Next Article in Journal
Effects of Pectinase on Bacterial Succession during Hemp Retting
Previous Article in Journal
An Integrated Approach of Fuzzy AHP-TOPSIS for Multi-Criteria Decision-Making in Industrial Robot Selection
Previous Article in Special Issue
Modifications on the Processing Parameters of Traditional Pineapple Slices by Stabilized Sound Pressure of Multiple Frequency Ultrasonic-Assisted Osmotic Dehydration
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Processes
Volume 12
Issue 8
10.3390/pr12081724
processes-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

Deep Learning Prediction of Moisture and Color Kinetics of Apple Slices by Long Short-Term Memory as Affected by Blanching and Hot-Air Drying Conditions

by
Zehui Jia
,
Yanhong Liu
* and
Hongwei Xiao
College of Engineering, China Agricultural University, HaiDianDistrict, 17 Qinghua Donglu, Beijing 100083, China
*
Author to whom correspondence should be addressed.
Processes 2024, 12(8), 1724; https://doi.org/10.3390/pr12081724
Submission received: 12 June 2024 / Revised: 25 July 2024 / Accepted: 7 August 2024 / Published: 16 August 2024
(This article belongs to the Special Issue Drying Kinetics and Quality Control in Food Processing, 2nd Edition)
Browse Figures
Versions Notes

Abstract

:
This study aimed to improve apple slices’ color and drying kinetics by optimizing the hot-air drying process, utilizing machine and deep learning models. Different steam blanching times (30, 60, 90, and 120 s), drying temperatures (50, 55, 60, 65, and 70 °C), and humidity control methods (full humidity removal or temperature–humidity control) were examined. These factors significantly affected the quality of apple slices. 60 s blanching, 60 °C temperature, and full dehumidification represented the optimal drying conditions for apple slices’ dehydration, achieving better drying kinetics and the best color quality. However, the fastest drying process (40 min) was obtained at a 60 °C drying temperature combined with complete dehumidification after 90 s blanching. Furthermore, machine and deep learning models, including backpropagation (BP), convolutional neural network–long short-term memory (CNN-LSTM), temporal convolutional network (TCN), and long short-term memory (LSTM) networks, effectively predicted the moisture content and color variation in apple slices. Among these, LSTM networks demonstrated exceptional predictive performance with an R2 value exceeding 0.98, indicating superior accuracy. This study provides a scientific foundation for optimizing the drying process of apple slices and illustrates the potential application of deep learning in the agricultural processing and engineering fields.

1. Introduction

Apple is the third most widely cultivated fruit globally, with an estimated production of 95.84 million tons worldwide in 2022 (FAOSTAT) [1]. In addition to eating them fresh, apples can also be consumed as dried slices [2]. Hot-air drying is the most widely adopted technology for dry apple slices, due to the high drying rate and simple operation [3]. When subjected to hot air drying, the surface of apple slices tends to undergo pronounced browning and shrinkage [4], due to prolonged exposure to hot-dry air. Therefore, it is necessary to explore a novel approach to maximize efficiency and appearance quality during hot-air drying [5].
Various studies have shown the critical effect of the temperature and relative humidity on the hot-air drying time and the color of the dried apple slices [6]. Nagaya et al. [7] indicated that continuous dehumidification increased the likelihood of browning in apple slices, while staged humidity control reduced the drying time by approximately 81.7%, mitigating the occurrence of browning. Matys et al. [8] found that the optimal color retention was at 55 or 85 °C drying temperatures and an air humidity of 10 g water/m3 dry air. Blanching can effectively inactivate oxidative enzymes to protect color [9]. Aradwad et al. [9] found that blanching for 60 s lowered the browning of apple slices, reducing the color difference by 16% compared to the non-blanched group. The aforementioned research demonstrates that browning can be suppressed by optimizing the drying temperature and humidity as well as the blanching time, which can also shorten the drying time. During drying, the color of apple slices dynamically changes with time, moisture content, and drying intensity. Adjusting the drying parameters promptly when color deterioration becomes evident or when the drying rate slows down can optimize both the color [10].
Given the significant discrepancies in color and moisture content trends under various drying conditions, the reliability of simple mathematical models in predicting these variables remains limited. In the field of drying, simple mathematical models are typically employed to describe processes occurring under constant conditions [11]. However, the current trend in agricultural product drying tends towards variable processes. Zeng et al. [12] investigated the variation in the internal moisture content of peanuts under the conditions of varying drying temperatures at different stages of the drying process. In this context, empirical models are unable to accurately predict the outcome, prompting researchers to utilize on-line prediction based on real-time data. Li et al. [13] explored the use of real-time image information to assess the appearance quality of shiitake mushrooms, examining the changes in shrinkage, roundness, surface wrinkles, and texture characteristics during the drying process. However, machine learning technology, based on statistical algorithms, offers an effective alternative for model construction. This technology empowers machines to learn from vast amounts of historical data [14], applying this knowledge to predict and control processes. Currently, the most prevalent machine learning algorithms in the drying field include linear regression models, polynomial regression models, and others. In a study by [15], multivariate linear regression was employed to model the quality change during banana drying based on electronic nose sensor data, yielding a goodness of fit up to 0.7914. However, this fitting method, based on linear relationships, may be less effective for data comprising multiple responsible dimensions. Furthermore, it is sensitive to anomalies and outliers, potentially affecting the results. Deep learning offers an effective solution to these challenges. The technique is capable of learning complex nonlinear relationships in large datasets. Sun et al. [16] employed a back propagation (BP) neural network to predict water content changes in carrot, banana and apricot mushrooms during microwave vacuum drying, with R2 exceeding 0.99 in all cases. However, contrary to moisture prediction, when applying this model to color, color fluctuations caused the BP neural network to become stuck in local minima and then led to training failures. The long short-term memory (LSTM) networks have the capability to more comprehensively extract and understand the features within input data through its internal memory units and gating mechanisms. This enables more accurate predictions. Sabat et al. [17] employed a long short-term memory (LSTM) neural network to forecast the moisture content of potato flakes at varying drying temperatures. The resulting R2 values exceeded 0.97 in all cases. Chen et al. [18] successfully predicted the excess moisture of wood materials during the drying process with an R2 of 0.9846 by employing an LSTM model, based on factors such as feed rate and inlet air temperature. This study aimed to investigate the effects of blanching time, drying medium temperature, and humidity on the drying kinetics and color of apple slices. Secondly, we established a database and trained the LSTM model based on changes in moisture content and color over time during the experimental process. Finally, this LSTM model was utilized to predict the changes in moisture content and color of apple slices, determining the optimal hot-air drying conditions. The objective of this study was to provide a process monitoring method for improving the color of dried apple slices and reducing the drying time.

2. Materials and Methods

2.1. Raw Materials

Fresh apples (Qixia small Fuji) were purchased from a local supermarket in Beijing, China and stored at in a refrigerator at 4 ± 1 °C. The initial moisture content (on wet basis, w.b.) of the apples was determined by oven drying at 105 °C for 12 h [19], yielding a value of 84.03 ± 0.10%. The cleaned apples were sliced into thin slices with a thickness of 3 ± 0.5 mm using an electric slicer machine (Model: Electric Use Type) manufactured by Foshan Tiansui Electrical Appliance Co., LTD, Foshan, China.

2.2. Experimental Equipment and Design for Drying

A customized drying system, as illustrated in Figure 1, was utilized for all experiments conducted in this study. The system was designed and constructed in the laboratory to facilitate on-line inspection of the appearance quality of agricultural products and collect temperature, humidity, and speed data during the drying process [20].
The experimental protocol is outlined in Table 1 [10,21]. The cleaned apples were sliced into thin slices with a thickness of 3 ± 0.5 mm using a slicer (Model: Electric Use Type, manufactured by Foshan Tiansui Electrical Appliance Co., LTD., Foshan, China). During the experiment, one apple slice was placed in the top tray for the inspection of appearance data, while six apple slices were spread uniformly on each of the next three trays. The bottom panel was utilized for on-line weighing to acquire information on the materials’ drying characteristics. The experiment involved four different levels: (1) Different drying temperatures (50, 55, 60, 65, and 70 °C) were examined for the dehydration of fresh slices (treatments 1–5 in Table 1), with continuous fanning at an air velocity of 2 m/s. This range of temperatures was selected based on preliminary studies to cover common drying conditions and identify the most effective temperature for maintaining quality and drying efficiency. (2) Different steam blanching durations (30, 60, 90, and 120 s) were examined by subjecting the apple slices to ambient pressure, steam and drying at the aforementioned optimum temperature (treatments 6–9) while the moisture exhaust fan remained on and the wind speed was 2 m/s. These durations were chosen to assess the impact of blanching on drying kinetics and color retention, as preliminary tests indicated these time frames could significantly influence the outcomes. (3) Temperature—humidity controlled (THC) condition was conducted according to the pre-experiment results (treatment 10). The temperature with the smallest change in color and luster was applied, and the humidity was set to 30% for 30 min, after which it was reduced to 15% until the end of the drying process. This condition was designed based on pre-experimental results to minimize color change and optimize drying efficiency under controlled humidity conditions. (4) The combined effect of blanching and THC was examined (treatment 11). The apple slices were steam-blanched for the optimum duration and dried under the aforementioned THC conditions (120 s +THC). This treatment aimed to explore potential synergistic effects of blanching and controlled humidity on the drying process and quality of apple slices. Once the moisture content of the apple slices had fallen below 15% (w.b.), the data were saved, and the samples were cooled to room temperature and then stored in a sealed bag. Each set of experiments was repeated three times. In this experiment, the white balance of the camera was calibrated to standardize the illumination conditions and minimize the impact of surrounding light on the browning index calculation. The experimental setup included controlled light transmittance to ensure consistent lighting across all samples.

2.3. Drying Kinetics

The moisture ratio (MR) indicator is used to indicate the moisture content of a material at different drying moments under specific drying conditions. The moisture ratio (MR) was calculated according to Equation (1) [22]:
M R = M t − M e M 0 − M e
where solids are defined as the dry matter of the sample except for moisture, and moisture removed by drying is defined as the liquid phase, the value of M 0 represents the initial moisture content, while the value of M t represents the moisture content at a specific point in time. The value of M e represents the equilibrium moisture content, which is considerably smaller than the values of M t and M 0 . Consequently, Equation (1) can be simplified to Equation (2) [23]:
M R = M t M 0
The drying rate (g water/(g solid·h)) (DR) is the quantity of water evaporated per unit of time. DR of apple slices was calculated according to the methodology outlined in Equation (3) [24]:
D R = M t 1 − M t 2 t 2 − t 1
where the water content at moments t 1 and t 2 designated as M t 1 and M t 2 , respectively.

2.4. Determination of Color

The image of apple slices undergoing drying was segmented in real-time using the DeepLabV3+ [25] semantic segmentation model embedded inside the device, as illustrated in Figure 2. This process yields the region of the apple slices indicated in the image, taking the center of the region as the midpoint; coordinates were taken at eight points (all 200-pixel values away from the midpoint) and denoted as R, G, and B values uniformly around the region in Figure 3, which were then converted into L * (light/dark), a * (red/green), b * (yellow/blue), and the total color difference ΔE was calculated according to Equation (4) [26]:
Δ E = ( L 0 * − L 1 * ) 2 + ( a 0 * − a 1 * ) 2 + ( b 0 * − b 1 * ) 2
where ∆E represents the total color difference between the fresh apple slices and the dried ones. L 0 * , a 0 * , and b 0 * represent the color factors of the fresh apple slices, while L 1 * , a 1 * , and b 1 * represent those of the dried ones.
The browning index (BI) was calculated using Equations (5) and (6) [27]:
B I = 100 × ( X − 0.31 0.17 )
X = ( a * + 1.75 L * ) ( 6.645 L * + a * − 3.012 b * )

2.5. Neural Networks

2.5.1. Construction of Dataset

The dataset primarily consists of information regarding the changes in moisture content (d.b.) and color metrics of apple slices during hot-air drying at varying temperatures, along with different pretreatments and humidity control strategies. In this study, due to the inherent unpredictability of apple slices during the drying process, both the training and experiment sets comprise all available data.

2.5.2. Deep Learning Model

A BP neural network [28] and a TCN neural network [29] were utilized to predict the color and moisture content of the apple slices during drying under different conditions. The structure of the BP model and the TCN model are depicted in Figure 4 and Figure 5. They comprise input, hidden, and output layers. The BP model is designed for training feedforward neural networks and excels at handling static and nonlinear problems, but it struggles with sequence data. In contrast, the TCN model is tailored for sequence data, capturing long-term dependencies through causal and dilated convolutions, making it more effective for time-series analysis.
LSTM, as a distinctive recurrent neural network (RNN) structure, plays a pivotal role in addressing time-series issues and can effectively resolve gradient vanishing and gradient explosion concerns [30]. The cell model structure is illustrated in Figure 6. The three gating units represent the input, output, and forget gates. The formulas for the three gating units are presented below (7)–(12):
f t = σ ( W f ⋅ [ h t − 1 , x t ] + b f )
i t = σ ( W i ⋅ [ h t − 1 , x t ] + b i )
o t = σ ( W o ⋅ [ h t − 1 , x t ] + b o )
C t ' = tanh ( W c [ h t − 1 , x t ] + b c )
C t = f t C t − 1 + i t C t '
h t = o t tanh ( c t )
In the Equation, the forgetting, input, and output gates are represented by f t , i t , and o t , respectively, while σ represents the sigmoid activation function. Equation (7) uses ht−1 and xt with a sigmoid layer to determine data inclusion. After passing through the tanh layer using ht−1 and xt, fresh data are obtained using Equation (10). Equation (11) combines long-term memory Ct−1 and current data C’t. The input gate bias is bf, while the weight matrices are Wi. A sigmoid layer and a dot product, together with the forget gate, enable selective data transmission. Equation (8) decides whether to forget details from a previous cell, using Wi, Bi. Equations (9) and (12)’s ht−1 and xt inputs for the LSTM’s output unit, processing fresh data Ct by the tanh layer to obtain the outcome [31].
The LSTM model structure is depicted in Figure 7. The model comprises the input, hidden, and linear layers, as well as a PReLU activation layer and a Dropout layer, which have been incorporated to enhance the model’s generalization. The Adam algorithm was employed to train the model.
A CNN-LSTM neural network [32], depicted in Figure 8, was utilized to forecast apple slices’ color and moisture content during drying under various conditions. The CNN-LSTM model consists of one input layer, three hidden layers, and one output layer, along with a PReLU activation layer and Dropout to enhance nonlinearity and prevent overfitting. The integration of CNN and LSTM involves using CNN to extract local features from the data, which are then passed to LSTM to capture long-term dependencies over time, achieving comprehensive feature extraction and sequence modeling.

2.5.3. Training of the Model

The training environment for the applied models is presented in Table 2. The training parameters were as follows: the input datasets were all in CSV format, the model input dimension was 1, and the output dimension was 1. The model batch size was 128, adaptive moment estimation (Adam) was used as the optimizer, the model learning rate was 0.001, and the number of iterations was 5000.

2.6. Stability Analysis

All results presented in this paper were expressed as mean ± standard deviation. Analysis of variance (ANOVA) was performed using SPSS (version 26.0), with statistically significant differences defined as p < 0.05 [33]. Each experiment was repeated three times for each group.

3. Results and Analyses

3.1. Effect of Different Temperatures on Drying Characteristics and Color of Apple Slices

3.1.1. Effect of Temperature on Drying Characteristics of Apple Slices

Figure 9a illustrates the variation in moisture ratio (MR) of apple slices with drying time under different drying temperature conditions. The MR exhibited an exponential decreasing trend, being faster at the beginning, slower in the middle, and decreasing slowly to level off later. In the initial stages of drying, the considerable difference in temperature and humidity between the apple slices and the surrounding hot air, along with sufficient water on the surface of the slices, led to an increased drying rate. However, as the drying process continues, the temperature and humidity gradient between the slices and the surrounding hot air gradually diminishes, and the surface moisture content reduces, declining the drying rate [34]. It can be observed that the higher the temperature, the greater the slope of the drying curve, dehydrating the apple slices at a faster rate and shortening their drying time. When the drying temperature was 70 °C, the drying time of the apple slices was only 65 min, 35% shorter than the drying time at 50 °C. Figure 9b illustrates the significant impact of drying temperature on the drying rate of apple slices. It can be observed that the higher drying temperature enhanced the drying rate of apple slices due to the intensification of water migration within the slices [34].

3.1.2. Effect of Different Temperatures on Color Parameters of Apple Slices

The color parameters of apple slices dried by hot air at different drying temperatures are presented in Table 3. As the drying temperature increased, the color change in apple slices exhibited a trend of increasing, then decreasing, and finally increasing. Furthermore, the color change (ΔE) was significantly affected by the temperature change (p < 0.05). This trend is consistent with the results of Liu et al. [35], who studied the effect of temperature on the color of kiwifruit. They found that at a drying temperature of 60 °C, kiwifruits exhibited the lowest color and were closest to that of fresh kiwifruits. This may be because low temperatures prolong drying, increasing the time of oxidation, which in turn leads to a greater variation in color. High-temperature conditions, despite the brief drying time, resulted in a more pronounced color of the apple slices due to oxidative browning [36]. The observed color change exhibits a decreasing and then increasing trend with increasing temperature, reaching a minimum at 60 °C, which may be attributed to oxidative browning reactions occurring within the apple slices, which ultimately led to the appearance of the apple slices shown in the table [37]. This, in turn, resulted in a higher rate of decomposition of pigments, enzymes, and other chemicals within the apple slices. Moreover, this study revealed a significant difference (p < 0.05) in yellow-blueness b * , which exhibited a tendency to increase and then decrease with increasing temperature. This may be the reason for the significant difference in ΔE and BI of apple slices with increasing temperature. Consequently, a drying temperature of 60 °C was selected as the standard temperature for all subsequent studies to ensure better drying efficiency and color of dried products.

3.2. Effect of Steam Blanching Time on Drying Characteristics and Color of Apple Slices

3.2.1. Effect of Blanching Time on Drying Characteristics of Apple Slices

Figure 10a illustrates the variation in the MR of apple slices with drying time at 60 °C under different blanching times and under THC conditions. Until the moisture ratio of the apple slices reached 20% (w.b.), the moisture content did not continue to decrease with time, and the endpoint was not reached. Similar findings were also observed in the study of [34]. Figure 10b demonstrates the drying rate of the apple slices, showing that the blanching time significantly affects the drying rate of apple slices. The time required to reach the drying endpoint decreased with the increase in blanching time in the range of 30 to 90 s. This may be attributed to the exclusion of air from the apple tissues during blanching, which ruptures the cell wall and thus increases the permeability of the cell membrane, thereby promoting the diffusion of water [38,39].
Figure 11 illustrates the microstructure of apple slices subjected to different drying processes. After blanching, the surface tissues of apple slices exhibited shrinkage, cellular collapse, the disruption of cellular structure, and increased permeability. This is because of the weak connections between pectin and the cellulose micro primary fibers, so the stability of the skeleton (composed of cellulose and hemicellulose) is disturbed under high temperatures [40]. Additionally, the deep pores were observed in the scalded samples. These pores may be related to the expulsion of air from the cellular interstitial space during the scalding process [41], thereby increasing the heat and mass transfer processes during drying [40] and improving the drying rate.
However, when the blanching time was 120 s, the drying time increased instead. This unexpected outcome could be attributed to the fact that excessive blanching (120 s) resulted in the severe disruption of the cell wall structure of the apple slices, whereby intracellular sugars were released and then tightly bound to the water, which ultimately impeded the migration of water. This is visually manifested in the appearance of sticky fragments. Furthermore, prolonged blanching results in severe damage to the structure, leading to the excessive softening of the texture and increased resistance to water movement, thus prolonging the drying time of the apple slices and preventing them from eventually drying to the endpoint [42]. In conclusion, the shortest drying time (40 min) was achieved when the blanching time was 90 s, which was 20% shorter than that achieved with a 30 s blanching time. These results are in agreement with those previously reported by Wang et al. [43].

3.2.2. Effect of Blanching Time on the Color of Apple Slices

Previous studies have demonstrated that polyphenol oxidase (PPO) is responsible for the browning of apple slices during hot-air drying [44,45]. The color parameters of hot air-dried apple slices at 60 °C and different blanching times are presented in Table 4. A significant difference (p < 0.05) in ΔE was observed, affected by blanching time. As the blanching time increased, the ΔE of the apple slices exhibited a trend of increasing and then decreasing. This can be due to the PPO inactivation during blanching that increases with increasing blanching time [46]. This reduced the degree of browning of the dried apple slices [47], resulting in a smaller color change. However, prolonged blanching times can lead to excessive texture and nutrient loss, as well as the color degradation of apple slices [47]. These findings are aligned with those of Zhang et al. [47]about garlic pomace under steam blanching. Consequently, after 90 s blanching, the apple slices exhibited a lower color, superior drying characteristics, and an optimal drying process.

4. Construction of the Apple Slice Prediction Model

4.1. Results of Moisture Content by Predicted Models

The prediction results are shown in Figure 12 and Table 5. The study concluded that the XGB model outperformed other machine learning algorithms. At the same time, these results also substantiated that the deep learning algorithm outperforms the machine learning algorithm regarding fitting effectiveness, and the LSTM model has better accuracy than other deep learning models.
In Figure 12 and Figure 13, the BP model fits worse than the LSTM model when the drying temperature is 70 °C and the drying time is between 20 and 50 min. This because the LSTM model is better at processing sequence data with long-term dependencies and can capture the complex patterns and regularities in the data more accurately. In contrast, the BP neural network has difficulty fully learning and exploiting these patterns and regularities when dealing with this type of sequence data, providing a relatively large prediction error [28]. Similarly, the CNN-LSTM model did not fit well with the LSTM model at a temperature of 60 °C and a drying time of 50 to 70 min. This may because of the insufficient data sample size and insufficient training of the CNN-LSTM model [48]. Furthermore, at 90 s, the fitting effect of the TCN model was poor, and the error between the predicted and true values was large. This may be ascribed to the unstable effect of TCN [49] in predicting longer sequences. Nevertheless, the differences between the predicted and true values of the TCN model under other blanching time conditions were relatively small. This phenomenon suggests that TCN is not stable in prediction. In contrast, the LSTM model revealed a superior prediction.

4.2. Results of ΔE by Predicted Models

The prediction results are shown in Figure 14 and Figure 15. The study concluded that the XGB model outperformed other machine learning algorithms. This because XGB captures the nonlinear relationships of the data by combining multiple trees. However, the nonlinear modeling capability of decision trees is relatively limited. At the same time, these results substantiated that the deep learning algorithm outperforms the machine learning algorithm regarding fitting effectiveness. Deep learning models can learn more complex nonlinear functions through multi-layer nonlinear transformations to capture higher-order interactions in the data. The comparison of fitting effectiveness metrics in Table 6 underscored the superior accuracy of the LSTM model compared to other deep learning models. Notably, the LSTM model, standing out as the best-fitting algorithm among the deep learning approaches, demonstrated notable improvements over the other models. These results prove that the LSTM model surpasses other deep learning models, demonstrating its efficacy in predictive tasks.
The comparison of fitting effectiveness metrics in Figure 15 underscored the superior accuracy of the LSTM model compared to other deep learning models. Notably, the LSTM model, standing out as the best-fitting algorithm among the deep learning approaches, demonstrated notable improvements over the other models. This is because the BP network assumes that the data are independent and homogeneously distributed, limiting its ability to model non-stationary data [50]. In contrast, LSTM [51] can effectively capture and exploit long-term dependencies in the data by introducing gating mechanisms and internal states. Additionally, LSTM dynamically adjusts the parameters of the network through the gating mechanism, enabling it to adapt to the non-smoothness of the data and capture the dynamic changes in the dataset [52]. Therefore, the LSTM model’s superior performance in fitting effectiveness is evident based on the results. When dealing with a single feature, the CNN-LSTM [48] model might not have enough information to support accurate fitting due to the limited amount of information contained in that feature, resulting in poorer predictions. Additionally, while CNN [53] is adept at extracting spatial features, it might not be as effective in dealing with numerical sequences for prediction. The observed performance difference may be attributed to the TCN’s utilization of causal convolution. This mechanism might have limitations in pre-prediction future outcomes due to the reduced availability of historical information [49]. As the datasets expand and the model accesses more contextual information, predictions might improve gradually over time. Therefore, the LSTM model’s superior performance in fitting effectiveness is evident based on the results.
As verified by the validation set, LSTM reduces MAE and RMSE by 0.0015 and 0.0063, respectively; and increases R2 and R by 0.21% and 0.24%, respectively, compared to other best-fitting deep learning models.

5. Conclusions

This research investigates the influence of drying temperature, blanching time, and humidity control methods on the drying kinetics, color, and moisture content of apple slices during hot-air drying. The results show that blanching the apple slices for 60 s and setting the drying temperature at 60 °C, with the humidity control mode set at whole process dehumidification, resulted in minimal color differences in the dried apple slices and excellent drying characteristics. Moreover, when the drying temperature was 60 °C, the blanching time was 90 s, and the humidity control mode was set at complete moisture removal, the drying rate was the fastest, taking only 40 min for dehydration. Furthermore, machine learning and deep learning models were applied to predict the moisture content and color change in apple slices during the drying process. The findings affirm that deep learning models significantly outperform traditional machine learning models. Among the deep learning models, the LSTM model exhibited the best fitting effect and prediction accuracy, with R2 values exceeding 0.98. This highlights the LSTM model’s reliability for optimizing the drying process and predicting apple slices’ water content and color change. In conclusion, this study lays a scientific foundation for optimizing hot-air drying for apple slices’ dehydration and underscores the potential of deep learning applications in food engineering.

Author Contributions

Z.J.: Investigation, data curation, software, writing—original draft, methodology. Y.L.: funding acquisition, methodology, supervision, writing—review and editing. H.X.: guidance and revision of the paper. All authors have read and agreed to the published version of the manuscript.

Funding

This research was financially supported by Guangdong Institute of Modern Agricultural Equipment, China (NO. 2018B020241003-04).

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request. Code sharing is not applicable to this article as the project is not yet fully completed.

Conflicts of Interest

The author hereby declares that this is a research article, researched by the author himself, without conflicts of interest with any individual or collective.

References

  1. FAOSTAT. Available online: https://www.fao.org/faostat/en/#data/QCL/visualize (accessed on 2 May 2024).
  2. Tepe, T.K.; Tepe, B. The Comparison of Drying and Rehydration Characteristics of Intermittent-Microwave and Hot-Air Dried-Apple Slices. Heat Mass Transf. 2020, 56, 3047–3057. [Google Scholar] [CrossRef]
  3. Zarein, M.; Samadi, S.H.; Ghobadian, B. Investigation of Microwave Dryer Effect on Energy Efficiency during Drying of Apple Slices. J. Saudi Soc. Agric. Sci. 2015, 14, 41–47. [Google Scholar] [CrossRef]
  4. Michalska, A.; Lech, K.; Figiel, A.; Łysiak, G.P. The Influence of Selected Drying Methods on the Physical Properties of Dried Apples Cv. Jonagold Grown in Different Locations in Europe. Int. J. Food Eng. 2017, 13, 20160312. [Google Scholar] [CrossRef]
  5. Teymori-Omran, M.; Askari Asli-Ardeh, E.; Taghinezhad, E.; Motevali, A.; Szumny, A.; Nowacka, M. Enhancing Energy Efficiency and Retention of Bioactive Compounds in Apple Drying: Comparative Analysis of Combined Hot Air–Infrared Drying Strategies. Appl. Sci. 2023, 13, 7612. [Google Scholar] [CrossRef]
  6. Shewale, S.R.; Rajoriya, D.; Hebbar, H.U. Low Humidity Air Drying of Apple Slices: Effect of EMR Pretreatment on Mass Transfer Parameters, Energy Efficiency and Quality. Innov. Food Sci. Emerg. Technol. 2019, 55, 1–10. [Google Scholar] [CrossRef]
  7. Nagaya, K.; Li, Y.; Jin, Z.; Fukumuro, M.; Ando, Y.; Akaishi, A. Low-Temperature Desiccant-Based Food Drying System with Airflow and Temperature Control. J. Food Eng. 2006, 75, 71–77. [Google Scholar] [CrossRef]
  8. Matys, A.; Witrowa-Rajchert, D.; Parniakov, O.; Wiktor, A. Assessment of the Effect of Air Humidity and Temperature on Convective Drying of Apple with Pulsed Electric Field Pretreatment. LWT 2023, 188, 115455. [Google Scholar] [CrossRef]
  9. Aradwad, P.P.; Thirumani Venkatesh, A.K.; Mani, I. Infrared Drying of Apple (Malus Domestica) Slices: Effect on Drying and Color Kinetics, Texture, Rehydration, and Microstructure. J. Food Process. Eng. 2023, 46, e14218. [Google Scholar] [CrossRef]
  10. Dajbych, O.; Kabutey, A.; Mizera, Č.; Herák, D. Investigation of the Effects of Infrared and Hot Air Oven Drying Methods on Drying Behaviour and Colour Parameters of Red Delicious Apple Slices. Processes 2023, 11, 3027. [Google Scholar] [CrossRef]
  11. Rurush, E.; Alvarado, M.; Palacios, P.; Flores, Y.; Rojas, M.L.; Miano, A.C. Drying Kinetics of Blueberry Pulp and Mass Transfer Parameters: Effect of Hot Air and Refractance Window Drying at Different Temperatures. J. Food Eng. 2022, 320, 110929. [Google Scholar] [CrossRef]
  12. Zeng, S.; Du, Z.; Lv, W.; Li, D.; Su, D.; Lv, H. Experimental Study on the Hygrothermal Dynamics of Peanut (Arachis hypogaea Linn.) in the Process of Superposition and Variable Temperature Drying. Dry. Technol. 2022, 40, 1463–1477. [Google Scholar] [CrossRef]
  13. Li, X.; Zhang, Y.; Xie, Y.; Yu, X.; Zhu, G.; Gao, Z.; Xiao, H.; Liu, Y. Effects of relative humidity on the exterior quality of shiitake mushrooms (Lentinus edodes) during hot air drying. Trans. Chin. Soc. Agric. Eng. 2020, 36, 281–291. (In Chinese) [Google Scholar]
  14. Martynenko, A.; Misra, N.N. Machine Learning in Drying. Dry. Technol. 2020, 38, 596–609. [Google Scholar] [CrossRef]
  15. Sanaeifar, A.; Mohtasebi, S.S.; Ghasemi-Varnamkhasti, M.; Ahmadi, H. Application of MOS Based Electronic Nose for the Prediction of Banana Quality Properties. Measurement 2016, 82, 105–114. [Google Scholar] [CrossRef]
  16. Sun, Q.; Zhang, M.; Yang, P. Combination of LF-NMR and BP-ANN to Monitor Water States of Typical Fruits and Vegetables during Microwave Vacuum Drying. LWT 2019, 116, 108548. [Google Scholar] [CrossRef]
  17. Sabat, M.; Kotwaliwale, N.; Shelake, P.S. Real-Time Determination of Moisture Content of Potato Slices during Drying Using Long Short-Term Memory from Image Morphometric Parameters. Food Bioprod. Process. 2023, 140, 132–143. [Google Scholar] [CrossRef]
  18. Chen, F.; Gao, X.; Xia, X.; Xu, J. Using LSTM and PSO Techniques for Predicting Moisture Content of Poplar Fibers by Impulse-Cyclone Drying. PLoS ONE 2022, 17, e0266186. [Google Scholar] [CrossRef] [PubMed]
  19. Yang, R.-L.; Li, Q.; Hu, Q.-P. Physicochemical Properties, Microstructures, Nutritional Components, and Free Amino Acids of Pleurotus Eryngii as Affected by Different Drying Methods. Sci. Rep. 2020, 10, 121. [Google Scholar] [CrossRef] [PubMed]
  20. Mowafy, S.; Guo, J.; Lei, D.; Liu, Y. Application of Novel Blanching and Drying Technologies Improves the Potato Drying Kinetics and Maintains Its Physicochemical Attributes and Flour Functional Properties. Innov. Food Sci. Emerg. Technol. 2024, 94, 103648. [Google Scholar] [CrossRef]
  21. Xiao, B.; Xiao, B.; Liu, X. Drying characteristic, quality, and process optimization of apple slices during two-stage relative humidity-controlled drying. Food Ferment. Ind. 2023, 49, 237–246. [Google Scholar] [CrossRef]
  22. Noshad, M.; Ghasemi, P. Influence of Freezing Pretreatments on Kinetics of Convective Air-Drying and Quality of Grapes. Food Biosci. 2020, 38, 100763. [Google Scholar] [CrossRef]
  23. Süfer, Ö.; Palazoğlu, T.K. A Study on Hot-Air Drying of Pomegranate. J. Therm. Anal. Calorim. 2019, 137, 1981–1990. [Google Scholar] [CrossRef]
  24. Li, M.; Chen, Y.; Wang, X.; Cheng, S.; Liu, F.; Huang, L. Determination of Drying Kinetics and Quality Changes of Panax Quinquefolium L. Dried in Hot-Blast Air. LWT 2019, 116, 108563. [Google Scholar] [CrossRef]
  25. Xie, J.; Jing, T.; Chen, B.; Peng, J.; Zhang, X.; He, P.; Yin, H.; Sun, D.; Wang, W.; Xiao, A.; et al. Method for Segmentation of Litchi Branches Based on the Improved DeepLabv3+. Agronomy 2022, 12, 2812. [Google Scholar] [CrossRef]
  26. Pei, Y.; Li, Z.; Song, C.; Li, J.; Xu, W.; Zhu, G. Analysis and Modelling of Temperature and Moisture Gradient for Ginger Slices in Hot Air Drying. J. Food Eng. 2022, 323, 111009. [Google Scholar] [CrossRef]
  27. Karimi, S.; Layeghinia, N.; Abbasi, H. Microwave Pretreatment Followed by Associated Microwave-Hot Air Drying of Gundelia tournefortii L.: Drying Kinetics, Energy Consumption and Quality Characteristics. Heat Mass Transf. 2021, 57, 133–146. [Google Scholar] [CrossRef]
  28. Bai, H.; Chu, Z.; Wang, D.; Bao, Y.; Qin, L.; Zheng, Y.; Li, F. Predictive Control of Microwave Hot-Air Coupled Drying Model Based on GWO-BP Neural Network. Dry. Technol. 2023, 41, 1148–1158. [Google Scholar] [CrossRef]
  29. Jiang, G.; Duan, Z.; Zhao, Q.; Li, D.; Luan, Y. Remaining Useful Life Prediction of Rolling Bearings Based on TCN-MSA. Meas. Sci. Technol. 2023, 35, 25125. [Google Scholar] [CrossRef]
  30. Berhich, A.; Belouadha, F.-Z.; Kabbaj, M.I. An Attention-Based LSTM Network for Large Earthquake Prediction. Soil Dyn. Earthq. Eng. 2023, 165, 107663. [Google Scholar] [CrossRef]
  31. Rayan, A.; Holyl Alruwaili, S.; Alaerjan, A.S.; Alanazi, S.; Taloba, A.I.; Shahin, O.R.; Salem, M. Utilizing CNN-LSTM Techniques for the Enhancement of Medical Systems. Alex. Eng. J. 2023, 72, 323–338. [Google Scholar] [CrossRef]
  32. Niu, Z.; Qi, H.; Sun, A.; Ren, Y.; He, M.; Gao, B. Efficient and Robust CNN-LSTM Prediction of Flame Temperature Aided Light Field Online Tomography. Sci. China Technol. Sci. 2024, 67, 271–284. [Google Scholar] [CrossRef]
  33. Moscetti, R.; Raponi, F.; Ferri, S.; Colantoni, A.; Monarca, D.; Massantini, R. Real-Time Monitoring of Organic Apple (Var. Gala) during Hot-Air Drying Using near-Infrared Spectroscopy. J. Food Eng. 2018, 222, 139–150. [Google Scholar] [CrossRef]
  34. Liu, H.; Jiao, J.; Tian, Y.; Liu, J.; Yuan, P.; Wu, X. Drying Kinetics of Pleurotus eryngii Slices during Hot Air Drying. Open Phys. 2022, 20, 265–273. [Google Scholar] [CrossRef]
  35. Liu, Z.-L.; Wei, Z.-Y.; Vidyarthi, S.K.; Pan, Z.; Zielinska, M.; Deng, L.-Z.; Wang, Q.-H.; Wei, Q.; Xiao, H.-W. Pulsed Vacuum Drying of Kiwifruit Slices and Drying Process Optimization Based on Artificial Neural Network. Dry. Technol. 2020, 39, 405–417. [Google Scholar] [CrossRef]
  36. Huang, X.; Li, Y.; Zhou, X.; Wang, J.; Zhang, Q.; Yang, X.; Zhu, L.; Geng, Z. Prediction of Apple Slices Drying Kinetic during Infrared-Assisted-Hot Air Drying by Deep Neural Networks. Foods 2022, 11, 3486. [Google Scholar] [CrossRef]
  37. Zhang, J.; Zheng, X.; Xiao, H.; Shan, C.; Li, Y.; Yang, T. Quality and Process Optimization of Infrared Combined Hot Air Drying of Yam Slices Based on BP Neural Network and Gray Wolf Algorithm. Foods 2024, 13, 434. [Google Scholar] [CrossRef]
  38. Xiao, H.-W.; Yao, X.-D.; Lin, H.; Yang, W.-X.; Meng, J.-S.; Gao, Z.-J. Effect of Ssb (Superheated Steam Blanching) Time and Drying Temperature on Hot Air Impingement Drying Kinetics and Quality Attributes of Yam Slices. J. Food Process. Eng. 2012, 35, 370–390. [Google Scholar] [CrossRef]
  39. Ramesh, M.N.; Wolf, W.; Tevini, D.; Jung, G. Influence of Processing Parameters on the Drying of Spice Paprika. J. Food Eng. 2001, 49, 63–72. [Google Scholar] [CrossRef]
  40. Wang, J.; Chen, Y.; Wang, H.; Wang, S.; Lin, Z.; Zhao, L.; Xu, H. Ethanol and Blanching Pretreatments Change the Moisture Transfer and Physicochemical Properties of Apple Slices via Microstructure and Cell-Wall Polysaccharides Nanostructure Modification. Food Chem. 2022, 381, 132274. [Google Scholar] [CrossRef]
  41. Lin, Z.; Geng, Z.; Liang, W.; Zhu, H.; Ye, J.; Wang, J.; Xu, H. Steam Blanching and Ethanol Pretreatment Enhance Drying Rates and Improve the Quality Attributes of Apple Slices via Microstructure Modification. J. Food Process. Preserv. 2022, 46, e17256. [Google Scholar] [CrossRef]
  42. Xie, Y.-K.; Li, X.-Y.; Zhang, Y.; Zheng, Z.-A.; Huang, L.-Q.; Liu, D.-H.; Xiao, H.-W.; Liu, Y.-H. Effects of High-Humidity Hot Air Impingement Steaming on Gastrodia elata: Steaming Degree, Weight Loss, Texture, Drying Kinetics, Microstructure and Active Components. Food Bioprod. Process. 2021, 127, 255–265. [Google Scholar] [CrossRef]
  43. Wang, H.; Meng, J.-S.; Raghavan, G.S.V.; Orsat, V.; Yu, X.-L.; Liu, Z.-L.; Zheng, Z.-A.; Wang, S.-Y.; Xiao, H.-W. Vacuum-Steam Pulsed Blanching (VSPB) Enhances Drying Quality, Shortens the Drying Time of Gingers by Inactivating Enzymes, Altering Texture, Microstructure and Ultrastructure. LWT 2022, 154, 112714. [Google Scholar] [CrossRef]
  44. Dag, D.; Farmanfarmaee, A.; Kong, F.; Jung, J.; McGorrin, R.J.; Zhao, Y. Feasibility of Simultaneous Drying and Blanching Inshell Hazelnuts (Corylus avellana L.) Using Hot Air–Assisted Radio Frequency (HARF) Heating. Food Bioprocess Technol. 2023, 16, 404–419. [Google Scholar] [CrossRef]
  45. Moscetti, R.; Raponi, F.; Monarca, D.; Bedini, G.; Ferri, S.; Massantini, R. Effects of Hot-Water and Steam Blanching of Sliced Potato on Polyphenol Oxidase Activity. Int. J. Food Sci. Technol. 2019, 54, 403–411. [Google Scholar] [CrossRef]
  46. Zhang, A.-A.; Ha, B.-E.; Chen, C.; Xu, M.-Q.; Wang, Q.-H.; Xie, L.; Zheng, Z.-A.; Zhang, J.-S.; Lv, W.-Q.; Xiao, H.-W. Vacuum-Steam Pulsed Blanching: An Emerging Method to Enhance Texture Softening, Drying Behavior and Physicochemical Properties of Cornus officinalis. J. Food Sci. 2024, 89, 202–216. [Google Scholar] [CrossRef] [PubMed]
  47. Zhang, X.; Shi, Q.; Gao, T.; Zhang, Z.; Guo, C.; Fu, H.; Wang, Y. Developing Radio Frequency Blanching Process of Apple Slice. J. Food Eng. 2020, 273, 109832. [Google Scholar] [CrossRef]
  48. Li, Q.; Guan, X.; Liu, J. A CNN-LSTM Framework for Flight Delay Prediction. Expert Syst. Appl. 2023, 227, 120287. [Google Scholar] [CrossRef]
  49. Sun, P.; Wang, J.; Yan, Z. Ultra-Short-Term Wind Speed Prediction Based on TCN-MCM-EKF. Energy Rep. 2024, 11, 2127–2140. [Google Scholar] [CrossRef]
  50. Wang, H.; Che, G.; Wan, L.; Wang, X.; Tang, H. Combination of LF-NMR and BP-ANN to Monitor the Moisture Content of Rice during Hot-Air Drying. J. Food Process. Eng. 2022, 45, e14102. [Google Scholar] [CrossRef]
  51. Sabat, M.; Kotwaliwale, N.; Chakraborty, S.K.; Kumar, M. Long Short-Term Memory Based Real-Time Monitoring of Potato Slice Drying Using Image Chromatic Features. J. Food Process. Preserv. 2022, 46, e17232. [Google Scholar] [CrossRef]
  52. Zhou, H.; Huang, M.; Zhu, Q.; Zhang, M. Developing C-LSTM Model for Evaluating Moisture Content of Carrot Slices during Drying. Dry. Technol. 2022, 40, 2964–2974. [Google Scholar] [CrossRef]
  53. Jalal Uddin, M.; Li, Y.; Abdus Sattar, M.; Mistry, S. Climatic water balance forecasting with machine learning and deep learning models over Bangladesh. Int. J. Climatol. 2022, 42, 10083–10106. [Google Scholar] [CrossRef]
Processes 12 01724 g001
Figure 1. Schematic diagram (a) and photo (b) of the humidity-controlled hot-air drying system. 1—Raspberry Pi control component, 2—control box, 3—industrial camera, 4—light source, 5—control screen, 6—drying chamber, 7—air return pipe, 8—heating pipe, 9—centrifugal fan, 10—exhaust pipe, 11—water tank, 12—humidification device, 13—turbulence fan support frame, 14—weight sensor, 15—dehumidification fan, 16—turbulence fan. Reproduced with permission from Samir et al. [20], Innovative Food Science & Emerging Technologies; published by Elsevier, 2024.
Figure 1. Schematic diagram (a) and photo (b) of the humidity-controlled hot-air drying system. 1—Raspberry Pi control component, 2—control box, 3—industrial camera, 4—light source, 5—control screen, 6—drying chamber, 7—air return pipe, 8—heating pipe, 9—centrifugal fan, 10—exhaust pipe, 11—water tank, 12—humidification device, 13—turbulence fan support frame, 14—weight sensor, 15—dehumidification fan, 16—turbulence fan. Reproduced with permission from Samir et al. [20], Innovative Food Science & Emerging Technologies; published by Elsevier, 2024.
Processes 12 01724 g001
Processes 12 01724 g002
Figure 2. Structure diagram of the conducted DeepLabV3+ model [25].
Figure 2. Structure diagram of the conducted DeepLabV3+ model [25].
Processes 12 01724 g002
Processes 12 01724 g003
Figure 3. The diagram of the color collection of the apple slice.
Figure 3. The diagram of the color collection of the apple slice.
Processes 12 01724 g003
Processes 12 01724 g004
Figure 4. The structure of back propagation (BP) for predicting the color and moisture content of the apple slices.
Figure 4. The structure of back propagation (BP) for predicting the color and moisture content of the apple slices.
Processes 12 01724 g004
Processes 12 01724 g005
Figure 5. The structure of the temporal convolutional network (TCN) applied to forecast apple slices’ color and moisture content.
Figure 5. The structure of the temporal convolutional network (TCN) applied to forecast apple slices’ color and moisture content.
Processes 12 01724 g005
Processes 12 01724 g006
Figure 6. Schematic diagram of the long short-term memory LSTM unit structure.
Figure 6. Schematic diagram of the long short-term memory LSTM unit structure.
Processes 12 01724 g006
Processes 12 01724 g007
Figure 7. The structure of the long short-term memory (LSTM).
Figure 7. The structure of the long short-term memory (LSTM).
Processes 12 01724 g007
Processes 12 01724 g008
Figure 8. The structure of convolutional neural networks–long short-term memory (CNN-LSTM).
Figure 8. The structure of convolutional neural networks–long short-term memory (CNN-LSTM).
Processes 12 01724 g008
Processes 12 01724 g009
Figure 9. Water content (a) and drying rate (b) curves of apple slices at different drying temperatures.
Figure 9. Water content (a) and drying rate (b) curves of apple slices at different drying temperatures.
Processes 12 01724 g009
Processes 12 01724 g010
Figure 10. Water content (a) and drying rate (b) curves of apple slices at different pre-treatment processes.
Figure 10. Water content (a) and drying rate (b) curves of apple slices at different pre-treatment processes.
Processes 12 01724 g010
Processes 12 01724 g011
Figure 11. SEM micrographs of apples dried in hot air under different experimental conditions: (a1–a3) represents hot-air drying at a drying temperature of 60 °C; (b1–b3) represents hot-air drying at a blanching time of 30 s; (c1–c3) represents hot-air drying at a blanching time of 120 s; (d1–d3) represents hot-air drying at a blanching time of 120 s with THC; and (e1–e3) represents hot-air drying at a drying temperature of 60 °C with THC. The number following the letter represents the degree of magnification, with the numbers 1–3 indicating an increase in magnification.
Figure 11. SEM micrographs of apples dried in hot air under different experimental conditions: (a1–a3) represents hot-air drying at a drying temperature of 60 °C; (b1–b3) represents hot-air drying at a blanching time of 30 s; (c1–c3) represents hot-air drying at a blanching time of 120 s; (d1–d3) represents hot-air drying at a blanching time of 120 s with THC; and (e1–e3) represents hot-air drying at a drying temperature of 60 °C with THC. The number following the letter represents the degree of magnification, with the numbers 1–3 indicating an increase in magnification.
Processes 12 01724 g011
Processes 12 01724 g012
Figure 12. Results of predicted moisture content by models at (a) 50 °C, (b) 55 °C, (c) 60 °C, (d) 65 °C, (e) 70 °C.
Figure 12. Results of predicted moisture content by models at (a) 50 °C, (b) 55 °C, (c) 60 °C, (d) 65 °C, (e) 70 °C.
Processes 12 01724 g012
Processes 12 01724 g013
Figure 13. Results of predicted moisture content by models at (a) 30 s, (b) 60 s, (c) 90 s, (d) 120 s, (e) THC, (f) 120 s + THC.
Figure 13. Results of predicted moisture content by models at (a) 30 s, (b) 60 s, (c) 90 s, (d) 120 s, (e) THC, (f) 120 s + THC.
Processes 12 01724 g013
Processes 12 01724 g014
Figure 14. Results of predicted ΔE by models at (a) 50 °C, (b) 55 °C, (c) 60 °C, (d) 65 °C, (e) 70 °C.
Figure 14. Results of predicted ΔE by models at (a) 50 °C, (b) 55 °C, (c) 60 °C, (d) 65 °C, (e) 70 °C.
Processes 12 01724 g014
Processes 12 01724 g015
Figure 15. Comparison of predicted ΔE by models at (a) 30 s, (b) 60 s, (c) 90 s, (d) 120 s, (e) THC, (f) 120 s + THC.
Figure 15. Comparison of predicted ΔE by models at (a) 30 s, (b) 60 s, (c) 90 s, (d) 120 s, (e) THC, (f) 120 s + THC.
Processes 12 01724 g015
Table 1. Experimental design of drying apple slices, with all factors.
Table 1. Experimental design of drying apple slices, with all factors.
NO.Blanching Time
(s)		Temperature
(°C)		Humidity
10500
20550
30600
40650
50700
630600
760600
890600
9120600
1006030 ± 4% for 30 min—15 ± 4%
111206030 ± 4% for 30 min—15 ± 4%
Table 2. The experiment environment of the machine learning and deep learning models.
Table 2. The experiment environment of the machine learning and deep learning models.
ConfigurationParameter
CPUIntel Core i7-13700K
GPUNvidia GeForce RTX 4060Ti 16 GB
Operating systemWindows 11
Accelerated environmentCUDA 12.2
Table 3. Color metrics of apple slices at different drying temperatures.
Table 3. Color metrics of apple slices at different drying temperatures.
Temperature
(°C)		L*a*b*ΔEBI
5081.45 ± 1.29 a−1.70 ± 0.58 a14.91 ± 0.02 b6.98 ± 0.19 c18.16 ± 0.90 d
5579.01 ± 2.08 a−2.18 ± 0.29 a21.04 ± 0.29 b10.40 ± 1.07 a28.06 ± 0.67 a
6082.37 ± 1.32 a−2.46 ± 1.07 a15.07 ± 0.29 b5.16 ± 0.02 d17.46 ± 1.73 d
6580.76 ± 1.73 a−2.02 ± 1.41 a19.28 ± 1.68 a8.11 ± 0.42 bc24.71 ± 0.72 b
7080.10 ± 0.70 a−1.30 ± 1.42 a16.19 ± 0.70 a8.93 ± 0.18 b20.81 ± 0.50 c
Samples with the same lower case letter in the same column showed no statistically significant differences for their mean values at the 95.0% confidence level (Fisher LSD: 0.05).
Table 4. Color measurement of apple slices at different blanching times.
Table 4. Color measurement of apple slices at different blanching times.
TimeL*a*b*ΔEBI
30 s81.565 ± 3.71 a−3.445 ± 2.38 a14.63 ± 1.43 a2.4637 ± 0.63 b16.08
60 s81.08 ± 0.99 a−2.755 ± 0.06 a14.905 ± 1.25 a4.4437 ± 0.15 ab17.24
90 s81.38 ± 0.17 a−4.62 ± 0.48 a17.27 ± 0.69 a4.0828 ± 0.00 a18.90
120 s80.985 ± 0.71 a−2.7 ± 1.65 a16.425 ± 1.24 a3.3513 ± 0.88 a19.58
Samples with the same lower case letter in the same column showed no statistically significant differences for their mean values at the 95.0% confidence level (Fisher LSD:0.05).
Table 5. The performance of models in terms of moisture content parameters.
Table 5. The performance of models in terms of moisture content parameters.
Models60 °C90 sTHC
MAERMSER2RMAERMSER2RMAERMSER2R
XGB0.12400.15430.99000.99000.12580.17590.98930.99460.08750.10500.99600.9980
TCN0.02620.04040.99900.99950.02770.03810.99920.99960.04090.05390.99840.9996
LSTM0.00690.00910.99990.99990.02170.09910.99630.99810.00800.01010.99990.9999
Table 6. The performance of models for ΔE.
Table 6. The performance of models for ΔE.
Models65 °C120 sTHC
MAERMSER2RMAERMSER2RMAERMSER2R
XGB0.40350.62090.71440.82670.32280.42720.90380.95020.28670.41780.92990.9624
TCN0.01600.02200.99920.99960.08030.11530.99870.99930.10830.21550.88760.9839
LSTM0.01020.01340.99980.99990.01630.02180.99950.99970.01060.01380.99970.9998
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

Jia, Z.; Liu, Y.; Xiao, H. Deep Learning Prediction of Moisture and Color Kinetics of Apple Slices by Long Short-Term Memory as Affected by Blanching and Hot-Air Drying Conditions. Processes 2024, 12, 1724. https://doi.org/10.3390/pr12081724

AMA Style

Jia Z, Liu Y, Xiao H. Deep Learning Prediction of Moisture and Color Kinetics of Apple Slices by Long Short-Term Memory as Affected by Blanching and Hot-Air Drying Conditions. Processes. 2024; 12(8):1724. https://doi.org/10.3390/pr12081724

Chicago/Turabian Style

Jia, Zehui, Yanhong Liu, and Hongwei Xiao. 2024. "Deep Learning Prediction of Moisture and Color Kinetics of Apple Slices by Long Short-Term Memory as Affected by Blanching and Hot-Air Drying Conditions" Processes 12, no. 8: 1724. https://doi.org/10.3390/pr12081724

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