7260.5800.6450.690120.6700.3600.6730.42990.8770.3660.7420.683
0.2630.2880.3220.2760.2930.5650.6340.4100.2830.8840.4400.450
(b)
Odor IntensityCured aromaPig aroma
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-500110.9970.1380.6020.752141.0000.5090.6890.90091.0000.0070.6090.715
0.0160.3220.1970.1470.0000.1720.1330.0840.0000.1420.0880.065
Foss 5000 *100.8080.0140.7710.653100.9050.3940.6830.763100.8110.0870.7730.681
0.1320.3050.1410.1710.0860.2260.1350.1250.0540.1170.0430.066
microPHAZIR101.0000.1700.6790.683120.9380.4090.8340.80960.7680.1250.6000.379
0.0020.4240.1810.1790.0820.1160.2030.1140.0630.1420.1340.092
MicroNIR 1700150.9180.3780.9060.689150.5520.5080.6770.412150.7720.0400.7180.642
0.0880.2870.2900.1740.1660.3120.2490.2080.0610.1160.0810.075
Enterprise50.6530.1900.7460.503120.6700.3650.7050.32680.9600.4910.6320.737
0.1730.3100.2250.2070.1510.2530.3690.2140.0230.1000.1430.070
SCiO40.7580.0050.6230.588140.7710.3440.5820.23220.4170.1370.6650.405
0.1380.2590.2620.1840.1130.3010.4570.2320.0810.1070.1240.093
Rancid aromaAtypical aromaFlavor intensity
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-50040.5750.2270.6250.505140.9610.3140.6430.71880.9020.3790.5400.307
0.1390.1070.1660.1390.0310.1990.1030.0900.0910.2910.6240.277
microPHAZIR100.7070.1850.7100.42160.9410.1410.5570.55790.8450.5420.7610.717
0.1000.2790.1450.1480.0460.2090.1970.1180.1190.1980.2750.165
MicroNIR 1700130.5620.0170.7690.24550.9950.0480.6080.61640.8100.4370.5360.645
0.1300.2790.1950.1710.0160.2490.1060.1060.1270.2380.3060.184
Enterprise140.7930.2710.6430.598130.9220.5290.6540.63520.6610.1850.5950.462
0.0980.2070.1150.1230.0560.1030.2250.1070.1890.2880.3100.228
SCiO140.9180.1990.7470.344100.7920.1840.6180.47620.8220.7780.7500.753
0.0570.2060.4050.1820.0860.2360.2290.1460.1610.1450.1560.158
Fat flavor intensityCured flavorSaltinessSweetness
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-500111.0000.2090.6380.50890.9920.2400.5240.710150.9810.6440.7450.816150.8550.6660.7070.787
0.0000.6140.7940.3890.0270.3070.3340.1770.0620.2750.2920.1640.0950.1760.1520.120
Foss 5000 *120.7580.1200.7710.611100.9730.0760.6650.427100.9720.6620.5610.787110.8230.5510.6080.381
0.2720.4970.3380.3260.0620.3870.5210.2570.0600.2720.3010.1650.1240.1870.4470.214
microPHAZIR20.9910.5810.6530.527140.6560.0120.6350.37391.0000.6690.6000.69940.9910.5790.7730.872
0.1420.3440.9640.4140.1900.4610.2930.2650.0000.2120.4580.1960.0250.1690.1750.096
MicroNIR 170040.9990.4770.5760.727140.7410.2630.5600.38650.6460.1580.6050.510151.0000.5000.6390.851
0.0520.3700.6750.3010.1820.2150.4300.2400.2030.3880.2720.2500.0020.2060.1420.097
Enterprise110.8970.0260.7310.430110.9670.5730.7130.82690.9400.8080.8250.86750.7470.3030.6690.587
0.1670.9740.5130.4480.0560.2720.1890.1370.0860.1660.2230.1300.1220.2880.1580.163
SCiO150.6010.6040.5890.569140.7940.0550.8550.21410.9820.7640.5870.877130.7390.5730.6180.670
0.3290.3400.4050.3430.1330.3830.5400.2800.1890.2310.2990.2160.1250.1640.2050.146
SournessRancidityAftertasteAtypical flavor
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-50030.9140.0950.5070.72680.8090.2280.5650.599120.7890.0490.5990.59880.8000.3980.6400.643
0.0610.1760.1540.1040.1220.2720.1270.1550.1610.3630.3160.2300.1270.2580.2290.171
100.7840.1770.6860.61490.7690.1910.6970.220130.8890.6300.6820.74370.9390.5610.7870.378
0.1040.1900.1360.1260.1170.3810.3400.2210.1110.3530.2440.1900.0730.2000.5710.242
microPHAZIR111.0000.1450.6990.781150.9090.5930.6070.42250.9260.3810.6270.74070.8490.4490.6290.697
0.0050.2110.1340.0970.0840.1290.4650.2000.0980.2340.3820.1920.1130.1990.2430.154
MicroNIR 170050.9990.4990.6020.82750.6550.4720.7720.30220.9950.5240.5720.73710.9380.5630.7210.605
0.0050.1470.1730.0880.1440.1940.4050.2120.0310.2920.4710.2160.1110.1590.4010.191
Enterprise131.0000.1000.6060.386130.9150.1370.6720.16940.8140.6900.7790.67660.8800.0100.7410.319
0.0030.2120.4430.1900.0780.3110.6200.2760.1520.2790.3650.2190.1050.3120.5060.247
SCiO80.3950.0140.6610.257140.7910.4890.6060.473130.7480.1870.6460.61370.6530.1730.6880.154
0.1620.2010.1790.1710.1140.2580.2830.1760.1750.3430.3180.2330.1480.2640.5800.276
(c)
HardnessJuicinessFatnessFibrousness
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-500121.0000.7250.8600.90550.9990.5490.9070.919150.9710.0040.7130.596120.9820.5610.7930.830
0.0000.2550.4020.1840.0110.2300.2350.1280.0930.8670.3250.3670.0860.3650.4340.231
Foss 5000 *80.9970.5950.6860.42620.6720.5780.6630.610110.9820.1370.7280.80530.8490.3450.6900.705
0.0370.5081.0660.4580.2300.4080.2890.2730.0700.4900.2850.2270.1990.3530.4840.286
microPHAZIR20.9350.5400.9140.780130.9800.5460.7850.82210.9610.6840.6070.56880.9180.6870.7840.823
0.1350.5650.2850.2700.0590.3390.3080.1840.1690.3430.7200.3400.1380.4180.2500.221
MicroNIR 1700100.9550.7730.8680.85320.8750.6310.8750.79420.7410.2480.6830.627150.7950.6810.8040.711
0.1160.2340.4910.2320.1740.3270.1980.2080.2340.5080.3280.3050.2440.5120.1320.289
Enterprise10.9860.5940.7330.85210.8560.1270.7180.65420.5580.6300.7550.23990.8060.4120.6270.662
0.1130.3870.3350.2200.2180.3900.4850.3020.3210.3670.8180.4390.2330.5500.3370.317
SCiO70.7910.6150.9070.76040.9690.4690.8130.848110.6650.2870.7060.52530.8340.5690.7840.756
0.2990.3350.4100.3240.0720.3490.2060.1680.2790.3920.5420.3490.1890.3680.3800.259
ChewinessGumminessHeterogeneityChewing residue
NR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2TotalNR2TrainR2ValidR2TestR2Total
RMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSERMSE
NIRFlex N-500150.9750.6830.7610.870130.9330.2570.6090.647150.9780.7100.8130.857110.9970.2000.8040.795
0.0610.2720.2150.1430.0890.2520.4360.2090.0820.1920.5000.2190.0180.2950.2860.160
Foss 5000 *140.8200.0910.6910.584100.9410.0430.6220.587141.0000.3490.7070.814140.9160.0040.6890.293
0.1810.3970.3510.2550.0730.4540.2540.2110.0000.3830.5590.2630.0940.4350.5310.277
microPHAZIR150.8480.0530.5320.54070.9040.3060.6980.45590.8390.2460.6870.71280.7140.0430.6990.304
0.1610.5180.3050.2690.1060.2620.5030.2370.2420.3650.4130.2940.1760.3910.4300.269
MicroNIR 1700120.9580.5290.7240.782120.7040.0490.8200.558140.8740.4060.7690.72220.8620.3700.8130.708
0.0740.3590.2870.1890.1830.3690.1570.2180.1990.4870.3250.2810.1310.2690.2540.180
Enterprise90.9270.5020.8220.341110.8860.6090.6370.337110.9110.1080.7320.710150.8160.1050.5740.522
0.1140.4220.7830.3580.1090.2530.6520.2860.1570.5700.2900.2800.1370.4090.2370.216
SCiO70.9070.3380.9010.71110.6620.0110.7380.605150.7100.5090.7590.649100.9960.5720.5580.617
0.1140.4010.3780.2340.2250.1940.1600.2120.3030.3640.2790.3090.0200.2840.5030.224
N: number of neurons in the hidden layer. * Foss 5000 spectra were recorded at three different points (5 cm × 5 cm, including fat and lean meat), and their average was the whole slice spectra. The whole slice spectra for the rest of the devices were the mean value of 6 spectra of the lean meat and 4 spectra of the fat.
Table 5. Number of parameters predicted depending on the R2Total for each recording site and device.
Table 5. Number of parameters predicted depending on the R2Total for each recording site and device.
Lean MeatFatWhole Slice
DeviceR2TotalAppearance (6)Flavor (14)Texture (8)Total
(28)		Appearance (6)Flavor (14)Texture (8)Total
(28)		Appearance (6)Flavor (14)Texture (8)Total
(28)
NIRFlex N-500>0.5613726613827613827
>0.651252251082359721
>0.7464144661648618
>0.82136222632510
Foss 5000>0.5 59620
>0.6 38415
>0.7 1438
>0.8 0022
microPHAZIR>0.5512421613827510621
>0.641031761082458415
>0.726194451354413
>0.8210321141225
MicroNIR 1700>0.5612725612725610825
>0.65771961062268721
>0.734411343103429
>0.8212502152204
Enterprise>0.561272561372667518
>0.65861961172448413
>0.71416436133429
>0.8111310341214
SCiO>0.56107236862036817
>0.6266144241032712
>0.72441001120146
>0.8012301120101
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Revilla, I.; Vivar-Quintana, A.M.; Martínez-Martín, I.; Hernández-Ramos, P.; Hernández-Jiménez, M.; Grabska, J.; Beć, K.B.; Huck, C.W. The Influence of the Site of Recording and Benchtop and Portable NIRS Equipment on Predicting the Sensory Properties of Iberian Ham. Foods 2026, 15, 436. https://doi.org/10.3390/foods15030436

AMA Style

Revilla I, Vivar-Quintana AM, Martínez-Martín I, Hernández-Ramos P, Hernández-Jiménez M, Grabska J, Beć KB, Huck CW. The Influence of the Site of Recording and Benchtop and Portable NIRS Equipment on Predicting the Sensory Properties of Iberian Ham. Foods. 2026; 15(3):436. https://doi.org/10.3390/foods15030436

Chicago/Turabian Style

Revilla, Isabel, Ana María Vivar-Quintana, Iván Martínez-Martín, Pedro Hernández-Ramos, Miriam Hernández-Jiménez, Justyna Grabska, Krzysztof B. Beć, and Christian W. Huck. 2026. "The Influence of the Site of Recording and Benchtop and Portable NIRS Equipment on Predicting the Sensory Properties of Iberian Ham" Foods 15, no. 3: 436. https://doi.org/10.3390/foods15030436

APA Style

Revilla, I., Vivar-Quintana, A. M., Martínez-Martín, I., Hernández-Ramos, P., Hernández-Jiménez, M., Grabska, J., Beć, K. B., & Huck, C. W. (2026). The Influence of the Site of Recording and Benchtop and Portable NIRS Equipment on Predicting the Sensory Properties of Iberian Ham. Foods, 15(3), 436. https://doi.org/10.3390/foods15030436

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Article metric data becomes available approximately 24 hours after publication online.
Back to TopTop