Next Article in Journal
Preliminary Design and Testing of a Resetting Combination Anchor, Antenna, and Tether Mechanism for a Spherical Autonomous Underwater Vehicle
Previous Article in Journal
Drag Reduction by Riblets on a Commercial UAV
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Applied Sciences
Volume 12
Issue 10
10.3390/app12105071
applsci-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Application of Machine Learning Using Color and Texture Analysis to Recognize Microwave Vacuum Puffed Pork Snacks

by
Tomasz Pawlak
1,
Agnieszka A. Pilarska
2,
Krzysztof Przybył
1,*,
Jerzy Stangierski
3,
Antoni Ryniecki
1,
Dorota Cais-Sokolińska
1,
Krzysztof Pilarski
4 and
Barbara Peplińska
5
1
Department of Dairy and Process Engineering, Poznań University of Life Sciences, ul. Wojska Polskiego 31, 60-624 Poznań, Poland
2
Department of Hydraulic and Sanitary Engineering, Poznań University of Life Sciences, ul. Piątkowska 94A, 60-649 Poznań, Poland
3
Department of Food Quality and Safety Management, Poznań University of Life Sciences, ul. Wojska Polskiego 31, 60-624 Poznań, Poland
4
Department of Biosystems Engineering, Poznań University of Life Sciences, ul. Wojska Polskiego 50, 60-627 Poznań, Poland
5
NanoBioMedical Centre, Adam Mickiewicz University, ul. Wszechnicy Piastowskiej 3, 61-614 Poznań, Poland
*
Author to whom correspondence should be addressed.
Appl. Sci. 2022, 12(10), 5071; https://doi.org/10.3390/app12105071
Submission received: 13 April 2022 / Revised: 6 May 2022 / Accepted: 16 May 2022 / Published: 18 May 2022
(This article belongs to the Section Food Science and Technology)
Browse Figures
Review Reports Versions Notes

Abstract

:
The objective of the study was to create artificial neural networks (ANN) capable of highly efficient recognition of modified and unmodified puffed pork snacks for the purposes of obtaining an optimal final product. The study involved meat snacks produced from unmodified and papain modified raw pork (Psoas major) by means of microwave-vacuum puffing (MVP) under specified conditions. The snacks were then analyzed using various instruments in order to determine their basic chemical composition, color and texture. As a result of the MVP process, the moisture-to-protein ratio (MPR) was reduced to 0.11. A darker color and reduction in hardness of approx. 25% was observed in the enzymatically modified products. Multi-layer perceptron networks (MLPN) were then developed using color and texture descriptor training sets (machine learning), which is undoubtedly an innovative solution in this area.

1. Introduction

Natural food products have recently been gaining in popularity; such products are usually processed using efficient and, to the extent possible, non-invasive methods to suit the needs of the consumers [1]. It is worth noting that changes in the metabolic conversion of food products can be induced by the application of high temperature [2]. Using an adequate method of preserving the quality of food products can prevent their spoiling and the development of bacteria and other microorganisms [3,4,5]. Drying food products involves the simultaneous movement of mass and heat (the drying agent releases heat to a moist material while simultaneously absorbing moisture from the material). Absorbing a sufficient amount of water requires adequate parameters of the drying agent (moisture, temperature, and steam). Values of these parameters depend on the physical and chemical parameters of food products. Quick, accurate and automatic determination of properties of food products is a practical requirement in everyday life. The drying procedure itself is complex, at times highly non-linear, and determines the quality of finished food products. Modern techniques, including acoustic signal [6], machine learning (ML) and deep learning (DL) [7] supported by computer vision [2,8,9], spectroscopy, including Fourier transform infrared spectroscopy (FTIR) [8], etc., are widely used to detect the properties of food.
Many examples today show that ML may provide innovative insight into existing research problems [3,10,11]. ML is defined as the approach to creating data patterns with learning algorithms for creating models, which in turn may be used to solve actual problems [12]. Using ML helps set new standards of speed, flexibility and optimization with respect to products, which translates into cost savings, error reduction and quality decision making.
The dominant method of representation learning born out of ML is artificial neural networks (ANNs) [13]. Neural models are a network of artificial neurons, interconnected so as to exchange various information, including using numerical data [8], images [2,14] or sound [6]. To use the simplest description, a neural model consists of input, hidden and output layers. Each layer is a set of neurons, where a neuron from each layer generates a signal constituting one of the parameters for each neuron in the next layer. The method used to present the signal when designing the network determines the type of the artificial neural network, e.g., feedforward, recurrent or convolutional [9].
Given the above, assisting in the processes of drying, mass transfer or assessing the quality of meat (such as pork, mutton, beef, etc.) becomes an important part of ensuring that food products contain a sufficient amount of protein [15]. No literature exists on the use of artificial intelligence (AI) to assess the quality of pork meat snacks.
Meat snacks are an important part of our diet and one of the fastest developing segments in the food market. The rapid development of meat snacks coinciding with an increased demand for food products tailored to the needs of consumers may enable the introduction of innovative meat snacks on the market, opening new possibilities before representatives of the meat industry, consumers and professionals. The meat snack industry wants to improve the quality and stability of traditional meat snacks or introduce new products with better nutritional values, functional qualities, more convenient packaging and improved sensory properties, such as taste and aroma, on the market. Matters related to the preservation of products are equally important [16].
Dehydration is one of the oldest methods of preserving meat [17]. Every kind of meat—beef, pork, game meat, poultry and fish—can be dried. Jerky made of beef is a widely sold meat snack, sold mostly in the USA. These are products available in large supermarkets, petrol stations, and in shops selling accessories connected, for example, with hunting and angling. The popularity of traditional meat snacks can reflect the changes in meat consumption. Drying has numerous applications in the food industry, mostly because of the variety of techniques available, long-life of dried products and economic reasons. Dried products are stable at room temperature and have a long shelf life because drying allows water activity to be decreased (aw) to values below 0.5. Thanks to this, the development of microorganisms can be stopped, as well as the reduction in the rate of chemical reactions and hydrolysis in food. Drying also reduces the mass and volume of products, which in turn reduces shipping costs and facilitates storing, in addition to achieving desirable product characteristics that cannot be achieved by other methods [6,18,19,20].
In order to maintain the quality of dried products, alternative methods that will result in a product having high stability and nutritional properties are sought. Microwave-vacuum drying of food has recently gained extensive popularity among drying methods [21,22,23]. Most reports in the literature on this method of drying refer to vegetables and fruit [24,25,26,27,28,29]. The possibility of using this method is also desired in drying raw meat of animal origin, such as fish [30] or pork [31]. The use of microwave-vacuum drying in food preservation may improve the sensory quality of products in terms of texture and color and reduce thermal degradation of nutrients and pro-health ingredients occurring as a result of using conventional drying methods [30,32]. The parameters of dried products obtained by the microwave-drying method are often close to the same product when freeze-dried. Such an opportunity is offered to multi-stage drying using microwave-vacuum drying and hot-air drying [30,33].
Texture and color are important quality parameters of foods and reflect the differentiation of raw meat (pork) and effect of numerous technological factors. Most frequently, these properties are assessed in meat snack foods using descriptors supplemented with various score scales [5,34]. However, there have been few attempts to assess the quality of such meat snacks only using instrumental methods [34]. With regard to meat products such as puffed pork snacks, this issue remains still unrecognized.
Therefore, the aim of this study was to analyze the mechanical properties and color of pork meat snacks (ready-to-eat dried products) manufactured under non-commercial conditions using laboratory equipment and use the results of the analysis as a training set in a machine learning process (artificial neural networks) to train AI to recognize and design pork meat snacks based on their color and mechanical (textural) properties to obtain the optimal final product.

2. Materials and Methods

2.1. Sample Preparation

The object of the study was raw meat, pork longitudinal muscle (Musculus psoas major) bought from the local slaughterhouse (Sokołów S.A., Robakowo, Poland). The mass of pork longitudinal muscle used in the study was 500.0 ± 0.4 g. A detailed description of the meat preprocessing method involving papain was given by Pawlak et al. in the first part of ref. [35].
On the basis of the process optimization of microwave-vacuum puffed pork snacks presented in the first part of the paper [35], the following technological conditions were determined: the initial for the MVP process moisture content of the pork slices was 0.5 kg/kg wb, microwave energy generated by the magnetron was 1.409 kJ/g and the vacuum absolute pressure 5 kPa. In the first stage of the process of obtaining the products, a single layer of slices was exposed to a flow of hot air under atmospheric pressure with the aim being to create a partially dried and sealed layer on the surface of the muscle tissue. Next, several slices of pork meat muscle, which were dried to a specific moisture content, were exposed to the MVP process in a microwave-vacuum drier. As the next step, after obtaining the puffed snacks, the stage of final hot-air drying occurred. Detailed information regarding the MVP process was presented and described in the first part of the ref. [35].
“Puffing” the meat requires the tenderization of the raw material using meat tenderizing salt produced by McCormick Polska; the salt contains papain, a proteolytic enzyme also referred to as a thiol protease (also known as sulfhydryl or cysteine protease).

2.2. Examination of Morphological Structure Using Cryo-SEM

The research samples were investigated with a JSM-7001FTTLS LV scanning electron microscope with an SEI detector with an accelerated voltage of 10 kV (JEOL Ltd., Akishima, Japan). A detailed description of the meat preprocessing method using papain was given by Pawlak et al. [35].

2.3. Determination of Basic Chemical Composition

To determine the basic chemical composition of microwave-dried pork muscle slices, the following methods were applied: dry matter using the oven method (PN-ISO 1442:2000) [36], total protein content according to the Kjeldahl procedure assuming 6.25 as the value of the conversion factor (PN-75/A-040l8/Az3:2002) [37], total fat content using the Soxhlet method (PN-ISO 1444:2000) [38], and ash content using PN-ISO 936:2000 [39]. Using the method recommended by the Food Safety and Inspection Service (FSIS) USA [40], the moisture-to-protein ratio (MPR) was calculated to determine if the proper level of meat drying had been achieved. A level of 0.75 or less is the accepted standard that identifies a product as jerky.

2.4. Texture Analysis

The dried muscle slices of pork for each treatment were analyzed by a puncture (compression) test using a TA-XT2 Texture Analyzer (Stable Micro Systems, Godalming, Surrey, UK). To achieve this, a type P/5 addition (steel rod of 0.005 m diameter—cut flat) was used. Twelve slices with a total mass of 10–11 g were placed flat inside a plastic cylinder with an inner diameter of 0.015 m. Such a sample was penetrated by a spindle to a depth of 40% of its initial height (0.03 m) at a movement speed of 0.001 m·s−1. Force–time deformation curves were obtained with a 25 kg load cell applied. The other operating conditions of the apparatus were as follows: pre-test speed 0.015 m·s−1; post-test speed 0.015 m·s−1 data acquisition rate 200 PPS and applied force 0.1 N. Based upon the research, the following parameters were determined: hardness of the sample (N), slope of the penetration curve (N/s) and work of compression (Nxs). The hardness of the sample is the maximum registered force during the penetration, the slope of the penetration curve was determined for the straight line connecting the point of the curve in which the force is equal to zero with the maximum force (hardness) point. Work of compression (Nxs) was determined for the total area under the extrusion curve. During the spindle penetration of the sample, numerous cracks are caused by squeezing, crushing and breaking the sample. They are registered as peaks until the spindle starts the return movement.

2.5. Color Analysis

The instrumental color of the dried meat was measured using a Minolta Chroma meter CR-200 (Minolta Camera Co., Osaka, Japan), which was pressed into close contact with the surface of the material. CIE-Lab L*, a*, and b* values were recorded. L*, a*, and b* values [7] indicate lightness, redness (+)/(−) greenness, and yellowness (+)/(−) blueness, respectively (Commission Internationale de l’Éclairage, CIE, 1976). The unit was standardized with a white-colored calibration tile. Ten readings per treatment were taken. The color difference (ΔE) is the distance between two points in a three-dimensional space and is calculated using the following formula [41]:
Δ E = [ Δ L * 2 + Δ a * 2 + Δ b * 2 ] 1 2

2.6. Structure of Artificial Neural Networks

The most popular neural network type is the multi-layer perceptron network (MLPN) [7,42,43]. The operation of single-layer feedforward networks is based on learning with the use of a backpropagation algorithm. This method enables the choice of weighting so as to reduce the prediction error to a minimum [7,8,44]. In the present paper, the feedforward networks were designed based on a multi-layer perceptron (MLP) which included the input, hidden and output layer. The number of hidden layers, including the number of neurons and type of activation function in the hidden and output layer, was determined using Statistica 13.3 in the network designer mode. Two variants of neural networks were designed as part of the study. Each variant was based on a multi-layer perceptron and included input variables that specified selected quality parameters of the test material. Similarities between the variants of designed networks were specified by the same output layer with 2 neurons: unmodified pork (without preprocessing) and modified pork (preprocessed using papain and NaCl). One hidden layer was applied for each network variant with different number of neurons (5 to 15). This time when designing the network, the activation function in neurons was set only for the hyperbolic tangent (Tanh) hidden layer and different activation functions were used in neurons for the output layer (hyperbolic tangent (Tanh), logistic (Log) and linear (Lin) activation). The input layer was determined by a set of independent (quantitative) variables based on the analysis of color and texture. Input variables included the following parameters: L*, a*, b* concerning the color and hardness of the sample (N), the slope of the penetration curve (N/s) and work of compression (Nxs). The network (weighting) parameters were determined using the Broyden–Fletcher–Goldfarb–Shanno (BFGS) algorithm [8,45]. The quality of obtained networks was evaluated based on the error in the learning, testing and validation sets, and then determining the root mean square error (RMSE) [2,46]. During the learning process of each neural network, data were randomly divided into three sets: learning, testing and validation. The learning data set used in the learning process included 70% of all cases (30 learning cases). The other two data sets were the testing set used to evaluate the network during its learning process and the validation set not used in the design of the network, i.e., 15% of the full data set.

2.7. Statistical Analysis

The statistical analysis was performed using the TIBCO Statistica data analysis software, version 13.3 (TIBCO Software Inc., Palo Alto, CA, USA). Results were presented as the average standard deviation (SD) from 15 repetitions of each analysis. Statistical analysis was performed using the PCA, HCA [47] and ANOVA [48] methods for each quality attribute of the test material and the Tukey test was performed. A critical significance level of 0.05 was used during the entire study. In the case of ANN for unmodified pork (without preprocessing) and modified pork (preprocessed using papain and NaCl), the efficiency was determined based on the determination coefficient (R2) and RMSE.

3. Results and Discussion

3.1. Chemical Composition

The processing of seasoned and preserved raw meat makes it possible to produce a product with attributes meeting the individual preferences of consumers. Such an effect can also be achieved in the case of preparing jerky-type meat snacks under home and commercial conditions. However, meat puffing is a technological process, which is more difficult to perform and satisfy customer expectations toward the final product [49,50].
Table 1 lists results presenting changes in basic contents in pork muscle slices after MVP. The slices subject to analysis contained more than 80% total protein and less than 4% fat. Statistically significant (α = 0.05) differences were noticed in the case of total protein and ash mass indicators. The ash content in meat products varied between 2.07% and 3.83%. It is safe to assume that a near double amount of ash content in the sample that underwent preprocessing with the use of papain and NaCl was due to the presence of 5.23% (mean value) of NaCl in the meat products.
Chemical composition, especially water content, in meat snack and jerky-type products is decisive both for their texture and shelf life [50]. As shown in other studies [51], meat snacks of this type produced at home are at risk of infection with coliform bacteria (Escherichia coli) of 0157:H7 [52,53,54] serotype or Salmonella [55]. According to the recommendations of the American Food Safety and Inspection Service (FSIS, 2014) [40], microbially safe jerky type products are those with an MPR (moisture-to-protein ratio) not exceeding 0.75, dried meat 2.04 and tropic cure pork 3.25 [40], whereas in raw beef, the MPR value is on average 4.5. However, it has been determined that this is not an adequate interpretation of the level of water activity (aw) present in the product. A microbially safe jerky product depends both on the parameters of the drying process of the seasoned meat and on the appropriate selection of components used in the spice blend and functional additives. It also requires proper water activity that is equal to or less than 0.85 (HACCP, 1997) [56].
According to the International Commission of Microbiological Specifications for Foods (ICMSF), a value of 0.85 is the water activity limit for Staphylococcus aureus growth under aerobic conditions and 0.90 under anaerobic conditions. Manufacturers are required to control the ratio of moisture to protein in the final product. The moisture/protein ratio (MPR) is controlled by varying the amount of added water based on the overall product formulation, and primarily by the drying procedure. In some products, the MPR can affect the final microbiological stability of the product, in others, it is important to ensure some elements of the overall product quality, such as the texture. The minimum requirement for all products produced in FSIS-inspected facilities is that they must meet the FSIS policy standards for MPR. However, these prescribed treatments have proved to be insufficiently lethal for some bacterial pathogens. Thus, most of the industry has volunteered to implement more rigorous treatment.
As a result of changes in the chemical composition connected with the dehydration process, a gradual decrease in the MPR value up to 0.11 was recorded. For comparison, previous investigations by Konieczny et al. (2007) [50] found that commercially available jerky-type meat snacks exhibited protein contents of over 50%, a low fat content (approximately 3.6%) and relatively high table salt content (approximately 6.0%). The MPR values of jerky examined in this study varied between 0.34 and 0.39. The pork snacks acquired in the process of microwave drying had a significantly lower moisture content together with a significantly higher protein content. Due to their reduced water content (approximately below 10%) and water activity, aw, below 0.80, they are classified as intermediate moisture foods (IMF). The IMF are characterized by their aw range from 0.65 to 0.90.

3.2. Visualization of Morphological Structure Using Cryo-SEM

Illustrative photos of samples puffed pork snacks were taken at 330 magnification (Figure 1). Figure 1a,b shows the structure of slices of examined samples obtained from the psoas muscle of a pig in the puffing process, in two variants: (a) unmodified pork, without preprocessing, and (b) modified pork, preprocessed using papain and NaCl.
Studies indicate that the significant differences in the quality of processing between red and white groups of muscles that can be associated with differences in the functional properties of myofibrillar protein are related to the type of the fiber [57]. Intercellular spaces appear to be full of typical eutectic artifacts resulting from the aggregation of substances dissolved in water following sublimation caused by the use of the cryo-SEM technique [58]. Bundles of muscle fibers in modified pork—preprocessed using papain and NaCl—appear to be denser than unmodified (non-preprocessed) pork. For comparison, Ozuna et al. (2013) [59] were also able to demonstrate the increased desalination of the sample related to the “salting out” phenomenon. Apart from NaCl, papain, a plant enzyme (classified as a class 3 hydrolase) commercially known as “meat tenderizing salt”, was used in the study and successfully imparted the meat snacks (modified pork—preprocessed using papain and NaCl) with the required crispness and specific taste, and improved its storage life. Plant enzymes primarily degrade proteins in the connective tissue—collagen and elastin—and have only a slight effect on muscle fibers [60].

3.3. Texture Analysis

The results of measuring the three distinguishing features of texture of dried pork snacks are shown in Table 2. The penetration method, used together with the effect of squeezing meat products, shows that the lowest strength, evidenced by the highest destruction force (i.e., cracks registered during penetration), was present in the sample that was preprocessed with the use of papain and NaCl. Moreover, a lower value on the curve slope also indicates the lower strength of preprocessed samples. The mean values acquired for modified samples were significantly different from the ones that were not exposed to preprocessing. Preprocessing resulted in lower mean values of the distinguishing mechanical features (except for the number of peaks), which indicates that the meat products acquired were of higher, more desirable tenderness (crispness).
Analysis of the distinguishing features of meat products’ texture using the instrumental method for the first variant showed a positive link with the conducted sensory analysis that yielded the highest tenderness value [35]. It can be assumed that not exposing the raw meat to preprocessing would yield unsatisfactory results for consumers, who would not enjoy these less tender meat snacks.
The instrumental analysis of the texture of enzymatically pretreated meat products was in line with the sensory evaluation of the texture as shown in the first part of the paper [35]. The pretreatment of meat muscle contributed to a reduction in hardness of over 25% with a simultaneous increase in the crispness of the final product.
The irregular peaks observed in the force curves of the roasted samples, for example, characteristic of a crispy product, reflected multiple fracture events, and could be a result of a non-homogeneous morphology and microstructural elements, such as water or oil phase distribution, causing the anisotropy of the structure. However, the particular importance of these elements in the perception of crispiness is still unknown [61].
Enzymatic methods are applied with an increasing frequency to control the technological properties of food raw meat. They are used to obtain hydrolysates with specific properties through enzymatic proteolysis of proteins or modification of their amino acid composition in the plastein reaction. One of the groups of proteases are cysteine proteases, with the best-known example being the plant enzyme papain [62]. It functions in a neutral environment and is used for meat tenderization. Figure 2 shows average curves acquired during a pork puffing mechanical properties study. The positive impact of enzymatic meat tenderization on lowering the mechanical strength of the samples analyzed can be observed.
Myofibrillar proteins (actin and myosin), intramuscular connective tissue, and perimysium played major roles in the mechanical resistance of muscle fiber [63]. Connective tissue and longitudinal muscle fiber contracted during the heat-induced denaturation of proteins, resulting in water leaking out of cells. This caused a change in hardness and in the force required for biting [64]. A similar effect can be seen in the case of studies in which microwave-vacuum drying of pork muscle slices was used.

3.4. Color Analysis

The color of meat is one of the fundamental distinguishing features of its quality and an important criterion for consumer desirability [59]. The decisive factor in creating the color of meat is the quantity, composition and transition of muscle tissue pigments. Other factors consist of the fat content, muscle tissue structure and active acidity of meat. The results for meat snack color are presented in Table 3. Measuring product color using the reflection method shows that there are significant differences between the parameters of the colors L*, a* and b* marked for the sample variants studied. Meat products that did not undergo preprocessing had a lighter color and lower values of a* and b*. The numerical value of the ΔE parameter gives an approximate difference of 6.64 between the color of unmodified (non-preprocessed) pork and modified pork (preprocessed with the use of papain and NaCl). The significant difference between the samples, which was evident and perceived as two different colors, may indicate that the value of the R2 coefficient of determination for both MLPN was above 0.90.
The results of sensory evaluation presented in the first part of the paper [35] led to the conclusion that a correlation exists between the diversity of color in the pork meat products obtained and the pretreatment of meat.
The instrumental analysis of the distinguishing features of unmodified meat samples proved a correlation with a conducted sensory color assessment. It can be assumed that not exposing the raw meat to preprocessing resulted in the meat products acquiring a lighter color, less satisfactory for consumers [57,58]. A darker color of modified pork snacks with a higher value of red color parameter a* was preferred by the evaluating panel. This may suggest that enzymatic modification of samples contributed to the preservation of the red color of pork meat that is desired by the consumers [58,60].

3.5. Machine Learning in Microwave Vacuum Puffed Pork Snacks

As part of machine learning of neural networks, a set of data with variables determining the quality properties of puffed pork snacks was imported into Statistica 13.3. Input and output data for the design of the network structure were determined. The input layer was used to send input variables (color or texture descriptors). Data were processed in at least one hidden layer and output layer. A total of 500 epochs were set in Statistica in the designer mode as part of the network learning process.
To determine the effectiveness of the recognition of puffed pork snacks, a hyperbolic tangent (Tanh) activation function was set for neurons in the hidden layer. The shape of the Tanh activation function resembles a sigmoid but differs from it in that it contains values in the range of [−1, 1]. This results in the fact that negative input values correspond to negative activation values, whereas positive input values correspond to positive activation values. On the other, output values with Tanh hover around zero, resulting in the fast learning of neural networks [65]. A reduction in weightings for neurons in hidden and output layers was used in the learning processed. As in similar studies, involving the examination of quality of fruit powders, the Broyden–Fletcher-Goldfarb–Shanno (BFGS) was used in machine learning. The BFGS algorithm is one of the most popular algorithms used to optimize data sets and is particularly useful in adapting machine learning algorithms [8,65]. Results for the networks obtained are shown in Table 4 and Table 5.
The first ANN obtained (MLP 3-12-1) uses Tanh transfer in the hidden layer (Figure 3) and linear function in the output layer of the network when identifying puffed pork snacks based on color (Lab). The other variant of the designed ANN (MLP 3-10-1), used to analyze texture, uses Tanh activation in the hidden layer (Figure 3) and output layer of the network. Taking into account the results for the learning, testing and validation set, RMSE values for both networks were obtained with MLP 3-12-1 at a level of 0.096 and with MLP 3-10-1 at a level of 0.038, respectively. The value of the R2 determination coefficient in the learning, testing and validation set for both networks exceeded 0.90 (with MLP 3-10-1 the value of R2 = 0.91 and with MLP 3-12-1 the value of R2 = 0.95). The high coefficient of determination R2, coupled with a low RMSE, indicates a very high efficiency of adaptation to study data (Table 5). Given the previous studies, an attempt was made to use color and texture parameters to evaluate puffed pork snacks. Future research will focus on the use of artificial neural networks to design a smart system or a real-time monitoring system of puffed pork snacks produced using the microwave vacuum process based on selected quality parameters (indirectly using the color and texture attributes).

3.6. Statistical Analysis

Principal component analysis (PCA) was used to sum up information on descriptors found in artificial neural networks. The purpose of PCA is to identify principal components, in this case, the data dimension was reduced to two principal components. Own vectors defined each principal component, PC1–PC6. The first component in the data set explained 60.95% of variance. The second component explained a much lower variance, i.e., 21.06%. Two component variables transferred over 80% of variability of the data set. This means that a good analysis in the two primary components of the selected data set is possible. Color and texture descriptors in Figure 4, supported with a correlation analysis (Table 6), indicate that brightness of color has a significant impact on the identification of puffed pork snacks, present in the data set of the obtained MLP 3-12-1 model. This translates to a negative correlation of L* (brightness) between parameters b* (yellow to blue) (R2 = −0.745) and a* (R2 = −0.501). It also appears that hardness (R2 = 0.636) and slope (R2 = 0.637) equally strongly correlate with the brightness of color L* for puffed pork snacks. However, a strong correlation of R2 = 0.854 exists between textural parameters (hardness and slope), which translates to adaptation during the learning process of the data set with textural parameters for the MLP 3-10-1 artificial neural net, obtaining a high coefficient of determination in the validation data set (0.999).
For comparison, an HCA analysis was performed using the Ward methods. Distances between concentrations were estimated, accounting for the variance analysis approach. As a result of the analysis in data sets concerning textural parameters and the Lab color space model, it was observed that significant similarities exist primarily between the brightness descriptor and textural parameters (Figure 5), which correlate with each other. Both HCA and PCA analysis can confirm the similarity between the discovery samples due to the brightness of the produced puffed pork snacks. We can conclude that the choice of product by the consumer will depend on the brightness of the produced puffed pork snacks [35,59].

4. Conclusions

The experiment has showed the potential for continued tests on evaluating the quality of puffed snacks. The neural networks designed enabled estimating the quality of recognition based on parameters of color and texture of puffed pork snacks. A high coefficient of determination R2 in machine learning was obtained for neural networks based on a multi-layer perceptron. The value of the R2 coefficient of determination for both MLPNs was above 0.90. The results indicated that it is possible to use ANN to test puffed pork snacks, which in future research could be used to control the acquisition of the final product.
On the other hand, the value of the moisture-to-protein ratio (MPR) determined in the study for pork muscle snacks was 0.11, which suggests the high microbiological safety of such a product, in accordance with FSIS.
The process of tenderizing raw meat with the use of protease ensured that the puffed snacks obtained crispness. Instrumental analysis showed that the suggested pretreatment substantially decreased the hardness by over 25% and led to an increase in the crispness of the final product. This was confirmed by a high number of cracks in the structure of snacks registered during the instrumental analysis of the texture.
Because of the color measurements conducted on two sample types, differentiations in the set parameters of color, L*, a*, and b* were stated. It can be presumed that no pretreatment of the meat would lead to snacks with the kind of too-bright colors less desired by consumers. This was also statistically confirmed using PCA and HCA analyses, which clearly showed a high correlation for the color parameter L in relation to other quality parameters in the data set.

Author Contributions

Conceptualization, T.P., K.P. (Krzysztof Przybył), J.S. and A.R.; methodology, T.P., K.P. (Krzysztof Przybył), J.S., B.P. and A.R.; software, K.P. (Krzysztof Przybył); validation, A.A.P., J.S., A.R.; formal analysis, A.A.P. and D.C.-S.; investigation, T.P., J.S. and B.P.; resources, T.P., A.A.P., K.P. (Krzysztof Przybył), J.S., K.P. (Krzysztof Pilarski) and B.P.; data curation, A.A.P., K.P. (Krzysztof Przybył), J.S. and A.R.; writing—original draft preparation, T.P., J.S., K.P. (Krzysztof Przybył) and A.R.; writing—review and editing A.A.P., K.P. (Krzysztof Przybył), D.C.-S. and K.P. (Krzysztof Pilarski); visualization, T.P. and K.P. (Krzysztof Przybył); supervision, A.A.P., J.S. and A.R.; project administration, T.P., J.S., A.R., D.C.-S.; funding acquisition, J.S., K.P. (Krzysztof Przybył), D.C.-S., K.P. (Krzysztof Pilarski). All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

Not applicable.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Lule, S.U.; Xia, W. Food Phenolics, Pros and Cons: A Review. Food Rev. Int. 2006, 21, 367–388. [Google Scholar] [CrossRef]
  2. Przybył, K.; Samborska, K.; Koszela, K.; Masewicz, Ł.; Pawlak, T. Artificial neural networks in the evaluation of the influence of the type and content of carrier on selected quality parameters of spray dried raspberry powders. Measurement 2021, 186, 110014. [Google Scholar] [CrossRef]
  3. Przybył, K.; Gawałek, J.; Koszela, K.; Wawrzyniak, J.; Gierz, L. Artificial neural networks and electron microscopy to evaluate the quality of fruit and vegetable spray-dried powders. Case study: Strawberry powder. Comput. Electron. Agric. 2018, 155, 314–323. [Google Scholar] [CrossRef]
  4. Przybył, K.; Gawałek, J.; Koszela, K.; Przybył, J.; Rudzińska, M.; Gierz, Ł.; Domian, E. Neural image analysis and electron microscopy to detect and describe selected quality factors of fruit and vegetable spray-dried powders—Case study: Chokeberry powder. Sensors 2019, 19, 4413. [Google Scholar] [CrossRef] [Green Version]
  5. Przybył, K.; Pilarska, A.; Duda, A.; Wojcieszak, D.; Frankowski, J.; Koszela, K.; Boniecki, P.; Kujawa, S.; Mueller, W.; Gierz, Ł. Health properties and evaluation of quality of dried strawberry fruit produced using the convective drying method with neural image analysis. In Proceedings of the Eleventh International Conference on Digital Image Processing (ICDIP 2019), Guangzhou, China, 10–13 May 2019; SPIE—The International Society for Optical Engineering: Bellingham, WA, USA, 2019; Volume 11179. [Google Scholar]
  6. Przybył, K.; Duda, A.; Koszela, K.; Stangierski, J.; Polarczyk, M.; Gierz, Ł. Classification of dried strawberry by the analysis of the acoustic sound with artificial neural networks. Sensors 2020, 20, 499. [Google Scholar] [CrossRef] [Green Version]
  7. Przybył, K.; Wawrzyniak, J.; Koszela, K.; Adamski, F.; Gawrysiak-Witulska, M. Application of Deep and Machine Learning Using Image Analysis to Detect Fungal Contamination of Rapeseed. Sensors 2020, 20, 7305. [Google Scholar] [CrossRef]
  8. Przybył, K.; Koszela, K.; Adamski, F.; Samborska, K.; Walkowiak, K.; Polarczyk, M. Deep and machine learning using SEM, FTIR, and texture analysis to detect polysaccharide in raspberry powders. Sensors 2021, 21, 5823. [Google Scholar] [CrossRef]
  9. Przybył, K.; Masewicz, Ł.; Koszela, K.; Duda, A.; Szychta, M.; Gierz, Ł. An MLP artificial neural network for detection of the degree of saccharification of Arabic gum used as a carrier agent of raspberry powders. In Proceedings of the Thirteenth International Conference on Digital Image Processing (ICDIP 2021), Singapore, 20–23 May 2021; Jiang, X., Fujita, H., Eds.; SPIE: Bellingham, WA, USA, 2021; Volume 11878, p. 93. [Google Scholar]
  10. Sejnowski, T.J.; Cypryański, P. Deep Learning: Głęboka Rewolucja: Kiedy Sztuczna Inteligencja Spotyka Się z Ludzką; Publishing House Poltext: Warsaw, Poland, 2019. [Google Scholar]
  11. Boniecki, P.; Idzior-Haufa, M.; Pilarska, A.A.; Pilarski, K.; Kolasa-Wiecek, A. Neural classification of compost maturity by means of the self-organising feature map artificial neural network and learning vector quantization algorithm. Int. J. Environ. Res. Public Health 2019, 16, 3294. [Google Scholar] [CrossRef] [Green Version]
  12. Sturm, T.; Gerlach, J.; Pumplun, L.; Mesbah, N.; Peters, F.; Tauchert, C.; Nan, N.; Buxmann, P. Coordinating human and machine learning for effective organizational learning. MIS Q. 2021, 45, 1581–1602. [Google Scholar] [CrossRef]
  13. Boniecki, P.; Raba, B.; Pilarska, A.A.; Sujak, A.; Zaborowicz, Z.; Pilarski, K.; Wojcieszak, W. Neural reduction of image data in order to determine the quality of malting barley. Sensors 2021, 21, 5696. [Google Scholar] [CrossRef]
  14. Pilarska, A.A.; Boniecki, P.; Idzior-Haufa, M.; Zaborowicz, M.; Pilarski, K.; Przybylak, A.; Piekarska-Boniecka, H. Image analysis methods in classifying selected malting barley varieties by neural modelling. Agriculture 2021, 11, 732. [Google Scholar] [CrossRef]
  15. Cheng, J.H.; Sun, D.W. Partial Least Squares Regression (PLSR) Applied to NIR and HSI Spectral Data Modeling to Predict Chemical Properties of Fish Muscle. Food Eng. Rev. 2016, 9, 36–49. [Google Scholar] [CrossRef]
  16. Kumar, P.; Verma, A.K.; Kumar, D.; Umaraw, P.; Mehta, N.; Malav, O.P. Meat Snacks: A Novel Technological Perspective (Chapter 11). In Innovations in Traditional Foods; Woodhead Publishing: Sawston, UK, 2019; pp. 293–321. [Google Scholar]
  17. Dhaliwal, H.K.; Gänzle, M.; Roopesh, M.S. Influence of drying conditions, food composition, and water activity on the thermal resistance of Salmonella enterica. Food Res. Int. 2021, 147, 110548. [Google Scholar] [CrossRef]
  18. Mujumdar, A.S. Handbook of Industrial Drying; CRC Press: Boca Raton, FL, USA, 2006. [Google Scholar] [CrossRef]
  19. Pawlak, T.; Ryniecki, A.; Stangierski, J. Puffing of pork: Effects of process conditions on expansion ratio. Przem. Spożywczy 2016, 1, 15–17. [Google Scholar] [CrossRef]
  20. Salehi, F.; Kashaninejad, M. Effect of drying methods on rheological and textural properties, and color changes of wild sage seed gum. J. Food Sci. Technol. 2015, 52, 7361–7368. [Google Scholar] [CrossRef]
  21. Walkowiak, K.; Masewicz, Ł.; Baranowska, H.M. LF NMR studies of microwave modified starch witch lysozyme. Sci. Nat. Technol. 2018, 12, 341–351. [Google Scholar]
  22. Walkowiak, K.; Przybył, K.; Baranowska, H.M.; Koszela, K.; Masewicz, Ł.; Piątek, M. The Process of Pasting and Gelling Modified Potato Starch with LF-NMR. Polymers 2022, 14, 184. [Google Scholar] [CrossRef]
  23. Vadivambal, R.; Jayas, D.S. Changes in quality of microwave-treated agricultural products—A review. Biosyst. Eng. 2007, 98, 1–16. [Google Scholar] [CrossRef]
  24. Cui, Z.W.; Xu, S.Y.; Sun, D.W.; Chen, W. Temperature changes during microwave-vacuum drying of sliced carrots. Dry. Technol. 2007, 23, 1057–1074. [Google Scholar] [CrossRef]
  25. Sutar, P.P.; Prasad, S. Modeling microwave vacuum drying kinetics and moisture diffusivity of carrot slices. Dry. Technol. 2007, 25, 1695–1702. [Google Scholar] [CrossRef]
  26. Clary, C.D.; Wang, S.; Petrucci, V.E. Fixed and incremental levels of microwave power application on drying grapes under vacuum. J. Food Sci. 2005, 70, E344–E349. [Google Scholar] [CrossRef]
  27. Böhm, V.; Kühnert, S.; Rohm, H.; Scholze, G. Improving the nutritional quality of microwave-vacuum dried strawberries: A preliminary study. Food Sci. Technol. Int. 2016, 12, 67–75. [Google Scholar] [CrossRef]
  28. Figiel, A. Drying kinetics and quality of beetroots dehydrated by combination of convective and vacuum-microwave methods. J. Food Eng. 2010, 98, 461–470. [Google Scholar] [CrossRef]
  29. Wojdyło, A.; Figiel, A.; Legua, P.; Lech, K.; Carbonell-Barrachina, Á.A.; Hernández, F. Chemical composition, antioxidant capacity, and sensory quality of dried jujube fruits as affected by cultivar and drying method. Food Chem. 2016, 207, 170–179. [Google Scholar] [CrossRef]
  30. Zhang, J.; Zhang, M.; Shan, L.; Fang, Z. Microwave-vacuum heating parameters for processing savory crisp bighead carp (Hypophthalmichthys nobilis) slices. J. Food Eng. 2007, 79, 885–891. [Google Scholar] [CrossRef]
  31. Laopoolkit, P.; Suwannaporn, P. Effect of pretreatments and vacuum drying on instant dried pork process optimization. Meat Sci. 2011, 88, 553–558. [Google Scholar] [CrossRef]
  32. Hu, Q.-G.; Zhang, M.; Mujumdar, A.S.; Xiao, G.-N.; Sun, J.-C. Drying of edamames by hot air and vacuum microwave combination. J. Food Eng. 2006, 77, 977–982. [Google Scholar] [CrossRef]
  33. Pawlak, T.; Ryniecki, A.; Siatkowski, I. Optimization of process parameters for microwave-vacuum puffing of black radish slices using the response surface method. Acta Sci. Pol. Technol. Aliment. 2013, 12, 253–262. [Google Scholar]
  34. Lee, S.W.; Kang, C.S. Effects of moisture content and drying temperature on the physicochemical properties of ostrich jerky. Food/Nahrung 2003, 47, 330–333. [Google Scholar] [CrossRef]
  35. Pawlak, T.; Gawałek, J.; Ryniecki, A.; Stangierski, J.; Siatkowski, I.; Peplińska, B.; Pospiech, E. Microwave vacuum drying and puffing of the meat tissue—Process analysis. Dry. Technol. 2019, 37, 156–163. [Google Scholar] [CrossRef]
  36. PN-ISO 1442:2000; Meat and Meat Products—Determination of Moisture Content (Polish Standard). PKN: Warsaw, Poland, 2000.
  37. PN-75/A-04018:1975/Az3:2002; Agricultural Food Products—Determination of Nitrogen Content by the Kjeldahl Method and Expressing as Protein (Polish Standard). PKN: Warsaw, Poland, 2002.
  38. PN-ISO 1444:2000; Meat and Meat Products—Determination of Free Fat Content (Polish Standard). PKN: Warsaw, Poland, 2000.
  39. PN-ISO 936:2000; Meat and Meat Products—Determination of Total Ash Content (Polish Standard). PKN: Warsaw, Poland, 2000.
  40. FSIS. Compliance Guideline for Meat and Poultry Jerky Produced by Small and Very Small Establishments; Compliance Guideline; USDA: Washington, DC, USA, 2014; pp. 1–54. [Google Scholar]
  41. Yan, W.Q.; Zhang, M.; Huang, L.L.; Tang, J.; Mujumdar, A.S.; Sun, J.C. Studies on different combined microwave drying of carrot pieces. Int. J. Food Sci. Technol. 2010, 45, 2141–2148. [Google Scholar] [CrossRef]
  42. Gierz, Ł.; Przybył, K.; Koszela, K.; Duda, A.; Ostrowicz, W. The use of image analysis to detect seed contamination—A case study of triticale. Sensors 2021, 21, 151. [Google Scholar] [CrossRef] [PubMed]
  43. Boniecki, P.; Zaborowicz, M.; Pilarska, A.; Piekarska-Boniecka, H. Identification process of selected graphic features apple tree pests by neural models type MLP, RBF and DNN. Agriculture 2020, 10, 218. [Google Scholar] [CrossRef]
  44. Keshtegar, B.; Piri, J.; Kisi, O. A nonlinear mathematical modeling of daily pan evaporation based on conjugate gradient method. Comput. Electron. Agric. 2016, 127, 120–130. [Google Scholar] [CrossRef]
  45. Biegalski, J.; Cais-Sokolińska, D.; Wawrzyniak, J. Effect of packaging and portioning on the dynamics of water–fat serum release from fresh pasta filata soft cheese. Foods 2022, 11, 296. [Google Scholar] [CrossRef]
  46. Hasar, H.; Hasar, U.C.; Kaya, Y.; Ozturk, H.; Izginli, M.; Ozbek, I.Y.; Oztas, T.; Canbolat, M.Y.; Ertugrul, M. Prediction of water-adulteration within honey by air-line de-embedding waveguide measurements. Measurement 2021, 179, 109469. [Google Scholar] [CrossRef]
  47. Wang, W.; Yu, W.; Zhao, L.; Chai, T. PCA and neural networks-based soft sensing strategy with application in sodium aluminate solution. J. Exp. Theor. Artif. Intell. 2011, 23, 127–136. [Google Scholar] [CrossRef]
  48. Do Nascimento, M.Z.; Martins, A.S.; Azevedo Tosta, T.A.; Neves, L.A. Lymphoma images analysis using morphological and non-morphological descriptors for classification. Comput. Methods Programs Biomed. 2018, 163, 65–77. [Google Scholar] [CrossRef] [Green Version]
  49. Nummer, B.A.; Harrison, J.A.; Harrison, M.A.; Kendall, P.; Sofos, J.N.; Andress, E.L. Effects of Preparation Methods on the Microbiological Safety of Home-Dried Meat Jerky. J. Food Prot. 2004, 67, 2337–2341. [Google Scholar] [CrossRef]
  50. Konieczny, P.; Stangierski, J.; Kijowski, J. Physical and chemical characteristics and acceptability of home style beef jerky. Meat Sci. 2007, 76, 253–257. [Google Scholar] [CrossRef]
  51. Calicioglu, M.; Sofos, J.N.; Samelis, J.; Kendall, P.A.; Smith, G.C. Effect of acid adaptation on inactivation of Salmonella during drying and storage of beef Jerky treated with marinades. Int. J. Food Microbiol. 2000, 89, 51–65. [Google Scholar] [CrossRef]
  52. Lim, J.Y.; Yoon, J.W.; Hovde, C.J. A Brief Overview of Escherichia coli O157:H7 and Its Plasmid O157. J. Microbiol. Biotechnol. 2010, 20, 5. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  53. Yoon, J.W.; Hovde, C.J. All blood, No stool: Enterohemorrhagic Escherichia coli O157:H7 infection. J. Vet. Sci. 2008, 9, 219–231. [Google Scholar] [CrossRef] [PubMed]
  54. Oporto, B.; Esteban, J.I.; Aduriz, G.; Juste, R.A.; Hurtado, A. Escherichia coli O157:H7 and non-O157 Shiga toxin-producing E. coli in healthy cattle, sheep and swine herds in Northern Spain. Zoonoses Public Health 2008, 55, 73–81. [Google Scholar] [CrossRef]
  55. Pohlman, S.R.; Kalchayanand, N.; Means, W.J.; Field, R.A.; Wolf, A.W. Destruction of Non-Pathogenic Escherichia coli in Beef Jerky Made With Home-Style Dehydrators; University of Wyoming, Annual Animal Science Research Report; Department of Animal Science, University of Wyoming: Laramie, WY, USA, 2003; pp. 100–105. [Google Scholar]
  56. HACCP. Generic Model-Dried Meats (Beef Jerky); ACIA Report; CFIA: Ottawa, ON, Canada, 1997; pp. 1–35. [Google Scholar]
  57. Xiong, Y.L. Myofibrillar protein from different muscle fiber types: Implications of biochemical and functional properties in meat processing. Crit. Rev. Food Sci. Nutr. 2009, 34, 293–320. [Google Scholar] [CrossRef]
  58. Nollet, L.M.L.; Toldra, F. Handbook of Muscle Foods Analysis; CRC Press: Boca Raton, FL, USA, 2008. [Google Scholar] [CrossRef]
  59. Ozuna, C.; Puig, A.; García-Pérez, J.V.; Mulet, A.; Cárcel, J.A. Influence of high intensity ultrasound application on mass transport, microstructure and textural properties of pork meat (Longissimus dorsi) brined at different NaCl concentrations. J. Food Eng. 2013, 119, 84–93. [Google Scholar] [CrossRef]
  60. Powell, T.H.; Hunt, M.C.; Dikeman, M.E. Enzymatic assay to determine collagen thermal denaturation and solubilization. Meat Sci. 2000, 54, 307–311. [Google Scholar] [CrossRef]
  61. Lillford, P.; van Vliet, T.; van de Velde, F. Discussion session on solid foods. Food Hydrocoll. 2006, 20, 432–437. [Google Scholar] [CrossRef]
  62. Ashie, I.N.A.; Sorensen, T.L.; Nielsen, P.M. Effects of papain and a microbial enzyme on meat proteins and beef tenderness. J. Food Sci. 2002, 67, 2138–2142. [Google Scholar] [CrossRef]
  63. Damez, J.L.; Clerjon, S. Meat quality assessment using biophysical methods related to meat structure. Meat Sci. 2008, 80, 132–149. [Google Scholar] [CrossRef]
  64. Tornberg, E. Effects of heat on meat proteins—Implications on structure and quality of meat products. Meat Sci. 2005, 70, 493–508. [Google Scholar] [CrossRef] [PubMed]
  65. Krohn, J.; Beyleveld, G.; Bassens, A. Deep Learning Illustrated: A Visual, Interactive Guide to Artificial Intelligence; Addison-Wesley Data and Analytics Series; Pearson Education Canada Inc.: North York, ON, Canada, 2020. [Google Scholar]
Applsci 12 05071 g001 550
Figure 1. Cryo-SEM images of the longitudinal structure of the puffed pork snacks MVP (a) unmodified pork—without preprocessing (b) modified pork—preprocessing with the use papain and NaCl (magnification: 330). Note: Cryo-SEM, cryo-scanning electron microscopy; MVP, microwave-vacuum puffed.
Figure 1. Cryo-SEM images of the longitudinal structure of the puffed pork snacks MVP (a) unmodified pork—without preprocessing (b) modified pork—preprocessing with the use papain and NaCl (magnification: 330). Note: Cryo-SEM, cryo-scanning electron microscopy; MVP, microwave-vacuum puffed.
Applsci 12 05071 g001
Applsci 12 05071 g002 550
Figure 2. Force vs. distance curves acquired during meat puffing texture analysis (average curves, n = 15) (sample description: see Table 1).
Figure 2. Force vs. distance curves acquired during meat puffing texture analysis (average curves, n = 15) (sample description: see Table 1).
Applsci 12 05071 g002
Applsci 12 05071 g003 550
Figure 3. Scheme of artificial neural network structure using Tanh activation function in hidden layer.
Figure 3. Scheme of artificial neural network structure using Tanh activation function in hidden layer.
Applsci 12 05071 g003
Applsci 12 05071 g004 550
Figure 4. PCA analysis of variables occurring in neural models.
Figure 4. PCA analysis of variables occurring in neural models.
Applsci 12 05071 g004
Applsci 12 05071 g005 550
Figure 5. Hierarchical dendrogram with grouping using the Ward method, illustrating the impact of simultaneously analyzed properties of color and texture present in learning data sets.
Figure 5. Hierarchical dendrogram with grouping using the Ward method, illustrating the impact of simultaneously analyzed properties of color and texture present in learning data sets.
Applsci 12 05071 g005
Table
Table 1. Chemical composition and MPR ratio in experimental microwave-vacuum dried pork.
Table 1. Chemical composition and MPR ratio in experimental microwave-vacuum dried pork.
Type of SampleWater (%)Protein (%)Fat (%)Ash (%)Moisture-to-Protein Ratio (MPR)
Unmodified pork9.03 ± 0.0784.97 a ± 0.203.91 ± 0.122.07 b ± 0.060.11
Modified pork9.07 ± 0.0983.11 b ± 0.183.94 ± 0.063.83 a ± 0.270.11
Means in the same column with different letters as superscripts indicate that there is a significant difference at α = 0.05 (n = 6). Type of sample: unmodified pork—without preprocessing; modified pork—preprocessing with the use papain and NaCl.
Table
Table 2. Mean values of distinguishing features of puffed pork snacks texture.
Table 2. Mean values of distinguishing features of puffed pork snacks texture.
Type of SampleHardness [N]Slope [N/s]Work of Compression [Nxs]
Unmodified pork133.53 a ± 12.434.60 a ± 0.511788.51 a ± 214.67
Modified pork108.40 b ± 9.643.80 b ± 0.411632.95 a ± 207.91
Means in the same column with different letters as superscripts indicate that there is a significant difference at α = 0.05 (n = 15) (sample description: see Table 1).
Table
Table 3. Mean values of distinguishing features of puffed pork snacks color.
Table 3. Mean values of distinguishing features of puffed pork snacks color.
Type of SampleColor Parameters
L*a*b*
Unmodified pork57.49 a ± 2.313.21 b ± 0.415.42 b ± 0.75
Modified pork51.67 b ± 2.074.35 a ± 0.858.41 a ± 1.07
Means in the same column with different letters as superscripts indicate that there is significant difference at α = 0.05 (n = 15) (Sample description: see Table 1).
Table
Table 4. The information on the structure and training, testing and validation error values of ANN.
Table 4. The information on the structure and training, testing and validation error values of ANN.
Structure of ANNActivation Function
Hidden/Output Layer		Function ErrorTraining of ErrorTesting of ErrorValidation
of Error
MLP 3-12-1Tanh/Linsum of squares0.00560.03380.1046
MLP 3-10-1Tanh/Tanhsum of squares0.00060.04930.0007
Table
Table 5. RMSE and R2 used to evaluate the performance of the ANN.
Table 5. RMSE and R2 used to evaluate the performance of the ANN.
ANNStatistical IndexTrainingTestingValidation
MLP 3-12-1 (by Lab analysis)RMSE0.01120.06760.2092
Coefficient of determination (R2)0.97870.88110.8666
Learning cases2866
MLP 3-10-1 (by Texture analysis)RMSE0.00120.09860.0140
Coefficient of determination (R2)0.98670.98190.9046
Learning cases2866
Table
Table 6. Correlation of color and texture variables in ANN.
Table 6. Correlation of color and texture variables in ANN.
L*a*b*Hardness [N]Work of Compression [Nxs]Slope [N/s]
L*1.00000
a*−0.500981.00000
b*−0.745490.708701.00000
Hardness0.63644−0.30142−0.659721.00000
Work of compression0.26711−0.01180−0.186560.584991.00000
Slope0.6372−0.39516−0.602720.854970.565711.000000
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

Pawlak, T.; Pilarska, A.A.; Przybył, K.; Stangierski, J.; Ryniecki, A.; Cais-Sokolińska, D.; Pilarski, K.; Peplińska, B. Application of Machine Learning Using Color and Texture Analysis to Recognize Microwave Vacuum Puffed Pork Snacks. Appl. Sci. 2022, 12, 5071. https://doi.org/10.3390/app12105071

AMA Style

Pawlak T, Pilarska AA, Przybył K, Stangierski J, Ryniecki A, Cais-Sokolińska D, Pilarski K, Peplińska B. Application of Machine Learning Using Color and Texture Analysis to Recognize Microwave Vacuum Puffed Pork Snacks. Applied Sciences. 2022; 12(10):5071. https://doi.org/10.3390/app12105071

Chicago/Turabian Style

Pawlak, Tomasz, Agnieszka A. Pilarska, Krzysztof Przybył, Jerzy Stangierski, Antoni Ryniecki, Dorota Cais-Sokolińska, Krzysztof Pilarski, and Barbara Peplińska. 2022. "Application of Machine Learning Using Color and Texture Analysis to Recognize Microwave Vacuum Puffed Pork Snacks" Applied Sciences 12, no. 10: 5071. https://doi.org/10.3390/app12105071

APA Style

Pawlak, T., Pilarska, A. A., Przybył, K., Stangierski, J., Ryniecki, A., Cais-Sokolińska, D., Pilarski, K., & Peplińska, B. (2022). Application of Machine Learning Using Color and Texture Analysis to Recognize Microwave Vacuum Puffed Pork Snacks. Applied Sciences, 12(10), 5071. https://doi.org/10.3390/app12105071

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

Article Metrics

Back to TopTop