Optimization of the Appearance Quality in CO2 Processed Ready-to-Eat Carrots through Image Analysis

A high-pressure CO2 process applied to ready-to-eat food products guarantees an increase of both their microbial safety and shelf-life. However, the treatment often produces unwanted changes in the visual appearance of products depending on the adopted process conditions. Accordingly, the alteration of the visual appearance influences consumers’ perception and acceptability. This study aims at identifying the optimal treatment conditions in terms of visual appearance by using an artificial vision system. The developed methodology was applied to fresh-cut carrots (Daucus carota) as the test product. The results showed that carrots packaged in 100% CO2 and subsequently treated at 6 MPa and 40 °C for 15 min maintained an appearance similar to the fresh product for up to 7 days of storage at 4 °C. Mild appearance changes were identified at 7 and 14 days of storage in the processed products. Microbiological analysis performed on the optimal treatment condition showed the microbiological stability of the samples up to 14 days of storage at 4 °C. The artificial vision system, successfully applied to the CO2 pasteurization process, can easily be applied to any food process involving changes in the appearance of any food product.


Introduction
In the last decade, the consumption of ready-to-eat (RTE) products has widely increased. Among the RTE products, such as fresh-and-cut fruits and vegetables, carrots' market increased by 5.5% between 2020 and 2021, being the 5th most produced vegetable in the US in 2020 [1]. Raw RTE carrots are usually packaged in a modified atmosphere packaging (MAP) which extends the storage time and increases the product shelf life. Due to the recent outbreaks in RTE foods [2,3], microbial safety is of primary importance and new strategies should be implemented to reduce the risk of food contamination. Recently, MAP has been coupled with innovative low temperature treatments to increase the microbiological safety, such as gamma radiation [4], ultraviolet light [5], ozone [6], high-voltage electrostatic fields [7], and high pressure [8]. Spilimbergo et al. [9] proposed a new method to inactivate the microbial population in food at low temperature which combines the advantages of supercritical CO 2 inactivation within a MAP (ScCO 2 -MAP). Similar to traditional supercritical CO 2 (ScCO 2 ) treatment [10,11], ScCO 2 -MAP uses ScCO 2 to inactivate microorganisms, but acts directly inside the packaging, thus reducing the risk of cross-contamination after processing.
The adoption of novel preservation technologies by industry has to comply not only with the microbial safety of the food, but also with the maintenance of fresh food character-istics. Indeed, the sensorial characteristics of a RTE product, such as visual appearance, texture, moisture content, and flavor, are the major driving factors affecting sensory perception and consumer preferences [12,13]. In this context, high-pressure technologies affect the texture and color of the products and these effects depend on the adopted processing conditions and the specific product [14]. Fast, cheap, and straightforward methods to identify the optimal processing conditions which preserve and/or minimize the visual changes in the food are of paramount importance.
The visual appearance of food products is one of the most difficult parameters to assess in an objective and repeatable manner. The preservation of a fresh-like appearance is usually evaluated through color measurement [15], because color changes have a primary role in food acceptability and preference [16]. However, the analytical methodology for color measurement based on a colorimeter is time-consuming and suffers from several drawbacks, mostly related to the high sensitivity to color inhomogeneities in the surface of the inspected product [17]. Since the color measurements are performed in specific spots, they capture only local color information, completely disregarding information on the superficial structure of the food matrix. This structural information, captured by the spatial distribution of color over the food surface, is another important factor affecting consumers' perception [18].
Artificial vision systems are effective tools for the evaluation of color changes in fruits, vegetables, and meat. In fact, image analysis is an accurate, repeatable, and inexpensive way to detect even minor differences in product appearance. Artificial vision systems have been used for color measurements [19,20], for the estimation of chemical parameters [21], for food authentication [22], and for defects identification [23]. For example, in recent years, artificial vision systems have been used to evaluate the quality and to detect contaminants in the soybean industry [24], to grade the maturity level of mangoes [25], and to evaluate the quality and freshness of saffron [26], while systems based on hyperspectral images have been used to detect defects and diseases of orange skin [27]. Artificial vision systems exploit the information on the color distribution over the product surface to significantly improve their recognition and detection capabilities [22,28]. However, exploiting the information on the changes in the color distribution over the food surface is not straightforward, and appropriate multivariate techniques [29] are required to deal with the large variability in food appearance. Despite the use of artificial vision systems being widespread in the food industry, to our knowledge, they have not been previously applied to food undergoing ScCO 2 treatment in any form. The use of artificial vision systems in ScCO 2 food treatments could provide a reliable methodology to characterize the effect of these innovative preservation technologies on the visual appearance quality of food products. Furthermore, artificial vision systems could be exploited to reduce the experimental burden of microbiological tests, allowing the microbiological inactivation to be verified only in the samples which meet the consumers' acceptability criteria in terms of visual appearance.
In this study we developed an artificial vision system to evaluate, through multivariate hypothesis testing, the appearance of the food after the ScCO 2 -MAP food preservation process is carried out. The methodology is applied on fresh-cut carrots (Daucus carota) and is aimed at identifying the optimal treatment conditions in terms of pressure, temperature, time, and %CO 2 in the MAP to preserve a fresh-like appearance after processing and during storage (up to 14 days). In this way, the experimental campaign to verify the microbiological inactivation is considerably reduced and carried out in the restricted domain of the optimal process conditions identified by the artificial vision system as the ones that ensure a fresh-like product. Microbiological analysis is performed for the naturally present microorganisms (total mesophilic bacteria, yeasts, and molds) and inoculated E. coli.

Materials and Methods
In this study, the analyses are carried out in five steps: (i) preparation of the samples; (ii) definition of the experimental design; (iii) ScCO 2 -MAP treatment and storage; (iv) evaluation of the food appearance through the artificial vision system and identification of the optimal operating conditions to reduce the experimental domain for microbiological testing; and (v) execution of the microbiological testing on the optimal conditions identified by the artificial vision system and statistical analysis.

Sample Preparation and Storage
Fresh orange carrots (Daucus carota) harvested in mid-April, with approximatively the same dimensions and mature level, were purchased from a local farm market (Padova, Italy). Carrots were preliminary packaged by the vendor in plastic packages with no modified atmosphere and stored at 4 • C. After purchasing, carrots were uniformly stored at 4 • C and processed within 2 h.
Carrots were cut into round slices of 3.25 ± 0.25 g weight, 3.00 ± 0.25 cm diameter, and 3 ± 1 mm thickness just before processing. After processing, carrots were stored at 4 • C up to 14 days to perform a storage test.
The ScCO 2 -MAP treatment conditions were selected according to a 2 3−1 fractional factorial design of experiments (DoE) [30] with three replicates of the central point. The entire set of treatment conditions defined by the DoE was carried out for each MAP composition in two replicates. Three factors affecting process performance were considered: process temperature, process pressure, and treatment time. Two factor levels were considered: 25 • C and 40 • C for temperature; 6 and 12 MPa for pressure; and 15 and 45 min for treatment time. Based on the preliminary tests, the ranges of the process operating conditions were chosen (i) to match typical treatment conditions that are effective for the preservation of foods using the traditional ScCO 2 treatment [10,11,31,32] and do not significantly affect the food quality [14,33,34], and (ii) to test the effect on product appearance of both subcritical and supercritical CO 2 . The ScCO 2 -MAP treatment conditions tested in this study are reported in Table 1.  TC2  40  6  15  TC3  25  6  45  TC4  25  12  15  TC5 40 12 45 At each treatment condition, four packages were treated simultaneously. Each package was analyzed at a different control time along the storage test performed up to 14 days at 4 • C and subsequently discarded. The control times were set to 0, 3, 7, and 14 days of storage. Control times were selected to analyze the appearance changes progressively along an average shelf life for carrots with more emphasis on the first week of storage. Consequently, in this study 168 carrot packages were prepared, treated, and analyzed (3 MAP compositions × (4 fractional factorial + 3 central) treatment conditions × 4 time controls × 2 replicates).
Untreated control samples were prepared for each MAP composition. The control samples' preparation and storage were identical to that of all other samples, but they were packaged and stored without any treatment; they were analyzed at the same control times as treated samples. In this study, 9 control packages were prepared (3 MAP compositions × 3 time controls (3,7, and 14 days)).

High-Pressure CO 2 Treatment
The ScCO 2 -MAP treatment [9] was carried out in two steps ( Figure 1): (i) food samples were packaged in CO 2 MAP and subsequently (ii) subjected to high pressure using water as driving force in a multi-batch apparatus.
Untreated control samples were prepared for each MAP composition. The control samples' preparation and storage were identical to that of all other samples, but they were packaged and stored without any treatment; they were analyzed at the same control times as treated samples. In this study, 9 control packages were prepared (3 MAP compositions × 3 time controls (3,7, and 14 days)).

High-Pressure CO2 Treatment
The ScCO2-MAP treatment [9] was carried out in two steps ( Figure 1): (i) food samples were packaged in CO2 MAP and subsequently (ii) subjected to high pressure using water as driving force in a multi-batch apparatus.

MAP Packaging
MAP packages with an internal chamber of 10 cm × 10 cm and a volume of 100 ± 10 mL were produced from a high gas barrier multilayer (PA/EVOH/PA/PE) film (Euralpack, Schoten, Belgium) with an electrical sealer (PFS-400, Plastic Film Sealer). Preliminary tests (not reported for the sake of conciseness) showed that the packaging material is resistant to high-pressure and supercritical CO2, and that these conditions determine neither mechanical damages on the material, nor variations in its barrier properties. Two carrot slices were packaged in each bag. The modified atmosphere was introduced into the packages through a gas mixer (MAP Mix 9001 ME, PBI Dansensor, Segate, Milano, Italy). The sealed packages with the product had a mass of 7.2 ± 0.3 g. To exclude gas leakage or water infiltration during the experiments, the mass and volume of the packages were measured with a scale (PS 6000.R2, Radwag, Radom, Poland) before and after treatment (data not shown). Packages with mass or volume change >5% (unlikely event) were considered damaged and discarded. The gas composition inside the packages was measured with a gas analyzer (Oxybaby M+i O2/CO2, WITT-Gastechnik GmbH, Witten, Germany). Gas composition was measured before the processing for control samples to ensure a correct gas composition inside the packages, and along the storage to assess the holding of the gas composition inside the packages. Gas composition details are provided in the Supplementary Material.

High Pressure Equipment
The high-pressure treatment was carried out in a multi-batch apparatus ( Figure S1). Two steel vessels (R-1 and R-2) with an internal volume of 320 mL each produced replicates under identical process conditions. Vessels are manually loaded at the same time. Each vessel was connected to the equipment piping through an on-off valve (V-1 and V-2, respectively). The vessels R-1 and R-2 were placed in a thermostatic bath to

MAP Packaging
MAP packages with an internal chamber of 10 cm × 10 cm and a volume of 100 ± 10 mL were produced from a high gas barrier multilayer (PA/EVOH/PA/PE) film (Euralpack, Schoten, Belgium) with an electrical sealer (PFS-400, Plastic Film Sealer). Preliminary tests (not reported for the sake of conciseness) showed that the packaging material is resistant to high-pressure and supercritical CO 2 , and that these conditions determine neither mechanical damages on the material, nor variations in its barrier properties. Two carrot slices were packaged in each bag. The modified atmosphere was introduced into the packages through a gas mixer (MAP Mix 9001 ME, PBI Dansensor, Segate, Milano, Italy). The sealed packages with the product had a mass of 7.2 ± 0.3 g. To exclude gas leakage or water infiltration during the experiments, the mass and volume of the packages were measured with a scale (PS 6000.R2, Radwag, Radom, Poland) before and after treatment (data not shown). Packages with mass or volume change >5% (unlikely event) were considered damaged and discarded. The gas composition inside the packages was measured with a gas analyzer (Oxybaby M+i O 2 /CO 2 , WITT-Gastechnik GmbH, Witten, Germany). Gas composition was measured before the processing for control samples to ensure a correct gas composition inside the packages, and along the storage to assess the holding of the gas composition inside the packages. Gas composition details are provided in the Supplementary Material.

High Pressure Equipment
The high-pressure treatment was carried out in a multi-batch apparatus ( Figure S1). Two steel vessels (R-1 and R-2) with an internal volume of 320 mL each produced replicates under identical process conditions. Vessels are manually loaded at the same time. Each vessel was connected to the equipment piping through an on-off valve (V-1 and V-2, respectively). The vessels R-1 and R-2 were placed in a thermostatic bath to control their temperature. Preheated water was pumped into the vessel by a high-pressure pump (P-1; LDC1, Lewa, Tokyo, Japan). The pressurization was achieved at 4 MPa/min with a water flow rate of 100 mL/min. The system was depressurized at the end of the process by acting on a micrometric valve (V-3; model 2S-4L-N-SS, Rotarex, Brescia, Italy).
The process temperature was controlled by the thermostatic bath heater (TBH; M408-BC, MPM Instruments s.r.l., Milano, Italy) through the manipulation of a thermal fluid flow inside a copper coil immersed in the thermostatic bath. Pressure was controlled by a pressure controller (PC; ATR241, Pixsys s.r.l., Venice, Italy), which acted on the pump motor. The apparatus was designed to operate at temperatures between 25 and 50 • C, and pressures between 4 and 20 MPa.

Image Acquisition
Unpackaged carrot images were collected with an internally developed acquisition apparatus [22] consisting of a box, which shades environmental light, equipped with an illumination system of 4 LED lamps (12V, 4000K, V-Tac, Sofia, Bulgaria) on the four sides of the box to provide uniform illumination. Images were taken with a commercial digital camera (Lumix TZ57 16 MP, Panasonic, Osaka, Japan) fixed on a horizontal support 50 cm above the bottom of the acquisition apparatus. The camera was set with a diaphragm aperture of f/6.3, an exposure time of 1/80 s, and an ISO sensibility of 200. The scene in the collected images contains a food sample (i.e., carrot slice), a white PFA standard reflectance reference (Ocean Optics Inc., Dunedin, FL, USA), a black background, and a dimensional indicator. A typical image used in this analysis is reported in Figure 2.
The process temperature was controlled by the thermostatic bath heater (TBH; M408-BC, MPM Instruments s.r.l., Milano, Italy) through the manipulation of a thermal fluid flow inside a copper coil immersed in the thermostatic bath. Pressure was controlled by a pressure controller (PC; ATR241, Pixsys s.r.l., Venice, Italy), which acted on the pump motor. The apparatus was designed to operate at temperatures between 25 and 50 °C, and pressures between 4 and 20 MPa.

Image Acquisition
Unpackaged carrot images were collected with an internally developed acquisition apparatus [22] consisting of a box, which shades environmental light, equipped with an illumination system of 4 LED lamps (12V, 4000K, V-Tac, Sofia, Bulgaria) on the four sides of the box to provide uniform illumination. Images were taken with a commercial digital camera (Lumix TZ57 16 MP, Panasonic, Osaka, Japan) fixed on a horizontal support 50 cm above the bottom of the acquisition apparatus. The camera was set with a diaphragm aperture of f/6.3, an exposure time of 1/80 s, and an ISO sensibility of 200. The scene in the collected images contains a food sample (i.e., carrot slice), a white PFA standard reflectance reference (Ocean Optics Inc., Dunedin, FL, USA), a black background, and a dimensional indicator. A typical image used in this analysis is reported in Figure 2.
The collected RGB (red, green, blue) images, , had a dimension of 3456 pixels × 4608 pixels × 3 color channels, where 1 pixel has the linear dimension of 33 ± 1 μm.
Images were taken separately for each carrot slice in the package: (i) at fresh condition 5 min prior the packaging and (ii) at treated condition immediately after unpackaging. Accordingly, = 168•2 = 336 fresh, = 336 treated, and = 36 control images were analyzed in this study. Hence, 708 carrot images were analyzed.

Image Analysis
Images were analyzed with an in-house developed software implemented in MATLAB ® 2019b (Matworks, Natick, MA, USA) through Image Processing Toolbox (Matworks, Natick, MA, USA), and PLS Toolbox 8.7 (Eigenvector Research Inc., Wenatchee, WA, USA). A three-step procedure was followed: (i) image standardization, The collected RGB (red, green, blue) images, , had a dimension of 3456 pixels × 4608 pixels × 3 color channels, where 1 pixel has the linear dimension of 33 ± 1 µm.
Images were taken separately for each carrot slice in the package: (i) at fresh condition 5 min prior the packaging and (ii) at treated condition immediately after unpackaging. Accordingly, O = 168 × 2 = 336 fresh, O = 336 treated, and O c = 36 control images were analyzed in this study. Hence, 708 carrot images were analyzed.

Image Analysis
Images were analyzed with an in-house developed software implemented in MATLAB ® 2019b (Matworks, Natick, MA, USA) through Image Processing Toolbox (Matworks, Natick, MA, USA), and PLS Toolbox 8.7 (Eigenvector Research Inc., Wenatchee, WA, USA). A three-step procedure was followed: (i) image standardization, (ii) extraction of the region of interest (i.e., carrot sample), and (iii) extraction of features on color and its distribution over the product surface. Images were initially adjusted [35,36] to compensate for all possible variations caused by lighting conditions and to reconstruct the real color of the image by means of the white and black references that were present in the image scene.
The region of the original image containing the carrot sample, named the region of interest, was segmented through k-means clustering [37]. The color information was extracted from the region of interest. Then, the largest rectangular portion of the region of interest was identified and used to extract features on the color spatial distribution (which defines the structural characteristics of products).
The features of the color and its spatial distribution were extracted from each image. M C = 12 color features were extracted as 4 statistical indices of the light intensity distribution of each color channel of an RGB image [22]: mean, standard deviation, skewness, and kurtosis. The features describing the spatial distribution of color were extracted through two different methods: (i) a wavelet texture analysis and (ii) a gray-level cooccurrence matrix (GLCM) from the gray-scale version Wavelet texture analysis [29,36,38,39]

Multivariate Hypothesis Testing for Appearance Characterization
Principal component analysis (PCA) [42] was used to test in a multivariate fashion if the treated or stored product preserves a fresh-like visual appearance through multivariate statistical control charts. PCA is a multivariate statistical method that compresses the information in the multivariate dataset of the M = 2576 color-structural features by projecting it into a reduced space of the A orthogonal (i.e., independent) principal components (PCs), which define the direction of maximum variability of the data.
Once calibrated on fresh samples, PCA provides two statistical indices to judge the appearance of new analyzed product samples: (i) the squared prediction error Q and (ii) Hotelling's T 2 . Q represents the mismatch between the new analyzed samples and the fresh reference, while T 2 describes the deviation of each sample from the average color-structural features on which the PCA model is built. Confidence limits for both the residual (Q lim ) and Hotelling's T 2 (T 2 lim ) can be used to verify whether a new observation could be adequately described by the model and is close to the average conditions (i.e., its appearance is similar to the one of the fresh product) or if it does not conform to the standard conditions in terms of the multivariate correlation structure among features of color and/or its spatial distribution. The 95% confidence limits for both Q and T 2 are defined as and where µ and ν are the mean and standard deviation of the Q residuals for the calibration dataset, α is the confidence limit (usually set at 95%), χ 2 is a chi-squared distribution, and F represents a F-distribution. Details on PCA are reported in the Appendix B.
In this study, a PCA model with A = 7 PCs was calibrated on features of color and its spatial distribution from fresh carrots X f . Accordingly, the multivariate control chart T 2 vs. Q was used to conduct a hypothesis testing on the carrot appearance. Specifically, T 2 o and Q o (the statistics for a treated sample o from X t ) were used to test the null hypothesis that the visual appearance of the treated carrot conforms to the one of fresh products with 95% confidence. If both T 2 o < T 2 lim and Q o < Q lim , we do not reject the null hypothesis that the treated sample conforms to the fresh vegetables with a confidence of 95%, meaning that the treated carrot cannot be distinguished from fresh products by the artificial vision system. Hence, it is likely that the human eye cannot perceive any differences between fresh and treated product. Conversely, if one of the two statistics T 2 o and Q o exceeds the respective confidence limits, the null hypothesis is rejected, meaning that the appearance of the treated sample does not conform to that of the fresh food with 95% confidence. In this case, the treated sample is subject to a machine detectable change in color and/or its spatial distribution and has a visual appearance that differs from 95% of the analyzed fresh carrots. Furthermore, the control chart indicates the extent of color and structural changes based on how large the deviations out of the confidence limits are. In fact, since T 2 o and Q o quantify how well a sample conforms to the model calibrated on fresh samples, large values of these indices relate to large appearance changes.

Microbiological Analyses
The Escherichia coli (Migula) Castellani and Chalmers (ATCC 25922) strain was used to inoculate samples to test the ScCO 2 -MAP performance to inactivate fecal contaminants, following the method described previously [43].
Some additional samples were inoculated with 40 µL of E. coli suspension (20 µL per each carrot side), to reach an inoculation level of 10 8 CFU/g [44]. After growing overnight, bacteria were first centrifuged (10 min-4696× g) and then pellets were resuspended in phosphate buffer saline (PBS, Oxoid, UK) before inoculation. The E. coli suspension was uniformly spread in several droplets (3-5 µL) over the carrot surface through a pipet. After inoculation, the samples were dried under a laminar flow hood for 30 min before packaging. Samples which were not inoculated were also analyzed to enumerate the natural microorganisms (i.e., total mesophilic bacteria, yeasts and molds, and E. coli) which were present.
For the enumeration, samples were diluted at a ratio of 1:10 in phosphate buffer saline (PBS, Oxoid, UK) and stomached for 1 min in sterile conditions. Appropriate serial dilutions were inoculated into plate count agar (PCA, Sacco, IT), rose bengal agar (RB, Sacco, IT), and chromogenic C-EC II agar (Sacco, IT), and incubated under one of the following conditions: 30 • C for 3 days, 22 • C for 5 days, or 37 • C for 18 h, to investigate the growth of total mesophiles, yeasts and molds, and E. coli, respectively.
The level of inactivation was determined by evaluating the log(N/N 0 ), where N 0 (CFU/g) is the number of colony forming units present in the untreated sample and N (CFU/g) is the number of survivors after the treatment.

Statistical Tests of the Microbiological Analyses
Analysis of variance (ANOVA) [30] was used for the statistical interpretation of the microbial analysis to assess the effect of pressure, temperature, and treatment time on the microbial population. Effects were considered significant if the ANOVA p-value was p ≤ 0.05, while effects were not considered statistically significant otherwise.

Results and Discussion
This Section shows the results on (i) the selection of the best MAP for untreated fresh carrots; (ii) the identification of the ScCO 2 -MAP operating conditions preserving the fresh appearance of carrots after treatment and evaluation of the appearance change during storage; and (iii) the verification of the microbiological safety for carrots processed at optimal ScCO 2 -MAP operating conditions only.

Identification of the Atmosphere Composition Preserving Fresh-like Product Appearance
The effect of the modified atmosphere composition on the product appearance during storage was preliminarily evaluated to identify which was the optimal MAP composition between MAP1, MAP2, and MAP3 for untreated carrots (i.e., stored samples without being processed with ScCO 2 -MAP) and to validate the multivariate method. For this purpose, multivariate hypothesis testing was performed on all the untreated control samples by projecting them into the PCA model built on fresh carrot samples X f (Section 2.4.3). The control chart (Figure 3a) reports Hotelling's T 2 on the x-axis and the residual Q on the y-axis, while the dashed lines are the 95% confidence limits T 2 lim and Q lim . The pairs (T 2 o , Q o ) related to all X f observations (for o = 1, 2, . . . , O) are plotted in the control chart as gray squares and serve as a reference of fresh conditions, while untreated control samples are plotted as colored shapes. The null hypothesis was rejected for all samples packaged in MAP2 and MAP3 (90% CO 2 with 10% O 2 and air, respectively) independently of the storage time, because the blue triangles and the black circles are located outside the 95% confidence limits. This indicates that the appearance of the untreated product packaged in the presence of O 2 (MAP2 and MAP3) did not conform to that of the fresh carrots and appearance changes occurred during storage. An example of what can be visually perceived for carrots packaged and stored in MAP2 and MAP3 after 3 days of storage is reported in Figure 3c,d. Conversely, the null hypothesis was not rejected for the untreated samples packaged in MAP1 (100% CO 2 ) up to 7 days of storage (green diamonds in Figure 3a). This means that samples packaged in MAP1 are not distinguishable from fresh carrots with 95% confidence after 7 days of storage. Instead, the null hypothesis was rejected for 14-day samples, because both replicates were located outside at least one of the statistical limits. Despite appearance changes being detected by the artificial vision system, these samples were located close to the statistical limits and to the 5% of non-standard fresh samples out of the 95% confidence limits. For this reason, these samples can still be considered hardly distinguishable from fresh carrots by human perception; hence, they can be considered to have a good appearance quality, and would likely be accepted by consumers. An example of what can be visually perceived for carrots packaged in MAP1 after 14 days of storage is reported in Figure 3b.
The analysis of the atmosphere composition for control samples confirmed that carrots were packaged with the correct MAP composition, and that this composition was correctly preserved in the packages up to 14 days (Supplementary Material, Figures S2 and S3). Only small variations in the O 2 concentration were found due to instrumental measurement error and tolerances in the MAP filling.
Our findings support the fact that the modified atmosphere composition has an effect on the appearance of carrots during storage [45,46], which are usually subject to surface whitening due to dehydration and lignification [47]. Furthermore, as previously observed [48], the results of MAP2 and MAP3 support the fact that low O 2 (~10%) has a negative effect on product quality, probably due to cellular respiration and enzymatic reactions. However, further tests should be performed to support these hypotheses and to determine the cause of appearance changes in MAP with low O 2 concentration. Our findings support the fact that the modified atmosphere composition has an effect on the appearance of carrots during storage [45,46], which are usually subject to surface whitening due to dehydration and lignification [47]. Furthermore, as previously observed [48], the results of MAP2 and MAP3 support the fact that low O2 (~10%) has a negative effect on product quality, probably due to cellular respiration and enzymatic reactions. However, further tests should be performed to support these hypotheses and to determine the cause of appearance changes in MAP with low O2 concentration.

Identification of the Processing Condition Preserving the Fresh-Like Product Appearance
The optimization of the ScCO2-MAP preservation process is intended to identify the treatment conditions that minimally affect the visual appearance of the products during storage. The analysis was performed for all treated carrots packaged with three MAPs (MAP1, MAP2, and MAP3) from by projecting them into the PCA model built on fresh carrot samples (Section 2.4.3). The results confirmed that 100% CO2 MAP (MAP1) induces a minimal effect on the visual appearance after the process and during storage, similar to what was presented in Section 3.1. For this reason, only the results obtained for MAP1 are presented and discussed. Results on MAP2 and MAP3 can be found in the Supplementary Material (Figures S5-S7).
The effect of the process conditions was analyzed after treatment (0 days of storage) to identify the conditions that preserve a fresh appearance. The control chart built for these samples is reported in Figure 4, where fresh carrots from are plotted as gray squares, and the dashed lines are the 95% confidence limits and Q . The null hypothesis

Identification of the Processing Condition Preserving the Fresh-like Product Appearance
The optimization of the ScCO 2 -MAP preservation process is intended to identify the treatment conditions that minimally affect the visual appearance of the products during storage. The analysis was performed for all treated carrots packaged with three MAPs (MAP1, MAP2, and MAP3) from X t by projecting them into the PCA model built on fresh carrot samples X f (Section 2.4.3). The results confirmed that 100% CO 2 MAP (MAP1) induces a minimal effect on the visual appearance after the process and during storage, similar to what was presented in Section 3.1. For this reason, only the results obtained for MAP1 are presented and discussed. Results on MAP2 and MAP3 can be found in the Supplementary Material (Figures S5-S7).
The effect of the process conditions was analyzed after treatment (0 days of storage) to identify the conditions that preserve a fresh appearance. The control chart built for these samples is reported in Figure 4, where fresh carrots from X f are plotted as gray squares, and the dashed lines are the 95% confidence limits T 2 lim and Q lim . The null hypothesis was rejected for all samples treated at TC5 (40 • C, 12 MPa, 45 min; black diamonds), which were located outside the statistical limits T 2 lim and Q lim , and were the most distant samples from the 95% confidence limits. Consequently, TC5 was the most severe treatment and caused a change in appearance with respect to fresh carrots. Similarly, TC3 (25 • C, 6 MPa, 45 min; green crosses) and TC4 (25 • C, 12 MPa, 15 min; cyan squares) produced appearance changes that oftentimes made the samples distinguishable from fresh carrots; in fact, for both TC3 and TC4 the null hypothesis was rejected for 50% of the samples (2 over 4 samples), which were located outside the statistical limits. On the contrary, 66.6% of the samples treated at TC1 (32.5 • C, 9 MPa, 30 min; blue triangles) remained similar to the fresh carrots after treatment. In fact, the null hypothesis was not rejected in 66.6% of the samples (8 over 12 samples), because they were located inside the statistical limits T 2 lim and Q lim . However, a large variability in the appearance of these samples was observed. In fact, three samples treated at TC1 showed a very high appearance change, compatible with a more severe treatment condition (that is, TC5). Samples treated at TC2 (40 • C, 6 MPa, 15 min; red dots) could not be distinguished from fresh carrots by the artificial vision system (with 95% confidence), because the null hypothesis was not rejected for all the samples (4 over 4 samples). Accordingly, TC1 and TC2 are the most promising conditions for preserving the fresh appearance of RTE carrots after ScCO 2 -MAP treatment. treatment and caused a change in appearance with respect to fresh carrots. Similar (25 °C, 6 MPa, 45 min; green crosses) and TC4 (25 °C, 12 MPa, 15 min; cyan s produced appearance changes that oftentimes made the samples distinguishab fresh carrots; in fact, for both TC3 and TC4 the null hypothesis was rejected for 50% samples (2 over 4 samples), which were located outside the statistical limits. contrary, 66.6% of the samples treated at TC1 (32.5 °C, 9 MPa, 30 min; blue tri remained similar to the fresh carrots after treatment. In fact, the null hypothesis w rejected in 66.6% of the samples (8 over 12 samples), because they were located ins statistical limits and Q . However, a large variability in the appearance o samples was observed. In fact, three samples treated at TC1 showed a ver appearance change, compatible with a more severe treatment condition (that is Samples treated at TC2 (40 °C, 6 MPa, 15 min; red dots) could not be distinguishe fresh carrots by the artificial vision system (with 95% confidence), because t hypothesis was not rejected for all the samples (4 over 4 samples). Accordingly, T TC2 are the most promising conditions for preserving the fresh appearance of RTE after ScCO2-MAP treatment. These results show that very high pressure (12 MPa) and long treatment ti min) have a stronger impact on carrot appearance than high temperature (40 ° influence of longer treatment times on carrots' color has previously been obser Ferrentino et al. [15]. According to the above-mentioned result, the control chart for samples treated during 14 days of storage was considered to understand the stability of the appe over time (Figure 5). At 3 days of storage (blue crosses), the null hypothesis w rejected in 50% of samples (2 over 4 samples,); however, for one of the samples t hypothesis was rejected even if it was located very close to the statistical limits, m that this sample was subject only to marginal appearance changes. At 7 days of (red triangles), the null hypothesis was rejected for 75% of samples (3 over 4 sa indicating that the majority of the samples showed some variation in color and its These results show that very high pressure (12 MPa) and long treatment time (45 min) have a stronger impact on carrot appearance than high temperature (40 • C). The influence of longer treatment times on carrots' color has previously been observed by Ferrentino et al. [15].
According to the above-mentioned result, the control chart for samples treated at TC2 during 14 days of storage was considered to understand the stability of the appearance over time (Figure 5). At 3 days of storage (blue crosses), the null hypothesis was not rejected in 50% of samples (2 over 4 samples,); however, for one of the samples the null hypothesis was rejected even if it was located very close to the statistical limits, meaning that this sample was subject only to marginal appearance changes. At 7 days of storage (red triangles), the null hypothesis was rejected for 75% of samples (3 over 4 samples), indicating that the majority of the samples showed some variation in color and its spatial distribution. Moreover, the null hypothesis was rejected for all samples (4 over 4 samples) at 14 days of storage (black diamonds), which were located outside the statistical limits. In summary, carrots treated at 40 • C, 6 MPa, and 15 min, and stored at 4 • C are often not distinguishable from fresh carrots in the initial days of storage, while after 7-14 days the appearance change from the fresh product of these samples is evident. An example of the appearance change that can be visually perceived during storage for carrots treated at TC2 is reported in Figure 6a-d. However, the extent of these appearance changes (i.e., the distance from the statistical limits T 2 lim and Q lim ) was often small compared to the appearance changes produced by heavier treatment conditions (i.e., higher pressures) as shown in Figure 3. For example, the treatment at 40 • C, 12 MPa, and 45 min (TC5) alone generally produced larger appearance changes compared to carrots stored for 14 days after being processed at 40 • C, 6 MPa, and 15 min (TC2). This result indicates that high pressure (12 MPa) produces a larger effect on appearance than the combined effect of lower pressure (6 MPa) and 14 days of storage.
Samples treated at TC1 (32.5 • C, 9 MPa, 30 min) with a storage time ranging from 3 to 14 days were distinguishable from fresh carrots (Supplementary Material, Figure S4), because the null hypothesis was rejected for 91.7% of samples (11 over 12 samples). In this case, even if the treatment itself produced moderate appearance changes, the higher pressure and longer treatment time produced modifications in the vegetable matrix that caused a larger appearance change during storage. However, additional studies should be performed to understand which factor among temperature, pressure, and treatment time mostly affect the appearance stability during storage.
The analysis of the atmosphere composition (Supplementary Material, Figures S2 and S3) confirmed that the atmosphere composition defined in the experimental plan was maintained in the above-mentioned conditions up to 14 days of storage.
Foods 2021, 10, x FOR PEER REVIEW distribution. Moreover, the null hypothesis was rejected for all samples (4 over 4 sa at 14 days of storage (black diamonds), which were located outside the statistical In summary, carrots treated at 40 °C, 6 MPa, and 15 min, and stored at 4 °C are of distinguishable from fresh carrots in the initial days of storage, while after 7-14 d appearance change from the fresh product of these samples is evident. An exampl appearance change that can be visually perceived during storage for carrots treated is reported in Figure 6a-d. However, the extent of these appearance changes (i distance from the statistical limits and Q ) was often small compared appearance changes produced by heavier treatment conditions (i.e., higher pressu shown in Figure 3. For example, the treatment at 40 °C, 12 MPa, and 45 min (TC5 generally produced larger appearance changes compared to carrots stored for 1 after being processed at 40 °C, 6 MPa, and 15 min (TC2). This result indicates th pressure (12 MPa) produces a larger effect on appearance than the combined e lower pressure (6 MPa) and 14 days of storage.

Microbial Analyses
Microbial analysis was performed only under the processing conditions identified by the artificial vision system (in Section 3.2) that ensure the preservation of the fresh appearance of RTE carrots. Accordingly, samples packaged in MAP1 and treated at TC1 (32.5 °C, 9 MPa, 30 min) and TC2 (40 °C, 6 MPa, 15 min) were tested for microbial safety, although TC1 did not provide a good preservation of product appearance during storage, while TC2 did. Nevertheless, the former was also analyzed to compare the inactivation performance of CO2 at both supercritical (TC1) and subcritical conditions (TC2).
The detected natural flora of untreated fresh-cut carrots was composed of 5.44 ± 0.20 log (CFU/g) mesophilic microorganisms and 4.90 ± 0.22 log (CFU/g) yeasts and molds, in accordance with other analyses performed on fresh-cut carrots [11]. No E. coli colonies were detected within the natural microbiota on the surface of the carrots.
The microbial population of fresh-cut carrots during shelf life in terms of mesophilic microorganisms is reported in Figure 7a. Control samples packaged in MAP1 and MAP3 exhibited a steady growth in the population of mesophilic microorganisms during shelf life, and after 1 week they both reached >6 log (CFU/g) for mesophilic microorganisms, usually considered a threshold value by the food industry. For untreated control samples, ANOVA indicates that the atmosphere composition did not have an impact on the microbial population during shelf life ( > 0.05), meaning that CO2 itself is not able to inhibit the growth of total mesophilic bacteria, nor does the presence of oxygen promote microbial proliferation. All the analyzed treated samples exhibited an initial inactivation >1 log (CFU/g): specifically, TC1 induced 1.2 log (CFU/g), while a higher inactivation of 2.9 log (CFU/g) was achieved with TC2. Only the carrots treated at TC2 were able to maintain the total microbial load <6 log (CFU/g) up to 14 days, while at less than 14 days of storage they were able to induce a substantial growth after processing at TC1. In these cases, additional tests would be required to accurately determine the actual shelf life of

Microbial Analyses
Microbial analysis was performed only under the processing conditions identified by the artificial vision system (in Section 3.2) that ensure the preservation of the fresh appearance of RTE carrots. Accordingly, samples packaged in MAP1 and treated at TC1 (32.5 • C, 9 MPa, 30 min) and TC2 (40 • C, 6 MPa, 15 min) were tested for microbial safety, although TC1 did not provide a good preservation of product appearance during storage, while TC2 did. Nevertheless, the former was also analyzed to compare the inactivation performance of CO 2 at both supercritical (TC1) and subcritical conditions (TC2).
The detected natural flora of untreated fresh-cut carrots was composed of 5.44 ± 0.20 log (CFU/g) mesophilic microorganisms and 4.90 ± 0.22 log (CFU/g) yeasts and molds, in accordance with other analyses performed on fresh-cut carrots [11]. No E. coli colonies were detected within the natural microbiota on the surface of the carrots.
The microbial population of fresh-cut carrots during shelf life in terms of mesophilic microorganisms is reported in Figure 7a. Control samples packaged in MAP1 and MAP3 exhibited a steady growth in the population of mesophilic microorganisms during shelf life, and after 1 week they both reached >6 log (CFU/g) for mesophilic microorganisms, usually considered a threshold value by the food industry. For untreated control samples, ANOVA indicates that the atmosphere composition did not have an impact on the microbial population during shelf life (p > 0.05), meaning that CO 2 itself is not able to inhibit the growth of total mesophilic bacteria, nor does the presence of oxygen promote microbial proliferation. All the analyzed treated samples exhibited an initial inactivation >1 log (CFU/g): specifically, TC1 induced 1.2 log (CFU/g), while a higher inactivation of 2.9 log (CFU/g) was achieved with TC2. Only the carrots treated at TC2 were able to maintain the total microbial load <6 log (CFU/g) up to 14 days, while at less than 14 days of storage they were able to induce a substantial growth after processing at TC1. In these cases, additional tests would be required to accurately determine the actual shelf life of the products. ANOVA indicates that TC1 and TC2 provided a significantly different inactivation of the microbial population (p < 0.001) and showed a significantly different microbial growth profile during shelf life (p < 0.001). Accordingly, TC1 provides a larger inactivation performance and a longer stability over storage time.
instead, showed a significant reduction in the population of yeasts and molds identified with ANOVA ( < 0.001), and the population remained stable up to 14 days. This indicates that yeasts and molds are more sensitive to the CO2 than mesophilic bacteria as previously observed [49], and the absence of oxygen reduces the growth of yeast and mold populations. The products treated at TC2 showed the highest inactivation for yeasts and molds, which resulted in them being below the detectability limits of the technique (<100 CFU/g) during the whole shelf life. However, a certain degree of inactivation was also achieved with treatment TC1, which inhibited microorganisms' growth during storage. TC1 and TC2 showed statistically different inactivation performances for the entire shelf life according to ANOVA ( < 0.001).
According to these results, the combination of high temperature (40 °C) and pressurized CO2 determines a more efficient inactivation than supercritical CO2 at low temperature (32.5 °C). The inactivation with ScCO2-MAP was also tested on inoculated E. coli. A 1.98 ± 0.48 log (CFU/g) and 2.27 ± 0.32 log (CFU/g) inactivation were achieved with treatments TC1 and TC2, respectively. Furthermore, no significant differences in inactivation performance (ANOVA > 0.05) were found between TC1 and TC2, indicating that the achieved E. coli inactivation is independent of the specific treatment condition, indicating that neither higher temperature (40 °C) nor supercritical CO2 provide higher performance in E. coli inactivation. This finding indicates that the ScCO2-MAP preservation process provides different inactivation performances for different microbial species (i.e., mesophilic microorganisms, yeasts, molds, and E. coli) as previously seen with the conventional ScCO2 treatment [50]. However, the process induces a 99% reduction in the initial E. coli population, thus increasing the product safety against fecal contamination. The microbial population of carrots during shelf life in terms of yeasts and molds is reported in Figure 7b. Control samples packaged in air (MAP3) showed a steadily growing profile in yeasts and molds population, and the total population of yeasts and molds was >6 log (CFU/g) at 14 days of shelf life. Control samples packaged in MAP1, instead, showed a significant reduction in the population of yeasts and molds identified with ANOVA (p < 0.001), and the population remained stable up to 14 days. This indicates that yeasts and molds are more sensitive to the CO 2 than mesophilic bacteria as previously observed [49], and the absence of oxygen reduces the growth of yeast and mold populations. The products treated at TC2 showed the highest inactivation for yeasts and molds, which resulted in them being below the detectability limits of the technique (<100 CFU/g) during the whole shelf life. However, a certain degree of inactivation was also achieved with treatment TC1, which inhibited microorganisms' growth during storage. TC1 and TC2 showed statistically different inactivation performances for the entire shelf life according to ANOVA (p < 0.001).
According to these results, the combination of high temperature (40 • C) and pressurized CO 2 determines a more efficient inactivation than supercritical CO 2 at low temperature (32.5 • C).
The inactivation with ScCO 2 -MAP was also tested on inoculated E. coli. A 1.98 ± 0.48 log (CFU/g) and 2.27 ± 0.32 log (CFU/g) inactivation were achieved with treatments TC1 and TC2, respectively. Furthermore, no significant differences in inactivation performance (ANOVA p > 0.05) were found between TC1 and TC2, indicating that the achieved E. coli inactivation is independent of the specific treatment condition, indicating that neither higher temperature (40 • C) nor supercritical CO 2 provide higher performance in E. coli inactivation. This finding indicates that the ScCO 2 -MAP preservation process provides different inactivation performances for different microbial species (i.e., mesophilic microorganisms, yeasts, molds, and E. coli) as previously seen with the conventional ScCO 2 treatment [50]. However, the process induces a 99% reduction in the initial E. coli population, thus increasing the product safety against fecal contamination.

Conclusions
In this study, we optimized the processing conditions of an innovative high-pressure CO 2 preservation process to reduce the appearance change during the maintenance of cut carrots at 4 • C using an artificial vision system. The developed methodology was able to characterize the appearance changes in terms of color and superficial structure and provided an easy and fast identification of the products with a fresh-like appearance, which will be easily accepted by consumers.
A 100% CO 2 modified atmosphere packaging alone was able to the preserve the visual appearance of ready-to-eat carrots, but it does not guarantee microbial stability. Processing in 100% CO 2 MAP at 40 • C, 6 MPa, and 15 min maintained the fresh appearance of carrots up to 7 days of storage, providing a 2.9 log (CFU/g) inactivation and a microbiologic stability over 14 days of storage at 4 • C. The process was able to inactivate 2 log (CFU/g) of inoculated E. coli on the surface. However, in the carrots after 7-14 days, perceivable appearance changes were detected; additional studies should be performed to demonstrate the acceptability of the product through panel/consumer testing. Moreover, tests on other sensorial and chemical characteristics, such as flavor, texture, humidity, and nutrient content, should be performed in future studies.
The developed artificial vision system has the potential to be applied to any food treatment involving changes in the appearance of products. Furthermore, future investigations should consider the development of an artificial vision system capable of directly analyzing the packaged product, which can be useful as routine qualitative tests. Funding: This research was funded by "Fondazione Cassa di Risparmio di Padova e Rovigo" ("CARIPARO Visiting Programme 2018"); AREA Science Park for the Innovation Voucher by the Interreg ADRION-ADRIATIC-IONIAN Programme "OISAIR" Proof of Concept Call; the Department of Industrial Engineering (University of Padova) ("Progetto Strategico di Dipartimento" (CoSID) 2018); and "Ministero delle Politiche Agricole Alimentari e Forestali" (MIPAAF) under the joint call 2019 SUSFOOD CORE Organic "Towards sustainable and organic food systems" within the project ID 46 HO-FOOD "Innovative High pressure process to Increase the preservation of ready-to-eat Organic FOOD" (CUP C95F20000930001).

Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author. The data are not publicly available due to privacy reasons.

Acknowledgments:
The authors thank Massimo Vetralla for the support during the experimental campaign. where t o is the score vector of the observation o calculated by projecting x o into the PCA model. Similarly, the sample deviation from the average observation is described by means of Hotelling's T 2 statistics:

Conflicts of Interest
where Λ −1 is a [A × A] diagonal matrix containing the inverse eigenvalues associated to the A PCs.