Next Article in Journal
Development of Inclined Applicators for Flattened Beam Intraoperative Radiotherapy
Previous Article in Journal
Inter-Segmental Coordination of the Swimming Start among Paralympic Swimmers: A Comparative Study between S9, S10, and S12 Swimmers
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Fruit Powder Analysis Using Machine Learning Based on Color and FTIR-ATR Spectroscopy—Case Study: Blackcurrant Powders

1
Department of Dairy and Process Engineering, Faculty of Food Science and Nutrition, Poznań University of Life Sciences, Wojska Polskiego 31, 60-624 Poznan, Poland
2
Department of Physics and Biophysics, Faculty of Food Sciences and Nutrition, Poznań University of Life Sciences, Wojska Polskiego 28, 60-637 Poznan, Poland
3
Institute of Food Sciences, Warsaw University of Life Sciences WULS-SGGW, Nowoursynowska 159c, 02-787 Warsaw, Poland
4
Department of Biosystems Engineering, Poznań University of Life Sciences, Wojska Polskiego 50, 60-625 Poznan, Poland
*
Author to whom correspondence should be addressed.
Appl. Sci. 2023, 13(16), 9098; https://doi.org/10.3390/app13169098
Submission received: 20 July 2023 / Revised: 2 August 2023 / Accepted: 6 August 2023 / Published: 9 August 2023

Abstract

:
Fruits represent a valuable source of bioactivity, vitamins, minerals and antioxidants. They are often used in research due to their potential to extend sustainability and edibility. In this research, the currants were used to obtain currant powders by dehumidified air-assisted spray drying. In the research analysis of currant powders, advanced machine learning techniques were used in combination with Lab color space model analysis and Fourier transform infrared spectroscopy (FTIR). The aim of this project was to provide authentic information about the qualities of currant powders, taking into account their type and carrier content. In addition, the machine learning models were developed to support the recognition of individual blackcurrant powder samples based on Lab color. These results were compared using their physical properties and FTIR spectroscopy to determine the homogeneity of these powders; this will help reduce operating and energy costs while also increasing the production rate, and even the possibility of improving the available drying system.

1. Introduction

In recent years, the production of fruits has increased, as a consequence of their appealing taste and significant role in the human diet [1]. Fruits are also rich in nutrients that provide essential substances for the proper functioning of the body, i.e., protein, carbohydrates, vitamins, minerals, fiber, antioxidants and water. Fruits are products that provide the body with necessary nutrients and play an important role in the prevention of many diseases [2,3,4,5]. In view of the above, their preservation seems crucial. This research focused on blackcurrant fruits (Ribes nigrum L.).
According to statistics (www.fao.org, accessed on 5 November 2022), currant production in Poland continues to represent an upward trend over several decades [6]. This is due to, among other things, the area of berry crops. In Poland, currants account for 55% of land planted with bushes, raspberries 28% and chokeberry bushes 7%. The interest in the production as well as consumption of these berries indirectly depends on their content of bioactive components and their impact through acting as antioxidants in the diet [3]. As a consequence of the continuous increase in the demand for antioxidants in health-promoting foods, a good indicator of the quality status of these products during food processing is sought. The consumption of blackcurrants improves health in humans [7,8,9]. This is due to the fact that they mainly contain anthocyanins, flavonols and phenolic acids. It has been unequivocally confirmed in the literature that anthocyanins have antioxidant and anticarcinogenic effects as well as improving vision [10,11,12,13].
Fruit drying is one of the methods used to preserve fruit qualities [14,15,16]. It is a complex, highly non-linear thermal process the underlying mechanisms of which are still not perfectly understood, in relation to controlling (monitoring) it. The complexity of the drying process becomes more complicated when it is affected by a number of elements, including simultaneous conditions of transients, heat and mass change, phase transformations, physicochemical reactions, fruit variation reactions over time or structural changes in the dried product. The use of the drying process also carries a number of benefits. The purpose of the process is to protect the product from unwanted microflora; which, if unprotected, can potentially inhibit consumers’ health or to slow down the rate of the body’s chemical and physical reactions. Spray drying is becoming a very popular food preservation step in the food industry [17,18]. This process is often responsible for the final stage of the quality state of the resulting product. Fruits, as well as vegetables, are very often seasonal products. Drying makes it possible to preserve them while keeping their quality parameters: color, appearance, structure (particle size), sensorics (taste and smell), texture and nutritional value. Spray drying is a step that allows, moreover, the production of powder dyes with high storage stability [19,20,21].
In view of this, the search for effective modeling, pattern detection, optimization, control and the ability to conduct a series of simulations during drying are critical aspects for drying technology. Engineers or researchers, taking into account the critical parameters, search for the optimal drying method for the indicated raw material, i.e., food powders.
As a result of the progress of research dealing with classification and regression problems made with artificial neural networks (ANN), qualitative changes have been observed for various food products, which are dictated by consumer expectations. ANNs were created by defining, in some spatial relationships, layers—sets of neurons not connected to each other within a set, often all receiving the same arguments. ANNs are powerful computational intelligence. They are flexible systems, similar to non-linear regression techniques with a high degree of robustness for mapping the non-linear structure of arbitrarily complex and dynamic phenomena that loosely mimic the actions of neurons in brain biology, especially in areas where conventional modeling methods fail. In addition, ANNs do not require knowledge, assumptions, predefined mathematical relationships, explicit expressions or input–output relationships. ANNs are generally classified, based on learning mode and architecture, into supervised or unsupervised mode and feedback networks. In the former mode of training with a “teacher”, the learning data set contains many training patterns. This is in contrast to the second, for which the training data set consists only of input training, requiring no external assistance for learning. Even though a lot of mathematical modeling methods have been performed in the course of research related to spray drying to support process simulation, mass exchange and quality condition assessment, they are not sufficient to effectively evaluate the physical and chemical properties of finished powders. The development of the computational intelligence of the dataset allows the application of new solutions in the analysis of food quality and authenticity [22].
In this project, an attempt was made to compare the influence of selected qualitative parameters of currant powders on the basis of FTIR spectroscopy and machine learning. The utilitarian purpose is to apply a non-invasive artificial intelligence method to develop a target intelligent system for assessing the qualities of selected food products for operation in online systems. The research will support the recognition of color descriptors by considering the amount and type of carrier in currant powders using FTIR spectroscopy and machine learning methods.

2. Materials and Methods

The research materials were blackcurrant powders obtained through dehumidified air-assisted drying in in a laboratory dryer at an inlet air temperature of 80 °C and outlet air temperature of 50 °C. During the drying process, the humidity of the air was 0.5 g-cm−3. The feed flow rate was 0.3 mL·s−1. Blackcurrant powders were obtained according to the procedure of Przybył K. et al., 2021 [23]. Currant powders, which were characterized by different percentages (50% and 30% s.s.) and carrier—gum arabic (GA), maltodextrin (MD), inulin (IN), milk whey protein (W), fiber (F) and microcrystalline cellulose (C). This study concentrates observation on the extreme proportions of the carrier contained in the currant solution, i.e., the smallest (30% d.s. content named as: 30) and the largest (50% d.s. content named as: 50) proportion of them. This will identify the point at which the best physicochemical properties of currant powders are obtained. As a result, studying currant solutions with extreme proportions of the chosen carrier can contribute to more effective scaling of the drying process under industrial conditions while maintaining other (intermediate) proportions. As a result, 12 variants of blackcurrant powders were obtained, assigned as: MD50, MD70, GA50, GA70, W50, W70, C50, C70, F50, F70 and IN50, IN70.

2.1. Moisture Content

The blackcurrant powders’ moisture contents were measured according to a reference method using a Sartorius AG (Göttingen, Germany) MA 30 series electronic moisture analyzer (based on a precision weighing balance at drying of 5 g sample at 95 °C to constant mass). The moisture content measurement accuracy was 0.0005 kg/kg wb (wet basis), as declared by Sartorius AG. The measurement procedure was performed similarly to raspberry powders according to Przybył et al., 2021 [23].

2.2. Water Activity

Performance characterization of blackcurrant powders was carried out to evaluate the qualitative properties of the obtained blackcurrant powders. The measurement of water activity was carried out using the ADA-7 diffusion and water activity analyzer (COBRABID, Poznań) with a system of automatic recording of the time course of water evacuation for individual samples [24]. Samples were carried out in 5 replicates (total n = 90). Water activity was performed according to the procedure for raspberry powders according to Przybył et al., 2021 [23].

2.3. Color Analysis

The color of the blackcurrant powders was measured using an X-Rite SP-60 spectrophotometer. In order to measure the color with the colorimeter, blackcurrant powders were placed on plastic Petri dishes. Samples were made in 30 replicates for each variant of blackcurrant powder (total n = 360).

2.4. Fourier Transform Infrared Spectroscopy

Identification of hydrogen bonds in the tested samples, i.e., blackcurrant powder samples, was performed using infrared spectrophotometry. The study was conducted at Poznań University of Life Sciences using a spectrophotometer equipped with an ATR adapter containing a diamond device as an internal reflection element made by PerkinElmer Ltd. (Waltham, MA, USA). The spectra of the test samples of 0.005 ± 0.005 g were recorded as a function of absorbance in the wave number range of 4000–500 cm−1, with a resolution of 0.9 cm−1. Each spectrum is the result of the superimposition of five replicates. The research conducted for blackcurrant powders was similarly conducted for raspberry powders according to the procedure of Przybył et al., 2021 [23].

2.5. Machine Learning

In the samples tested using the color parameter, 30 learning sets were prepared that represented the smallest and largest proportion of the carrier in solution (30% and 50% of its content). The purpose of this stage of the research was to apply machine learning to develop an adequate neural network topology, which turned out to be a multilayer perceptron (MLP) network [22] (Figure 1). The structure of each neural network was characterized by 3 neurons in the input layer, 3 to 10 neurons in the hidden layer and 2 neurons in the output layer. Within machine learning, the BFGS algorithm was used. In this stage of the research, optimal MLP networks were obtained, which were able to effectively identify certain types of blackcurrant powders using Lab color (Figure 1).

2.6. Statistical Analysis

The research performed homogeneity of the blackcurrant powder sample using analysis of variance (ANOVA) and applied Tukey’s test to more accurately depict the value of the averages between the blackcurrant powder samples [25]. Statistics made it possible to assess whether there were significant differences between groups. This translated into performing precise homogeneity for these samples.

3. Results and Discussion

3.1. Moisture Content

In the first step, the study was conducted to measure the moisture content of blackcurrant powder samples, which differed in the proportion and type of carrier (Figure 2). The purpose of the analysis was to compare the carriers between the samples used in this research action. In relation to previous research associated with raspberry powders [23,26], it was confirmed that carriers with a lower proportion in the fruit solution resulted in the lowest moisture content in the fruit powders. It was observed that raspberry powders enriched with inulin (IN) achieved equally low moisture levels in the current study of currant powders. As a result, blackcurrant powders as well as raspberry powders [26] enriched with inulin (IN) have one of the lowest water contents compared to other blackcurrant powder samples containing other carriers. The current studies also noted that microcrystalline cellulose (C) and fiber (F) achieved low moisture content in the samples. With the perspective of preserving these products for a longer shelf life, designing fruit powders with an inulin carrier for blackcurrant as well as raspberry seems to be a more favorable endeavor. The current research confirms unequivocally that blackcurrant powders with gum arabic (GA) achieve the highest moisture content in the samples, regardless of the carrier content in the fruit solution. However, within the framework of food preservation and the method of storage for a longer period of time, the drying of fruit powders with gum arabic seems to be a less effective operation [27,28,29]. The reason for the high moisture content of gum arabic was determined by its properties. In the current study of blackcurrant powders with a maltodextrin carrier, comparable moisture content results were obtained to those for raspberry powders [23,26]. It has been confirmed that the difference in moisture content change for maltodextrin (MD) of between 50% and 30% with the participation of this carrier is the largest.

3.2. Water Activity

As part of the research activities, water activity was examined for currant powders. Water activity is a measure of the availability of water in the raw material. It is an important parameter that significantly affects microbiological stability and safety during prolonged storage of food products. The higher the level of water activity, the greater the development of microorganisms in the food product [30,31,32]. Low-temperature spray drying resulted in the lowest water activity in currant powder samples [19,33,34]. Based on previous studies involving inulin, the lowest water activity was demonstrated in raspberry powders [23]. In the current study involving currant powders, low water activity levels were also obtained for inulin (IN), milk whey protein (W), microcrystalline cellulose (C) and fiber (F) (Figure 3). Thus, in order to extend shelf life and maintain high product quality, in addition to inulin, milk whey proteins, microcrystalline cellulose and fiber appear to be alternative carriers. In the aspect of the water activity value criterion addressed for raspberry powders [23] was also maintained in the current study, where low-temperature spray drying was used. Virtually all currant powders obtained low water activity, below 0.2 except for maltodextrin (0.25). This low activity criterion significantly affects the microbiological quality and shelf-life extension of food products (Figure 3).

3.3. Color Analysis

The statistical analysis was carried out using the ANOVA method, taking into account Tukey’s test; the results are mapped in Table 1 and Table 2. The data in Table 1 and Table 2 refer to the Lab color space model of currant powders. The Lab color space model allows the determination of three basic color components such as brightness (L), red green (a) and yellow blue (b) [35,36,37]. In the context of the Tukey test, subclasses were obtained for currant powders with 30% and 50% carrier. This means that some samples of currant powders are similar to each other, while others differ [37]. In Table 1, it was observed that the color component responsible for brightness showed that currant powder with 50% milk whey protein (W50) statistically finds similarity with inulin (IN50) and fiber (F50). In the case of component (a), currant powders with 50% of the carrier show statistically the greatest similarity among themselves, including milk whey protein, inulin, microcrystalline cellulose and fiber. In the case of component (b), currant powders with 50% milk whey protein (W50) were found to have statistically significant similarity with maltodextrin (MD50). On the other hand, Table 2 shows the analysis of variance for different samples with the lowest proportion of carrier in the currant solution. It turns out that the L component statistically shows significant differences between the currant powder samples [37]. In the case of components (a) and (b), there was a statistically significant similarity between fiber (F70) and microcrystalline cellulose (C70) in currant powders with 30% carrier. A comparison, mentioned above, of the results related to moisture and water activity made it possible to identify how the proportion of the carrier, i.e., inulin, is statistically significantly affected [23]. In the case of color when sampling currant powders due to the components of the Lab color space model, inulin becomes one of those carriers that determine a significant statistical similarity.

3.4. Fourier Transform Infrared Spectroscopy

FTIR infrared spectroscopy is an important method for identifying functional groups and compounds present in plant materials. Figure 4 shows the infrared spectra for 12 currant powder samples with different carrier contents. The broad absorption band at 3290 cm−1 and the narrower band at a wavenumber of 2900 cm−1 are due to OH stretching vibrations forming hydrogen bonds and from asymmetric stretching of CH2 [38,39]. The highest intensity of these bands is observed for the F70 sample, which contains the least amount of fiber. In contrast, the lowest absorption intensity in this spectral region is shown for sample MD70, which has the lowest maltodextrin content. Another absorption band in the entire spectral region falls at 1732 and 1621 cm−1 and was attributed to the absorption of ester carbonyl (COOR) groups and carboxylate ion stretching band (COO) [40]. The literature reports that a wavenumber of 1708 cm−1 is attributed to bonds responsible for lipids and in the range of 1705–1575 cm−1 to amide I. From all samples tested, the highest intensity of absorption bands in these areas is observed for sample GA50, i.e., with 50% gum arabic content. All spectra showed the typical C-O-C stretching vibration band attributed to the carbohydrate region (1200–900 cm−1) [41]. The absorption maximum in this region also falls on the sample with a content of 50% gum arabic (GA50), while the lower content of this carrier (GA70) causes a decrease in absorption intensity. Corresponding to the crystalline regions, the fingerprint region occurring below 900 cm−1, which indicates the conformational changes occurring in the material under study [42] is observed for a wavenumber of 764 cm−1, regardless of the type and amount of carrier added.

3.5. Machine Learning

As a result of the machine learning process, 30 MLPNs were obtained, which had a high determination index for the test set of 0.99. The MLPNs generated for the learning set showed an equally high level of efficiency in recognizing blackcurrant powder samples. According to the established principle in machine learning, the obtained models meet the criterion of good fit to the network (rule of overfitting by under fitting) [43]. In the MLPN structure, it is worth noting that the range of the number of neurons in the hidden layer had an impact on finding the optimal MLP network. In the case of activation functions for neurons in the hidden layer, the tanh and exponential functions turned out to be the most efficient giving the best results from learning these networks. Tanh and exponential appear to be more flexible activation functions than the others, resulting in a better model determination rate on the test set [29,44]. In practice, based on machine learning principles, the choice of activation functions depends on the decision problem at hand as well as the network topology. The nonlinearity of these activation functions can also affect the more effective recognition performance of these data.
The machine learning experiments also observed a slightly higher success rate for identifying blackcurrant powders when the amount of carrier in solution was 50% (Table 3) than for samples with 30% of the carrier (Table 4).
In the case of the type of carrier, the process of identifying individual samples was most successful with the selection of milk whey protein (W70), gum arabic (GA70), maltodextrin (MD70), inulin (IN70) and microcrystalline cellulose (C70), taking into account the value of the smallest error made on the test collection with its 30% (Table 4) and 50% (Table 3) share in the solution. Worse results for the recognition of blackcurrant powder samples were obtained with fiber. The smallest error made on the test set among the acquired MLPNs in the machine learning process for 30% carrier content was achieved by MLP network 3-6-1 (model C70zW70), for which the error reached a value of 0.0025. In comparison, when recognizing the carrier with its 50% content in solution, the smallest error on the test set was achieved by models such as MLP 3-8-1 (MD50zGA50), MLP 3-9-1 (IN50zGA50) and MLP 3-9-1 (W50zGA50), for which the error reached 0.000005 ± 0.000001. It was observed that MLPNs with 50% content of gum arabic (GA50) reproduced the results for three models well. This means that these MLPNs in the group with 50% carrier content effectively predict the results for gum arabic. As a comparison with the data obtained by the FTIR spectroscopy method, it can be concluded that as the carrier content increases, the absorption intensity and the number of correct MLPN models increases, which generalizes well the relationship between the input data and the result.
MLP-type networks were developed with high coefficient of determination and low error committed on the test set for currant powder samples involving milk whey protein (W70), gum arabic (GA70), maltodextrin (MD70), inulin (IN70) and microcrystalline cellulose (C70). Slightly worse learning results were achieved with fiber (F) relative to the other blackcurrant powder samples. The smallest error made on the test set among the acquired MLPNs was shown at 30% carrier content for the 3-6-1 MLP network (model C70zW70), for which the error reached a value of 0.0025. It was also shown that at 50% carrier content in solution, the smallest error on the test set was achieved by models such as MLP 3-8-1 (MD50zGA50), MLP 3-9-1 (IN50zGA50) and MLP 3-9-1 (W50zGA50), for which the error reached 0.000005 ± 0.000001.

4. Conclusions

In the results of this research, statistical correlations between samples of blackcurrant powders have been shown. The measurement of moisture content showed differences between carriers in currant powders. The lower the moisture content of fruit powders, the higher the shelf life and safety of food products. It was shown that blackcurrant powders with inulin (IN), microcrystalline cellulose (C) and fiber (F) achieved the lowest moisture content in the samples. Powders involving gum arabic were shown to achieve the highest moisture content in the powder.
In the case of water activity, the relationships that were achieved when measuring the moisture content of blackcurrant powders were confirmed. Low water activity of blackcurrant powders with inulin (IN), milk whey protein (W), microcrystalline cellulose (C) and fiber (F) was confirmed.
FTIR spectroscopy achieved the highest intensity of bands at a wavenumber of 2900 cm−1 for blackcurrant powders with fiber (F), and the lowest with maltodextrin (MD). In the compartment corresponding to lipid binding, the highest intensity of absorption bands in these areas was shown for sample GA50.
The data presented above suggest the possibility of using non-invasive methods to assist in the food preservation process. The mechanism learned allows to maintain authenticity and homogeneity of blackcurrant powders obtained by spray drying. Thus, machine learning appears to be an effective tool for controlling the authenticity and quality of food products.

Author Contributions

Conceptualization, K.P.; methodology, K.P.; software, K.P., Ł.M. and K.W.; validation, K.P.; formal analysis, KP., K.K. and K.S.; investigation, K.P., K.S. and A.J.; resources, K.P.; data curation, K.P.; writing—original draft preparation, K.P.; writing—review and editing, K.P., K.S., K.W. and K.K.; visualization, K.P., Ł.M. and K.W. and supervision, K.K., K.S., J.B., A.J. and T.P. All authors have read and agreed to the published version of the manuscript.

Funding

This research received funding by grant NCN registration number of MINIATURA5: 2021/05/X/NZ9/01014.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The data presented in this study are openly available in the Repository for Open Data—RepOD at https://doi.org/10.18150/YD3OIV, accessed on 20 July 2023.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Wallace, T.C.; Bailey, R.L.; Blumberg, J.B.; Burton-Freeman, B.; Chen, C.-Y.O.; Crowe-White, K.M.; Drewnowski, A.; Hooshmand, S.; Johnson, E.; Lewis, R.; et al. Fruits, vegetables, and health: A comprehensive narrative, umbrella review of the science and recommendations for enhanced public policy to improve intake. Crit. Rev. Food Sci. Nutr. 2020, 60, 2174–2211. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  2. Cosme, F.; Pinto, T.; Aires, A.; Morais, M.C.; Bacelar, E.; Anjos, R.; Ferreira-Cardoso, J.; Oliveira, I.; Vilela, A.; Gonçalves, B. Red Fruits Composition and Their Health Benefits—A Review. Foods 2022, 11, 644. [Google Scholar] [CrossRef] [PubMed]
  3. Mazzoni, L.; Ariza Fernández, M.T.; Capocasa, F. Potential Health Benefits of Fruits and Vegetables. Appl. Sci. 2021, 11, 8951. [Google Scholar] [CrossRef]
  4. Van Duyn, M.A.S.; Pivonka, E. Overview of the Health Benefits of Fruit and Vegetable Consumption for the Dietetics Professional: Selected Literature. J. Am. Diet. Assoc. 2000, 100, 1511–1521. [Google Scholar] [CrossRef] [PubMed]
  5. Del Río-Celestino, M.; Font, R. The Health Benefits of Fruits and Vegetables. Foods 2020, 9, 369. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  6. Panfilova, O.; Tsoy, M.; Golyaeva, O. Currant growing technology and mechanized harvesting-review. E3S Web Conf. 2021, 254, 07002. [Google Scholar] [CrossRef]
  7. Fanelli, F.; Cozzi, G.; Raiola, A.; Dini, I.; Mulè, G.; Logrieco, A.F.; Ritieni, A. Raisins and Currants as Conventional Nutraceuticals in Italian Market: Natural Occurrence of Ochratoxin A. J. Food Sci. 2017, 82, 2306–2312. [Google Scholar] [CrossRef]
  8. Hunt, J.E.A.; Coelho, M.O.C.; Buxton, S.; Butcher, R.; Foran, D.; Rowland, D.; Gurton, W.; Macrae, H.; Jones, L.; Gapper, K.S.; et al. Consumption of New Zealand Blackcurrant Extract Improves Recovery from Exercise-Induced Muscle Damage in Non-Resistance Trained Men and Women: A Double-Blind Randomised Trial. Nutrients 2021, 13, 2875. [Google Scholar] [CrossRef]
  9. Sakaki, J.; Melough, M.; Gil Lee, S.; Kalinowski, J.; Koo, S.I.; Lee, S.-K.; Chun, O.K. Blackcurrant Supplementation Improves Trabecular Bone Mass in Young but Not Aged Mice. Nutrients 2018, 10, 1671. [Google Scholar] [CrossRef] [Green Version]
  10. Mudd, A.M.; Gu, T.; Munagala, R.; Jeyabalan, J.; Fraig, M.; Egilmez, N.K.; Gupta, R.C. Berry anthocyanidins inhibit intestinal polyps and colon tumors by modulation of Src, EGFR and the colon inflammatory environment. Oncoscience 2021, 8, 120–133. [Google Scholar] [CrossRef]
  11. Bonesi, M.; Leporini, M.; Tenuta, M.C.; Tundis, R. The Role of Anthocyanins in Drug Discovery: Recent Developments. Curr. Cancer Drug Targets 2019, 17, 286–298. [Google Scholar] [CrossRef]
  12. Hair, R.; Sakaki, J.R.; Chun, O.K. Anthocyanins, Microbiome and Health Benefits in Aging. Molecules 2021, 26, 537. [Google Scholar] [CrossRef]
  13. Lin, B.-W.; Gong, C.-C.; Song, H.-F.; Cui, Y.-Y. Effects of anthocyanins on the prevention and treatment of cancer. Br. J. Pharmacol. 2017, 174, 1226–1243. [Google Scholar] [CrossRef] [Green Version]
  14. Radojčin, M.; Pavkov, I.; Kovačević, D.B.; Putnik, P.; Wiktor, A.; Stamenković, Z.; Kešelj, K.; Gere, A. Effect of Selected Drying Methods and Emerging Drying Intensification Technologies on the Quality of Dried Fruit: A Review. Processes 2021, 9, 132. [Google Scholar] [CrossRef]
  15. Morais, R.M.S.C.; Morais, A.M.M.B.; Dammak, I.; Bonilla, J.; Sobral, P.J.A.; Laguerre, J.-C.; Afonso, M.J.; Ramalhosa, E.C.D. Functional Dehydrated Foods for Health Preservation. J. Food Qual. 2018, 2018, 1739636. [Google Scholar] [CrossRef] [Green Version]
  16. Petikirige, J.; Karim, A.; Millar, G. Effect of Drying Techniques on Quality and Sensory Properties of Tropical Fruits. Int. J. Food Sci. Technol. 2022, 57, 6963–6979. [Google Scholar] [CrossRef]
  17. Parikh, D.M. Advances in Spray Drying Technology: New Applications for a Proven Process. Am. Pharm. Rev. 2008, 11, 34. [Google Scholar]
  18. Dyvelkov, K.N.; Overgaard, J.S.; Pedersen, S.J. New Advances in Spray-Drying Processes. In Microencapsulation in the Food Industry: A Practical Implementation Guide; Academic Press: Cambridge, MA, USA, 2022. [Google Scholar]
  19. Jedlińska, A.; Samborska, K.; Wiktor, A.; Balik, M.; Derewiaka, D.; Matwijczuk, A.; Gondek, E. Spray drying of pure kiwiberry pulp in dehumidified air. Dry. Technol. 2021, 40, 1421–1435. [Google Scholar] [CrossRef]
  20. Jangam, S.V.; Mujumdar, A.S. Drying of Foods, Vegetables and Fruits; CRC: Singapore, 2010; ISBN 978-981-08-6759-1. [Google Scholar]
  21. Anandharamakrishnan, C.; Ishwarya, S.P. Spray Drying Techniques for Food Ingredient Encapsulation; John Wiley & Sons, Ltd.: Chichester, UK, 2015; ISBN 9781118863985. [Google Scholar]
  22. Przybył, K.; Koszela, K. Applications MLP and Other Methods in Artificial Intelligence of Fruit and Vegetable in Convective and Spray Drying. Appl. Sci. 2023, 13, 2965. [Google Scholar] [CrossRef]
  23. Przybył, K.; Samborska, K.; Akoszela, K.; Masewicz, L.; 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]
  24. Baranowska, H.M.; Masewicz, Ł.; Kowalczewski, P.Ł.; Lewandowicz, G.; Piątek, M.; Kubiak, P. Water properties in pâtés enriched with potato juice. Eur. Food Res. Technol. 2018, 244, 387–393. [Google Scholar] [CrossRef] [Green Version]
  25. Bland, J.M.; Altman, D.G. Tukey Multiple Comparison Test. Br. Med. J. J. Clin. Nurs. 1995, 310, 170. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  26. 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] [PubMed]
  27. De Barros Fernandes, R.V.; Borges, S.V.; Botrel, D.A. Gum arabic/starch/maltodextrin/inulin as wall materials on the microencapsulation of rosemary essential oil. Carbohydr. Polym. 2014, 101, 524–532. [Google Scholar] [CrossRef]
  28. Ray, A.K.; Bird, P.B.; Iacobucci, G.A.; Clark, B.C. Functionality of gum arabic. Fractionation, characterization and evaluation of gum fractions in citrus oil emulsions and model beverages. Food Hydrocoll. 1995, 9, 123–131. [Google Scholar] [CrossRef]
  29. 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]
  30. Juarez-Enriquez, E.; Olivas, G.; Ortega-Rivas, E.; Zamudio-Flores, P.; Perez-Vega, S.; Sepulveda, D. Water activity, not moisture content, explains the influence of water on powder flowability. LWT 2019, 100, 35–39. [Google Scholar] [CrossRef]
  31. Domian, E.; Poszytek, K. Wheat Flour Flowability as Affected by Water Activity, Storage Time and Consolidation. Int. Agrophys. 2005, 19, 119–124. [Google Scholar]
  32. Pérez-Reyes, M.E.; Tang, J.; Barbosa-Cánovas, G.V.; Zhu, M.-J. Influence of water activity and dry-heating time on egg white powders quality. LWT 2021, 140, 110717. [Google Scholar] [CrossRef]
  33. Jedlińska, A.; Samborska, K.; Wieczorek, A.; Wiktor, A.; Ostrowska-Ligęza, E.; Jamróz, W.; Skwarczyńska-Maj, K.; Kiełczewski, D.; Błażowski, Ł.; Tułodziecki, M.; et al. The application of dehumidified air in rapeseed and honeydew honey spray drying—Process performance and powders properties considerations. J. Food Eng. 2019, 245, 80–87. [Google Scholar] [CrossRef]
  34. Samborska, K.; Jedlińska, A.; Wiktor, A.; Derewiaka, D.; Wołosiak, R.; Matwijczuk, A.; Jamróz, W.; Skwarczyńska-Maj, K.; Kiełczewski, D.; Błażowski, Ł.; et al. The Effect of Low-Temperature Spray Drying with Dehumidified Air on Phenolic Compounds, Antioxidant Activity, and Aroma Compounds of Rapeseed Honey Powders. Food Bioprocess Technol. 2019, 12, 919–932. [Google Scholar] [CrossRef] [Green Version]
  35. Kulapichitr, F.; Borompichaichartkul, C.; Fang, M.; Suppavorasatit, I.; Cadwallader, K.R. Effect of post-harvest drying process on chlorogenic acids, antioxidant activities and CIE-Lab color of Thai Arabica green coffee beans. Food Chem. 2022, 366, 130504. [Google Scholar] [CrossRef]
  36. 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. [Google Scholar] [CrossRef]
  37. Przybył, K.; Wawrzyniak, J.; Samborska, K.; Gierz, Ł.; Koszela, K.; Szychta, M. Application of Artificial Neural Networks in Recognizing Carrier Based on the Color of Raspberry Powders Obtained in the Spray-Drying Process. In Proceedings of the Fourteenth International Conference on Digital Image Processing (ICDIP 2022), Wuhan, China, 20–23 May 2022; Xie, Y., Jiang, X., Tao, W., Zeng, D., Eds.; SPIE: Bellingham, WA, USA, 2022; Volume 12342, p. 172. [Google Scholar]
  38. Wongphan, P.; Harnkarnsujarit, N. Characterization of starch, agar and maltodextrin blends for controlled dissolution of edible films. Int. J. Biol. Macromol. 2020, 156, 80–93. [Google Scholar] [CrossRef]
  39. Yu, M.; Chen, Y.; Liu, Y.; Xu, Y.; Wang, B. Efficient polysaccharides from Crinum asiaticum L.’s structural characterization and anti-tumor effect. Saudi J. Biol. Sci. 2019, 26, 2085–2090. [Google Scholar] [CrossRef]
  40. Aly, A.A.; Ali, H.G.M.; Eliwa, N.E.R. Phytochemical screening, anthocyanins and antimicrobial activities in some berries fruits. J. Food Meas. Charact. 2019, 13, 911–920. [Google Scholar] [CrossRef]
  41. Ying, D.; Hlaing, M.M.; Lerisson, J.; Pitts, K.; Cheng, L.; Sanguansri, L.; Augustin, M.A. Physical properties and FTIR analysis of rice-oat flour and maize-oat flour based extruded food products containing olive pomace. Food Res. Int. 2017, 100, 665–673. [Google Scholar] [CrossRef]
  42. Melanie, H.; Susilowati, A.; Iskandar, Y.M.; Lotulung, P.D.; Andayani, D.G. Characterization of Inulin from Local Red Dahlia (Dahlia sp. L) Tubers by Infrared Spectroscopy. Procedia Chem. 2015, 16, 78–84. [Google Scholar] [CrossRef] [Green Version]
  43. Bashir, D.; Montañez, G.D.; Sehra, S.; Segura, P.S.; Lauw, J. An Information-Theoretic Perspective on Overfitting and Under-fitting. In Proceedings of the Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Canberra, ACT, Australia, 29–30 November 2020; Volume 12576. [Google Scholar]
  44. Dubey, S.R.; Singh, S.K.; Chaudhuri, B.B. Activation functions in deep learning: A comprehensive survey and benchmark. Neurocomputing 2022, 503, 92–108. [Google Scholar] [CrossRef]
Figure 1. MLPN generation procedure based on machine learning.
Figure 1. MLPN generation procedure based on machine learning.
Applsci 13 09098 g001
Figure 2. Moisture content of powders containing 50% and 70% of blackcurrant concentrate solids obtained with gum arabic (GA), inulin (IN), maltodextrin (MD), microcrystalline cellulose (C), fiber (F) and whey milk proteins (W) by dehumidified air-assisted spray drying.
Figure 2. Moisture content of powders containing 50% and 70% of blackcurrant concentrate solids obtained with gum arabic (GA), inulin (IN), maltodextrin (MD), microcrystalline cellulose (C), fiber (F) and whey milk proteins (W) by dehumidified air-assisted spray drying.
Applsci 13 09098 g002
Figure 3. Water activity of powders containing 50% and 70% of blackcurrant concentrate solids obtained with gum arabic (GA), inulin (IN), maltodextrin (MD), microcrystalline cellulose (C), fiber (F) and whey milk proteins (W) by dehumidified air-assisted spray drying.
Figure 3. Water activity of powders containing 50% and 70% of blackcurrant concentrate solids obtained with gum arabic (GA), inulin (IN), maltodextrin (MD), microcrystalline cellulose (C), fiber (F) and whey milk proteins (W) by dehumidified air-assisted spray drying.
Applsci 13 09098 g003
Figure 4. The infrared spectra wavenumber for 12 blackcurrant powder samples with different carrier contents.
Figure 4. The infrared spectra wavenumber for 12 blackcurrant powder samples with different carrier contents.
Applsci 13 09098 g004
Table 1. Lab color of powders containing 50% of blackcurrant concentrate solids obtained with gum arabic (GA), inulin (IN), maltodextrin (MD), microcrystalline cellulose (C), fiber (F) and whey milk proteins (W) by dehumidified air-assisted spray drying.
Table 1. Lab color of powders containing 50% of blackcurrant concentrate solids obtained with gum arabic (GA), inulin (IN), maltodextrin (MD), microcrystalline cellulose (C), fiber (F) and whey milk proteins (W) by dehumidified air-assisted spray drying.
Blackcurrant
Powder Type
Lab
W5029.00±0.49 a36.92±0.39 a15.12±0.17 a
MD5027.84±0.10 b36.02±0.08 c15.35±0.07 a
GA5024.58±0.11 c34.00±0.12 b16.34±0.10 b
IN5029.58±0.07 a37.35±0.08 a16.57±0.08 b
C5028.25±1.34 b37.21±1.11 a17.60±0.59 c
F5029.35±1.92 a37.39±1.57 a18.15±0.88 d
a–d: the differences between mean values with the same letter in columns were statistically insignificant (p < 0.05).
Table 2. Lab color of powders containing 70% of blackcurrant concentrate solids obtained with gum Arabic (GA), inulin (IN), maltodextrin (MD), microcrystalline cellulose (C), fiber (F) and whey milk proteins (W) by dehumidified air-assisted spray drying.
Table 2. Lab color of powders containing 70% of blackcurrant concentrate solids obtained with gum Arabic (GA), inulin (IN), maltodextrin (MD), microcrystalline cellulose (C), fiber (F) and whey milk proteins (W) by dehumidified air-assisted spray drying.
Blackcurrant
Powder Type
Lab
F7024.37±1.60 c34.11±1.40 a17.64±0.89 a
C7023.40±1.00 b34.06±0.94 a17.54±0.55 a
W7027.77±0.25 f37.73±0.26 e18.98±0.16 e
IN7022.18±0.11 a33.22±0.10 b16.87±0.08 b
GA7025.33±0.18 d35.53±0.25 c18.17±0.20 c
MD7026.04±0.14 e36.44±0.15 d18.61±0.12 d
a–f: the differences between mean values with the same letter in columns were statistically insignificant (p < 0.05).
Table 3. Results of process machine learning for 50% of carrier content.
Table 3. Results of process machine learning for 50% of carrier content.
Blackcurrant
Powder Type
Name
ANN
Val
Quality
Test
Quality
Train
Quality
Train
Error
Test ErrorVal
Error
Train
Algorithm
Activation
Hidden
Activation
Output
F50zC50MLP 3-9-10.99930.99930.99850.0015460.0023260.004332BFGS 45TanhLinear
F50zW50MLP 3-9-10.99980.99970.99970.0018530.0030870.002819BFGS 41TanhLinear
F50zIN50MLP 3-7-10.99950.99910.99850.0104300.0189070.029667BFGS 26ExponentialLinear
F50zGA50MLP 3-8-10.99980.99990.99960.0080650.0028170.019212BFGS 189ExponentialLinear
F50zMD50MLP 3-8-10.99960.99960.99850.0216150.0285020.080596BFGS 28ExponentialLinear
C50zW50MLP 3-5-10.99930.99930.99930.0015740.0015150.001455BFGS 28ExponentialLinear
C50zIN50MLP 3-6-10.99960.99930.99960.0039920.0066220.003384BFGS 63ExponentialLinear
C50zGA50MLP 3-7-10.99990.99990.99990.0003770.0004380.001632BFGS 99TanhLinear
C50zMD50MLP 3-7-10.99990.99990.99990.0000250.0000330.001131BFGS 158TanhLinear
W50zIN50MLP 3-8-10.99990.99990.99990.0000170.0000390.000115BFGS 199ExponentialLinear
W50zGA50MLP 3-9-10.99990.99990.99990.0000110.0000060.000039BFGS 63TanhLinear
IN50zGA50MLP 3-9-10.99990.99990.99990.0000050.0000050.000049BFGS 160ExponentialLinear
IN50zMD50MLP 3-6-10.99890.99990.99880.0098640.0013570.017689BFGS 49ExponentialLinear
GA50zMD50MLP 3-8-10.99990.99990.99990.0000100.0000040.000276BFGS 124ExponentialLinear
W50zMD50MLP 3-9-10.99990.99990.99990.0003500.0001870.003510BFGS 83TanhLinear
Table 4. Results of process machine learning for 30% of carrier content.
Table 4. Results of process machine learning for 30% of carrier content.
Blackcurrant
Powder Type
Name
ANN
Val
Quality
Test
Quality
Train
Quality
Train
Error
Test ErrorVal
Error
Train
Algorithm
Activation
Hidden
Activation
Output
F70zC70MLP 3-6-10.99530.99800.99310.01060.01190.0158BFGS 160TanhLinear
F70zW70MLP 3-10-10.99800.99680.99550.01880.03790.0465BFGS 200TanhLinear
F70zIN70MLP 3-10-10.99880.99970.99890.02250.01150.0247BFGS 192TanhLinear
F70zGA70MLP 3-7-10.99830.99910.99340.06390.03160.2513BFGS 82TanhLinear
F70zMD70MLP 3-7-10.99980.99970.99990.00910.01450.0073BFGS 93TanhLinear
C70zW70MLP 3-6-10.99730.99950.99860.00600.00250.0101BFGS 64TanhLinear
C70zIN70MLP 3-7-10.98170.99080.99310.16620.11770.0686BFGS 68TanhLinear
C70zGA70MLP 3-10-10.99340.99710.96020.13250.06810.0369BFGS 184TanhLinear
C70zMD70MLP 3-8-10.98930.99670.99470.38360.14180.6706BFGS 26TanhLinear
W70zIN70MLP 3-7-10.99780.99920.99940.00490.00260.0022BFGS 22ExponentialLinear
W70zGA70MLP 3-7-10.99900.99950.99950.00880.00520.0051BFGS 105TanhLinear
IN70zGA70MLP 3-7-10.99770.99890.99900.00530.00370.0030BFGS 50ExponentialLinear
IN70zMD70MLP 3-10-10.99840.99800.99930.01470.02790.0132BFGS 149ExponentialLinear
GA70zMD70MLP 3-10-10.99340.98990.99250.01490.03090.0236BFGS 187ExponentialLinear
W70zMD70MLP 3-6-10.99440.99570.99550.11260.09360.1452BFGS 178ExponentialLinear
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

Przybył, K.; Walkowiak, K.; Jedlińska, A.; Samborska, K.; Masewicz, Ł.; Biegalski, J.; Pawlak, T.; Koszela, K. Fruit Powder Analysis Using Machine Learning Based on Color and FTIR-ATR Spectroscopy—Case Study: Blackcurrant Powders. Appl. Sci. 2023, 13, 9098. https://doi.org/10.3390/app13169098

AMA Style

Przybył K, Walkowiak K, Jedlińska A, Samborska K, Masewicz Ł, Biegalski J, Pawlak T, Koszela K. Fruit Powder Analysis Using Machine Learning Based on Color and FTIR-ATR Spectroscopy—Case Study: Blackcurrant Powders. Applied Sciences. 2023; 13(16):9098. https://doi.org/10.3390/app13169098

Chicago/Turabian Style

Przybył, Krzysztof, Katarzyna Walkowiak, Aleksandra Jedlińska, Katarzyna Samborska, Łukasz Masewicz, Jakub Biegalski, Tomasz Pawlak, and Krzysztof Koszela. 2023. "Fruit Powder Analysis Using Machine Learning Based on Color and FTIR-ATR Spectroscopy—Case Study: Blackcurrant Powders" Applied Sciences 13, no. 16: 9098. https://doi.org/10.3390/app13169098

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