Article

Exploration of Convective and Infrared Drying Effect on Image

Texture Parameters of ‘Mejhoul’ and ‘Boufeggous’ Date Palm
Fruit Using Machine Learning Models
Younes Noutfia and Ewa Ropelewska *

Fruit and Vegetable Storage and Processing Department, The National Institute of Horticultural Research,
Konstytucji 3 Maja 1/3, 96-100 Skierniewice, Poland; [email protected]
* Correspondence: [email protected]

Abstract: Date palm (Phoenix dactylifera L.) fruit samples belonging to the ‘Mejhoul’ and ‘Boufe-
ggous’ cultivars were harvested at the Tamar stage and used in our experiments. Before scanning,
date samples were dried using convective drying at 60 °C and infrared drying at 60 °C with a fre-
quency of 50 Hz, and then they were scanned. The scanning trials were performed for two hundred
date palm fruit in fresh, convective-dried, and infrared-dried forms of each cultivar using a flatbed
scanner. The image-texture parameters of date fruit were extracted from images converted to indi-
vidual color channels in RGB, Lab, XYZ, and UVS color models. The models to classify fresh and
dried samples were developed based on selected image textures using machine learning algorithms
belonging to the groups of Bayes, Trees, Lazy, Functions, and Meta. For both the ‘Mejhoul’ and
‘Boufeggous’ cultivars, models built using Random Forest from the group of Trees turned out to be
accurate and successful. The average classification accuracy for fresh, convective-dried, and infra-
red-dried ‘Mejhoul’ reached 99.33%, whereas fresh, convective-dried, and infrared-dried samples
of ‘Boufeggous’ were distinguished with an average accuracy of 94.33%. In the case of both cultivars
Citation: Noutfia, Y.; Ropelewska, E. and each model, the higher correctness of discrimination was between fresh and infrared-dried
Exploration of Convective and samples, whereas the highest number of misclassified cases occurred between fresh and convective-
Infrared Drying Effect on Image dried fruit. Thus, the developed procedure may be considered an innovative approach to the non-
Texture Parameters of ‘Mejhoul’ and destructive assessment of drying impact on the external quality characteristics of date palm fruit.
‘Boufeggous’ Date Palm Fruit Using
Machine Learning Models. Foods
Keywords: Phoenix dactylifera L.; image features; flatbed scanner; artificial intelligence; classification
2024, 13, 1602.
models
https://doi.org/10.3390/
foods13111602

Academic Editors: Zhiming Guo and

Weiqing Min 1. Introduction
Received: 29 April 2024 Date palm (Phoenix dactylifera L.) is a very important fruit crop grown in many re-
Revised: 15 May 2024 gions of the world, especially in hot and dry areas, and it is considered a promising tree
Accepted: 18 May 2024 for the irrigated dry zones of developing nations [1,2]. For many nations, date fruit is an
Published: 21 May 2024 essential subsistence food [1] and an important component of a healthy diet due to its
functional properties as well as its high sugar content, flavonoids, anthocyanins, phenols,
vitamins, minerals, and carotenoids [3]. Hence, the nutritional composition of date fruit
Copyright: © 2024 by the authors. differs according to several factors and can range from 44 to 88% for total sugar, 2.3–5.6%
Licensee MDPI, Basel, Switzerland. for protein, 6.4–11.5% for fibers, 0.2–0.5% for fat, and 0.2–0.5% for oil. Thus, both edible
This article is an open access article parts of date fruit and seeds can be used in food products as a rich source of biochemical
distributed under the terms and and functional compounds such as antioxidants and fibers [1], especially at the Tamar
conditions of the Creative Commons
stage.
Attribution (CC BY) license
At this complete stage of maturity, the fruit usually has a reduced moisture content
(https://creativecommons.org/license
of less than 20–25%, and it is considered appropriate for consumption in many regions
s/by/4.0/).
and for storage under specific and mastered conditions [4,5]. However, some physiologi-
cal and physical disorders can occur at this stage, mainly related to skin separation and

Foods 2024, 13, 1602. https://doi.org/10.3390/foods13111602 www.mdpi.com/journal/foods

Foods 2024, 13, 1602 2 of 13

sugar spots. These visual defects negatively impact the visual quality of the fruit and its
market value [1]. Therefore, drying can provide shelf-life extension [6] and postpone the
perishability and degeneration of date fruit.
Drying as a pretreatment, in association with efficient storage availability, ensures
high-quality date fruit all year round in markets and in places where they are not pro-
duced [7]. By removing water, transport efficiency is increased by reducing the product
volume as well as packaging costs. Besides this logistical improvement, moisture level,
fruit sugars, and other biochemical compounds are more concentrated, allowing for more
protection against deterioration factors, microorganism proliferation and growth, unde-
sired chemical reactions, and high risk of deterioration during storage [8], leading to
preservation of quality properties and shelf life extension. Moreover, date fruit may be
dried using various methods, such as open sun drying, solar drying, hot air oven drying,
vacuum drying, microwave drying, microwave vacuum drying, drum drying, or freeze
drying [9].
The sun drying of date fruit has been the most common way throughout history.
However, the process is challenging because of the inability to control drying parameters
that may fluctuate drastically depending on climatic conditions. Furthermore, the sun
drying process requires a long time while date fruit is under uncontrolled conditions of
illumination and airflow, in addition to a possible source of contamination with insects,
sand particles, soil, and dust. Thus, sun drying is not considered an effective method to
apply for date fruit in the perspective of quality preservation and amelioration prior to
further postharvest operations (e.g., storage). Consequently, other drying methods are
more suitable for maintaining and improving the internal and external qualities of date
fruit: homogenous skin color; adequate skin hardness; and high preservation of nutrient
compounds [10–13]. One of the most common and effective methods is convective drying,
which ensures a reduction in moisture content and quality preservation [14]. Also, infra-
red drying can be more beneficial compared with conventional drying due to rapid pro-
cessing, shorter heating times, chemical compound preservation, and reduced risk of fla-
vor loss [15]. Artificial intelligence and image analysis can be considered innovative and
alternative approaches to assess the quality of date fruit before and after drying instead
of tedious, time-consuming, and destructive analyses [5]. In doing so, machine vision en-
ables the quantitative analysis of the quality of the qualitative criteria of the sample. Ma-
chine learning (ML) models can be successful in the classification of different samples of
date fruit based on selected image parameters using several algorithms and models (e.g.,
traditional ML algorithms) [16]. Furthermore, image textures and determined geometric
parameters using image analysis can be useful in the objective characterization of date
fruit [17].
Considering the previous elements, the objective of this study was to compare the
effect of convective and infrared drying on the quality of ‘Mejhoul’ and ‘Boufeggous’ date
palm fruit in terms of external appearance. The assessment of the fruit quality was per-
formed objectively and non-destructively using innovative artificial intelligence models
developed based on selected image texture parameters.

2. Materials and Methods

2.1. Materials
Two well-known commercial cultivars of Moroccan date palm fruit, ‘Mejhoul’ and
‘Boufeggous’, were used in these experiments. Fruit samples were harvested at the Tamar
stage in November 2023 in an orchard located at Henabou-Erfoud in southeastern Mo-
rocco (31°26′10″ N, 4°13′58″ W) and stored in cardboard boxes in a cold room at 2–4 °C
until experiment starting. Before and after drying experiments, two hundred fruits of
‘Mejhoul’ and ‘Boufeggous’ without any visual defects were subjected to imaging.
Foods 2024, 13, 1602 3 of 13

2.2. Date Fruit Drying

Date fruit belonging to ‘Mejhoul’ and ‘Boufeggous’ cultivars were dried using a
CONVECO semi-industrial dryer (CONVECO Sp. z o.o., Glinianka, Poland) equipped
with a power supply cabinet, control cabinet, heat recovery (recuperation) units, and a
ventilation duct system. For both convective and infrared drying, two technological repe-
titions were carried out, and the same temperature of 60 °C was used. For the infrared
drying, the waving range was at a frequency of 10 min at a radiation distance of 170 mm
with a radiation power of 50 Hz.

2.2.1. Convective Drying

For each technological repetition, date fruit samples were spread out in one layer into
trays placed at different levels of the drying chamber rack (Figure 1). Each tray contained
approximately 3–4 kg of ‘Mejhoul’ or ‘Boufeggous’ to provide a total amount of dried
material of approximately 25–26 kg per cultivar and per technological repetition. This pro-
cess was performed through hot air at a temperature of 60 °C, and the airflow side (right
or left) was changed every 10 min. This process was carried out until reaching the desired
weight loss and estimated water content (achieved in 240 min).

Figure 1. Illustration of convective drying of ‘Mejhoul’ and ‘Boufeggous’.

2.2.2. Infrared Drying

The surfaces of the date samples were heated using lamps (radiant heaters) located
above trays with date fruit, with a distance (between lamps and fruit samples) of 17 cm
(Figure 2). The infrared drying was performed at parameters of 60 °C and 50 Hz. The
whole process lasted 150 min, and the date fruit was side-changed in the middle of the
process so that the lower side of the fruit was on top and heated consequently by the
lamps. The changing side of airflow (right or left) was every 10 min. As for convective
drying, infrared drying was performed in two technological repetitions, and the airflow
side (right or left) was changed every 10 min
Foods 2024, 13, 1602 4 of 13

Figure 2. Illustration of infrared drying of ‘Mejhoul’.

2.3. Image Analysis

The fresh, convective-dried, and infrared-dried date fruit were scanned using the Ep-
son Perfection flatbed scanner (Epson, Suwa, Nagano, Japan) on a white background cov-
ered by a box and saved in TIFF format. Twenty fruits were in one image, and the scanning
was performed for two hundred date palm fruit samples of ‘Mejhoul’ and ‘Boufeggous’
in fresh, convective-dried, and infrared-dried forms. Before image processing, the white
background was changed to black to facilitate image segmentation and separation of fruit
from the background, and images were saved in BMP format. Image processing, including
image segmentation, ROI (region of interest) determination, and texture extraction, was
performed using MaZda 4.7 software (Łódź University of Technology, Institute of Elec-
tronics, Łódź, Poland) [18–20]. The date fruit images were converted to individual color
channels R, G, B, L, a, b, X, Y, Z, U, V, and S. Exemplary original color images and images
in selected color channels of fresh, convective-dried, and infrared-dried date fruit of
‘Mejhoul’ and ‘Boufeggous’ are presented respectively in Figures 3 and 4. The dried fruit
samples were of better quality in terms of appearance and characterized by shinier skin,
especially in the case of infrared-dried fruit. The image segmentation was performed
based on the brightness threshold, and the lighter date fruit samples were separated from
the black background. Each whole date fruit was treated as an ROI, and for each ROI, 2160
image textures were computed based on the run-length matrix, co-occurrence matrix, his-
togram, gradient map, and autoregressive model.

Figure 3. Exemplary original color images and images in selected color channels of fresh,
convective-dried (CD), and infrared-dried (ID) date fruit ‘Mejhoul’. B refers to blue, and
L refers to lightness.
Foods 2024, 13, 1602 5 of 13

Figure 4. Exemplary original color images and images in selected color channels of fresh,
convective-dried, and infrared-dried date fruit ‘Boufeggous’. B refers to blue, and L refers
to lightness.

2.4. Mean Comparison of Selected Image Textures

Graphs of mean values of selected image textures and analysis of mean comparison
were carried out using STATISTICA 3.1 (Dell Inc., Tulsa, OK, USA, StatSoft Polska, Kra-
ków, Poland) software. The normality of the distribution was checked using Shapiro–
Wilk, Lilliefors, and Kolmogorov–Smirnov tests. The means were compared using
Tukey’s test at a significance level of p < 0.05.

2.5. Machine Learning Models for Distinguishing Fresh, Convective, and Infrared-Dried
Date Fruit
The classification models were built based on selected image texture parameters of
fresh and dried date fruit using WEKA 3.9 machine learning software (Machine Learning
Group, University of Waikato, Hamilton, New Zealand) [21–23]. The image textures with
the highest discriminative power were selected using the Best First and the correlation-
based feature selection subset evaluator. The models for distinguishing fresh, convective,
and infrared-dried date fruit were built using a 10-fold cross-validation mode by dividing
the dataset into 10 parts and considering nine parts as the training sets and one part as the
test set. This process was repeated 10 times for different training/test sets, and the result
was the average of 10 estimates. The machine learning algorithms belonging to the groups
of Bayes, Trees, Lazy, Functions, and Meta were tested to select algorithms that provided
highly accurate results. The confusion matrices with accuracies for individual classes, av-
erage accuracies, and the values of True Positive Rate (TPR), False Positive Rate (FPR),
Precision, Recall, F-Measure, Matthews Correlation Coefficient (MCC), Receiver Operat-
ing Characteristic Area (ROC Area), and Precision–Recall Area (PRC Area) (Equations (1)–
(8)) [24–27] were determined.
(TP + TN)
Accuracy = (1)
TP + TN + FN + FP

TP
TPR = Recall = (2)
TP + FN

FP
FPR = (3)
FP + TN

TP
Precision = (4)
TP + FP

2TP
F − Measure = (5)
2TP + FP + FN
Foods 2024, 13, 1602 6 of 13

(TP ∗ TN − FP ∗ FN)
MCC =
(6)
√((TP + FP)(TP + FN)(TN + FP)(TN + FN))

ROC Area = Area Under TPR vs. FPR Curve (7)

PRC Area = Area Under Precision vs. Recall Curve (8)

TP: True Positive; TN: True Negative; FP: False Positive; FN: False Negative

3. Results and Discussion

3.1. ‘Mejhoul’: Discrimination between Fresh, Convective, and Infrared Dried Fruit Based on
Machine Learning Models and Selected Image Texture Parameters
The selected image texture parameters of fresh, convective-dried, and infrared-dried
‘Mejhoul’ date fruit samples were compared, and graphs of mean values are shown in
Table 1. In the case of RHMean, when R was color channel R (red) from the RGB color
space, and HMean was histogram’s mean, there was no significant difference (p > 0.05)
between fresh and convective-dried fruit samples, and these samples were included in
one homogenous group. For other analyzed image textures, such as GHMean, BHMean,
LHMean, aHMean, bHMean, XHMean, YHMean, and ZHMean (the first letter means the
color channel as follows: G (green) and B (blue) from the RGB color space; L (lightness
component from black to white), a (red or green), and b (yellow or blue) from the Lab color
space; and X (component with color information), Y (lightness), and Z (component with
color information) from the XYZ color space, fresh, convective and infrared samples were
significantly different.

Table 1. Mean values of selected image texture parameters of fresh, convective-dried, and
infrared-dried ‘Mejhoul’ date fruit.

Image Texture Parameter Fresh ‘Mejhoul’ Convective-Dried ‘Mejhoul’ Infrared-Dried ‘Mejhoul’

RHMean 70.35 a 70.08 ab 69.44 b
GHMean 59.65 a 57.44 b 55.54 c
BHMean 55.51 a 53.45 b 51.60 c
LHMean 91.94 a 90.28 b 88.77 c
aHMean 130.02 a 130.79 b 131.24 c
bHMean 130.46 a 130.73 b 130.91 b
XHMean 12.74 a 12.41 b 11.85 c
YHMean 12.58 a 12.08 b 11.40 c
ZHMean 10.56 a 10.07 b 9.30 c
Note. In the same row, means with different letters are significantly different (p < 0.05) according to
Tukey HSD test.

Models developed based on selected texture parameters extracted from images in

different color channels R, G, B, L, a, b, X, Y, Z, U, V, and S allowed for distinguishing fresh
and dried ‘Mejhoul’ date fruit with very high correctness. The accuracy average of classi-
fication of fresh, convective-dried, and infrared-dried samples reached 99.33% for a model
built using Random Forest (Table 2). However, slightly lower accuracies were obtained
for models developed using IBk (99.25%), Multilayer Perceptron (98.67%), Logit Boost and
Bayes Net (98.42%), and PART (97.83%). The higher differences occurred between fresh
and infrared-dried ‘Mejhoul’ date fruit, whereas the greatest misclassification was ob-
served between fresh and convective-dried fruit.

Table 2. Distinguishing accuracies of fresh, convective, and infrared-dried ‘Mejhoul’ date fruit using
machine learning models.
Foods 2024, 13, 1602 7 of 13

Predicted Class (Number of Cases) Average

Algorithm Convective-Dried Infrared-Dried Actual Class Accuracy
Fresh ‘Mejhoul’
‘Mejhoul’ ‘Mejhoul’ (%)
trees 400 0 0 Fresh ‘Mejhoul’
Random 6 393 1 Convective-dried ‘Mejhoul’ 99.33
Forest 0 1 399 Infrared-dried ‘Mejhoul’
399 1 0 Fresh ‘Mejhoul’
rules
23 376 1 Convective-dried ‘Mejhoul’ 97.83
PART
0 1 399 Infrared-dried ‘Mejhoul’
meta 394 6 0 Fresh ‘Mejhoul’
Logit 12 387 1 Convective-dried ‘Mejhoul’ 98.42
Boost 0 0 400 Infrared-dried ‘Mejhoul’
400 0 0 Fresh ‘Mejhoul’
lazy
8 391 1 Convective-dried ‘Mejhoul’ 99.25
IBk
0 0 400 Infrared-dried ‘Mejhoul’
functions 397 3 0 Fresh ‘Mejhoul’
Multilayer 13 387 0 Convective-dried ‘Mejhoul’ 98.67
Perceptron 0 0 400 Infrared-dried ‘Mejhoul’
Bayes 391 9 0 Fresh ‘Mejhoul’
Bayes 9 390 1 Convective-dried ‘Mejhoul’ 98.42
Net 0 0 400 Infrared-dried ‘Mejhoul’

Other performance metrics, such as True Positive Rate (TPR), False Positive Rate
(FPR), Precision, Recall, F-Measure, Matthews Correlation Coefficient (MCC), Receiver
Operating Characteristic Area (ROC Area), and Precision–Recall Area (PRC Area) con-
firmed the highest correctness of the classification of infrared-dried ‘Mejhoul’ date fruit
(Table 3). The greatest differentiation of infrared-dried fruit in terms of selected image
textures from fresh and convective-dried samples was particularly visible in the case of
the machine learning model built using Multilayer Perceptron. The values of TPR, Preci-
sion, Recall, F-Measure, MCC, ROC Area, and PRC Area were equal to 1.000, and FPR was
0.000. It confirmed that there was no mixing of cases between infrared-dried and other
classes. In the case of each model, the lowest values of TPR, Recall, F-Measure, MCC, and
ROC Area and the highest FPR were obtained for convective-dried ‘Mejhoul’ fruit. These
results confirmed that convective-dried date fruit was classified with the lowest correct-
ness.

Table 3. Performance metrics of distinguishing fresh and dried date fruit ‘Mejhoul’ using machine
learning models developed based on texture parameters.

ROC PRC
Algorithm Class TPR FPR Precision Recall F-Measure MCC
Area Area
trees Fresh ‘Mejhoul’ 1.000 0.008 0.985 1.000 0.993 0.989 1.000 1.000
Random Convective-dried ‘Mejhoul’ 0.983 0.001 0.997 0.983 0.990 0.985 0.998 0.998
Forest Infrared-dried ‘Mejhoul’ 0.998 0.001 0.998 0.998 0.998 0.996 1.000 1.000
Fresh ‘Mejhoul’ 0.998 0.029 0.945 0.998 0.971 0.956 0.985 0.939
rules
Convective-dried ‘Mejhoul’ 0.940 0.003 0.995 0.940 0.967 0.951 0.966 0.970
PART
Infrared-dried ‘Mejhoul’ 0.998 0.001 0.998 0.998 0.998 0.996 0.998 0.994
Fresh ‘Mejhoul’ 0.985 0.015 0.970 0.985 0.978 0.966 0.999 0.998
meta
Convective-dried ‘Mejhoul’ 0.968 0.008 0.985 0.968 0.976 0.964 0.996 0.996
Logit Boost
Infrared-dried ‘Mejhoul’ 1.000 0.001 0.998 1.000 0.999 0.998 0.999 0.998
lazy Fresh ‘Mejhoul’ 1.000 0.010 0.980 1.000 0.990 0.985 0.991 0.966
IBk Convective-dried ‘Mejhoul’ 0.978 0.000 1.000 0.978 0.989 0.983 0.981 0.985
Foods 2024, 13, 1602 8 of 13

Infrared-dried ‘Mejhoul’ 1.000 0.001 0.998 1.000 0.999 0.998 0.999 0.998
functions Fresh ‘Mejhoul’ 0.993 0.016 0.968 0.993 0.980 0.970 0.996 0.990
Multilayer Convective-dried ‘Mejhoul’ 0.968 0.004 0.992 0.968 0.980 0.970 0.992 0.974
Perceptron Infrared-dried ‘Mejhoul’ 1.000 0.000 1.000 1.000 1.000 1.000 1.000 1.000
Bayes Fresh date fruit 0.978 0.011 0.978 0.978 0.978 0.966 0.999 0.998
Bayes Convective-dried date fruit 0.975 0.011 0.977 0.975 0.976 0.964 0.997 0.996
Net Infrared-dried date fruit 1.000 0.001 0.998 1.000 0.999 0.998 0.999 0.998
TPR—True Positive Rate; FPR—False Positive Rate; MCC—Matthews Correlation Coefficient; ROC
Area—Receiver Operating Characteristic Area; PRC Area—Precision–Recall Area.

3.2. ‘Boufeggous’: Discrimination between Fresh, Convective, and Infrared Dried Fruit Based on
Machine Learning Models and Selected Texture Parameters
In the case of the ‘Boufeggous’ cultivar, the mean comparison of selected image tex-
ture parameters of fresh, convective-dried, and infrared-dried samples revealed the great-
est similarity between fresh and convective-dried fruit. Among the nine image texture fea-
tures analyzed for fresh and convective-dried ‘Boufeggous’ samples, only RHMean,
GHMean, LHMean, and XHMean were in the same homogenous group with no statisti-
cally significant difference. Graphs of texture mean values are presented in Table 4.

Table 4. Mean values of selected image texture parameters of fresh, convective-dried, and
infrared-dried ‘Boufeggous’ date fruit.

Convective-Dried Infrared-Dried
Image Texture Parameter Fresh ‘Boufeggous’
‘Boufeggous’ ‘Boufeggous’
RHMean 65.32 a 65.25 a 63.70 b
GHMean 56.54 a 56.18 a 53.20 b
BHMean 49.58 a 51.53 b 48.22 c
LHMean 88.10 a 88.04 a 85.40 b
aHMean 129.06 a 129.58 b 130.07 c
bHMean 131.64 a 130.59 b 131.01 c
XHMean 11.56 a 11.47 a 10.81 b
YHMean 11.61 a 11.60 a 10.74 b
ZHMean 9.08 a 9.72 b 8.84 c
Note. In the same row, means with different letters are significantly different (p < 0.05) according to
Tukey HSD test.

For this date palm cultivar, classification accuracies were slightly lower (Table 5) than
‘Mejhoul’. The average accuracy of distinguishing fresh, convective-dried, and infrared-
dried date fruit reached 94.33% for a model built using Random Forest, compared to
99.33% for ‘Mejhoul’. Nevertheless, the lowest average accuracy for analyzed models was
observed for the IBk algorithm, and it was equal to 91.00%.
As was reported for ‘Mejhoul’, fresh and infrared-dried ‘Boufeggous’ date fruit were
distinguished with the highest correctness. Thus, the highest number of misclassified
cases was between fresh and convective-dried fruit. In the case of the model developed
using IBk, 57 cases belonging to the actual class of convective-dried ‘Boufeggous’ fruit
were misclassified as fresh samples, and 47 cases of fresh fruit were incorrectly classified
as convective-dried fruit.
Foods 2024, 13, 1602 9 of 13

Table 5. Distinguishing accuracies of fresh, convective, and infrared dried ‘Boufeggous’ date fruit
using machine learning models.

Predicted Class (Number of Cases) Average

Algorithm Fresh Convective-Dried Infrared-Dried Actual Class Accuracy
‘Boufeggous’ ‘Boufeggous’ ‘Boufeggous’ (%)
trees 377 22 1 Fresh ‘Boufeggous’
Random 42 356 2 Convective-dried ‘Boufeggous’ 94.33
Forest 0 1 399 Infrared-dried ‘Boufeggous’
361 38 1 Fresh ‘Boufeggous’
rules
48 348 4 Convective-dried ‘Boufeggous’ 92.25
PART
0 2 398 Infrared-dried ‘Boufeggous’
meta 367 31 2 Fresh ‘Boufeggous’
Logit 35 365 0 Convective-dried ‘Boufeggous’ 94.25
Boost 0 1 399 Infrared-dried ‘Boufeggous’
352 47 1 Fresh ‘Boufeggous’
lazy
57 342 1 Convective-dried ‘Boufeggous’ 91.00
IBk
1 1 398 Infrared-dried ‘Boufeggous’
functions 370 26 4 Fresh ‘Boufeggous’
Multilayer 41 359 0 Convective-dried ‘Boufeggous’ 93.92
Perceptron 1 1 398 Infrared-dried ‘Boufeggous’
Bayes 373 26 1 Fresh ‘Boufeggous’
Bayes 43 357 0 Convective-dried ‘Boufeggous’ 93.67
Net 1 5 394 Infrared-dried ‘Boufeggous’

The classification values of TPR, FPR, Precision, Recall, F-Measure, MCC, ROC Area,
and PRC Area of fresh, convective-dried, and infrared-dried ‘Boufeggous’ date fruit are
presented in Table 6. The highest classification accuracy of infrared-dried date fruit was
confirmed by the lowest FPR and the highest values of other metrics. The value of 1.000
was obtained in the case of ROC Area and PRC Area for a model built using Random
Forest and ROC Area for a model developed using Bayes Net. The lowest FPR was found
for a model built by Bayes Net.

Table 6. Performance metrics of distinguishing fresh and dried ‘Boufeggous’ date fruit using ma-
chine learning models developed based on texture parameters.

ROC PRC
Algorithm Class TPR FPR Precision Recall F-Measure MCC
Area Area
Fresh ‘Boufeggous’ 0.943 0.053 0.900 0.943 0.921 0.880 0.990 0.980
trees
Convective-dried
Random 0.890 0.029 0.939 0.890 0.914 0.873 0.990 0.980
‘Boufeggous’
Forest
Infrared-dried ‘Boufeggous’ 0.998 0.004 0.993 0.998 0.995 0.993 1.000 1.000
Fresh ‘Boufeggous’ 0.903 0.060 0.883 0.903 0.892 0.838 0.961 0.938
rules Convective-dried
0.870 0.050 0.897 0.870 0.883 0.826 0.967 0.927
PART ‘Boufeggous’
Infrared-dried ‘Boufeggous’ 0.995 0.006 0.988 0.995 0.991 0.987 0.995 0.984
Fresh ‘Boufeggous’ 0.918 0.044 0.913 0.918 0.915 0.873 0.988 0.975
meta
Convective-dried
Logit 0.913 0.040 0.919 0.913 0.916 0.874 0.987 0.978
‘Boufeggous’
Boost
Infrared-dried ‘Boufeggous’ 0.998 0.003 0.995 0.998 0.996 0.994 0.999 0.999
Fresh ‘Boufeggous’ 0.880 0.073 0.859 0.880 0.869 0.803 0.905 0.800
lazy
Convective-dried
IBk 0.855 0.060 0.877 0.855 0.866 0.800 0.899 0.804
‘Boufeggous’
Foods 2024, 13, 1602 10 of 13

Infrared-dried ‘Boufeggous’ 0.995 0.003 0.995 0.995 0.995 0.993 0.996 0.992
Fresh ‘Boufeggous’ 0.925 0.053 0.898 0.925 0.911 0.866 0.973 0.948
functions
Convective-dried
Multilayer 0.898 0.034 0.930 0.898 0.913 0.872 0.979 0.963
‘Boufeggous’
Perceptron
Infrared-dried ‘Boufeggous’ 0.995 0.005 0.990 0.995 0.993 0.989 0.998 0.995
Fresh ‘Boufeggous’ 0.933 0.055 0.894 0.933 0.913 0.869 0.982 0.964
Bayes
Convective-dried
Bayes 0.893 0.039 0.920 0.893 0.906 0.860 0.981 0.966
‘Boufeggous’
Net
Infrared-dried ‘Boufeggous’ 0.985 0.001 0.997 0.985 0.991 0.987 1.000 0.999
TPR—True Positive Rate; FPR—False Positive Rate; MCC—Matthews Correlation Coefficient; ROC
Area—Receiver Operating Characteristic Area; PRC Area—Precision–Recall Area.

The above-mentioned tables showed an important trend in the variation in the above-
mentioned texture parameters that can be ranged into two different groups: group with
high values for ID drying (aHMean and bHMean); and group with lower values for ID
drying (GHMean, BHMean, LHMean, XHMean, YHMean, and ZHMean). Nevertheless,
both groups are characterized by convective drying texture features comprising fresh and
infrared values. This trend may reflect significant texture variation between fresh, convec-
tive, and infrared dried ‘Mejhoul’ date fruit with high differences between ID and fresh
compared to CD and fresh fruit. For ‘Boufeggous’, these findings are applied only for the
case of aHMean and YHMean features.
Regarding developed models, Random forest was the most accurate and predicting
algorithm for both ‘Mejhoul’ and ‘Boufeggous’ while PART and lazy lBk were respectively
the lower predicting models for ‘Mejhoul’ and ‘Boufeggous’. Such constatation was re-
ported by [28] while comparing the effectiveness and accuracy of six algorithms used for
orange (Citrus sinensis) classification. Among Logistic Regression, Naïve Bayes, Decision
Trees, Neural Networks, K-nearest neighbors, and Random Forests models, authors
found that Logistic Regression was more accurate and precise, with an accuracy of 91%
and a precision score of 92%.
In another study related to cultivar date fruit classification, [29] employed algorithms
developed on the basis of ANN and LR approaches to discriminate between seven date
fruit cultivars. The lowest accuracies of 91 and 92.2% compared to the results of this paper
were obtained for distinguishing color, shape, and pomological features extracted from
images.
In relation to these findings and the scope of this study, the previous literature data
considered mainly the application of destructive analysis (instead of non-destructive
methods) to assess date fruit quality in terms of physicochemical properties under differ-
ent experimental conditions and various postharvest treatments [30–33]. In this way,
mathematical models were used in addition to “response surface methodology” to ex-
press the drying kinetics and behavior of date fruit [34–38] without considering machine
learning approaches for the specific case of date fruit drying. Nevertheless, machine learn-
ing models were employed for other fruit species to determine and assess dried fruit qual-
ity. For example, Raihen and Akter [39] developed deep-learning models for the classifi-
cation of dried grapefruit types. Sağlam and Çetin [40] applied machine learning models
to estimate the drying characteristics of dried apple slices by employing artificial neural
networks, k-nearest neighbors, random forests, etc. For the three apple cultivars, the high-
est correlation coefficients were about 0.98 for moisture ratio estimation using a Random
Forest algorithm. The effectiveness and usefulness of machine learning models in rapid
estimation of chemical properties of apples were also proven with five machine learning
algorithms [41]. Also, machine learning algorithms were used to predict the sweetness of
dried bananas using Linear Discriminant Analysis (LDA), K Nearest Neighbors (KNN),
Decision Tree (CART), Random Forest (RF), and Support Vector Machines (SVM) models.
Foods 2024, 13, 1602 11 of 13

Among these five algorithms, authors reported that RF and SVM allowed for a high pre-
diction accuracy of 86% [42]. The above-mentioned studies stated the importance of using
ML algorithms to predict and estimate the biochemical properties of dried fruit with high
accuracy, and the results presented in this paper are comparable in terms of high accuracy.
Our study presented in this paper revealed the usefulness of machine learning models in
determining the external quality of dried date fruit. However, further research can be re-
lated to the estimation of the physicochemical properties of date fruit using image pro-
cessing and artificial intelligence.

4. Conclusions
Through this work, the effect of mild convective and infrared drying techniques on
the changes in the external quality of date fruit was revealed in a non-destructive and
objective manner using a flatbed scanner. The developed machine learning models, using
selected texture parameters from images in different color channels R, G, B, L, a, b, X, Y,
Z, U, V, and S, classified accurately fresh, convective-dried, and infrared-dried ‘Mejhoul’
and ‘Boufeggous’ date palm fruit. Thus, samples were classified with high correctness of
99.33% and 94.33 for ‘Mejhoul’ and ‘Boufeggous’, respectively, using the Random Forest
algorithm, which was more accurate than other machine learning models tested under
this work. Furthermore, the usefulness of image features and models built using artificial
intelligence was confirmed and proved the greater effect of infrared drying on the im-
provement in the external appearance of date fruit for both cultivars.
From this perspective, selected texture features from date fruit images can be used as
inputs to develop more accurate models using traditional machine learning and deep-
learning algorithms that link those externally extracted features with other internal qual-
ity parameters such as biochemical and nutritional compounds. Additionally, other inno-
vative drying methods can be evaluated for other date fruit cultivars with a focus on the
elucidation of machine learning approaches that added value to the total quality assess-
ment.

Author Contributions: Conceptualization, E.R. and Y.N.; methodology, Y.N. and E.R.; software,
E.R.; validation, E.R.; investigation, E.R. and Y.N.; writing—original draft preparation, E.R.; writ-
ing—review and editing, Y.N. and E.R.; project administration, Y.N.; funding acquisition, Y.N. All
authors have read and agreed to the published version of the manuscript.
Funding: This research is part of project No. 2022/45/P/NZ9/03904, co-funded by the National Sci-
ence Centre and the European Union’s Horizon 2020 research and innovation program under the
Marie Skłodowska-Curie grant agreement no. 945339. Project title, “A novel approach to the assess-
ment of date fruit quality (Phoenix dactylifera L.) under different storage conditions, using innovative
models based on image analyses and machine learning” (M-LEARN4DATE).
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: The original contributions presented in the study are included in the
article, further inquiries can be directed to the corresponding author.
Conflicts of Interest: The authors declare no conflicts of interest.

References
1. Alsmairat, N.; Othman, Y.; Ayad, J.; Al-Ajlouni, M.; Sawwan, J.; El-Assi, N. Anatomical Assessment of Skin Separation in Date
Palm (Phoenix dactylifera L. var. Mejhoul) Fruit during Maturation and Ripening Stages. Agriculture 2023, 13, 38.
2. Rakesh Reddy, S.V.; Singh, R.S.; Meena, R.; Berwal, M.K.; Sarolia, D.K.; Palpandian, P. Impact of Hot Water Pre-Treatments on
the Drying Efficiency and Quality of Dates cv. Medjool. Horticulturae 2023, 9, 784.
3. Uchoi, J.; Nikhumbhe, P.H.; Kumar, A.; Patidar, A.; Harish, G.D. Impact of inclined solar drier for dehydration quality in
khadrawy dates during doka maturity stage at north western arid India. Int. J. Curr. Microbiol. Appl. Sci. 2020, 9, 119–125.
4. Muñoz-Bas, C.; Muñoz-Tebar, N.; Candela-Salvador, L.; Pérez-Alvarez, J.A.; Lorenzo, J.M.; Viuda-Martos, M.; Fernández-López,
J. Quality Characteristics of Fresh Date Palm Fruits of “Medjoul” and “Confitera” cv. from the Southeast of Spain (Elche Palm
Grove). Foods 2023, 12, 2659.
Foods 2024, 13, 1602 12 of 13

5. Noutfia, Y.; Ropelewska, E. What can artificial intelligence approaches bring to an improved and efficient harvesting and post-
harvest handling of date fruit (Phoenix dactylifera L.)? A review. Postharvest Biol. Technol. 2024, 213, 112926.
6. Bassey, E.J.; Cheng, J.H.; Sun, D.W. Novel nonthermal and thermal pretreatments for enhancing drying performance and im-
proving quality of fruits and vegetables. Trends Food Sci. Technol. 2021, 112, 137–148.
7. Alsmairat, N.; Al-Qudah, T.; El-Assi, N.; Mehyar, G.; Gammoh, I.; Othman, Y.; Araj, S.; Al-Antary, T. Effect of drying process
on physical and chemical properties of ‘Medjool’ date palm fruits. Fresenius Environ. Bull. 2019, 28, 1552–1559.
8. Özlem, E.; Yeliz, I. Modeling of drying processes of dates (Phoenix, Arecaceae) with oven or TGA and microbiological properties
of fresh and dried dates. Int. J. Fruit Sci. 2020, 20, S1530–S1538.
9. Elwakeel, A.E.; Tantawy, A.A.; Alsebiey, M.M.; Elliby, A.K. The date fruit drying systems: Acritical over review. Al-Azhar J.
Agric. Eng. 2022, 3, 26–36.
10. Falade, K.O.; Abbo, E.S. Air-drying and rehydration characteristics of date palm (Phoenix dactylifera L.) fruits. J. Food Eng. 2007,
79, 724–730.
11. Shahdadi, F.; Mirzaei, H.O.; Daraei Garmakhany, A. Study of phenolic compound and antioxidant activity of date fruit as a
function of ripening stages and drying process. J. Food Sci. Technol. 2015, 52, 1814–1819.
12. İzli, G. Total phenolics, antioxidant capacity, colour and drying characteristics of date fruit dried with different methods. Food
Sci. Technol. 2016, 37, 139–147.
13. Noutfia, Y.; Benali, A.; Alem, C.; Zegzouti, Y.F. Design of a solar dryer for small-farm level use and studying fig quality. Acta
Sci. Pol. Technol. Aliment. 2018, 17, 359–365.
14. Elghazali, M.N.; Tawfeuk, H.Z.; Gomaa, R.A.; Tantawy, A.A. Technological Studies on Aswan Dry Dates Products After Dehy-
dration. Assiut J. Agric. Sci. 2020, 51, 32–49.
15. Tireki, S. Effective diffusivity determination of date (Phoenix dactylifera L.) leather in infrared drying: Effect of cooking time.
NOHU J. Eng. Sci. 2023, 12, 1558–1565.
16. Noutfia, Y.; Ropelewska, E. Innovative models built based on image textures using traditional machine learning algorithms for
distinguishing different varieties of Moroccan date palm fruit (Phoenix dactylifera L.). Agriculture 2022, 13, 26.
17. Noutfia, Y.; Ropelewska, E. Comprehensive characterization of date palm fruit ‘Mejhoul’ (Phoenix dactylifera L.) using image
analysis and quality attribute measurements. Agriculture 2022, 13, 74.
18. Szczypiński, P.M.; Strzelecki, M.; Materka, A. Mazda-a software for texture analysis. In Proceedings of the 2007 International
Symposium on Information Technology Convergence (ISITC 2007), Jeonju, Republic of Korea, 23–24 November 2007; pp. 245–
249.
19. Szczypiński, P.M.; Strzelecki, M.; Materka, A.; Klepaczko, A. MaZda—A software package for image texture analysis. Comput.
Methods Programs Biomed. 2009, 94, 66–76. https://doi.org/10.1016/j.cmpb.2008.08.005.
20. Strzelecki, M.; Szczypiński, P.; Materka, A.; Klepaczko, A. A software tool for automatic classification and segmentation of
2D/3D medical images. Nucl. Instrum. Methods Phys. Res. Sect. A Accel. Spectrom. Detect. Assoc. Equip. 2013, 702, 137–140.
https://doi.org/10.1016/j.nima.2012.09.006.
21. Witten, I.H.; Frank, E. Data Mining: Practical Machine Learning Tools and Techniques, 2nd ed.; Elsevier: San Francisco, CA, USA,
2005.
22. Bouckaert, R.R.; Frank, E.; Hall, M.; Kirkby, R.; Reutemann, P.; Seewald, A.; Scuse, D. WEKA Manual for Version 3-9-1; University
of Waikato: Hamilton, New Zealand, 2016.
23. Frank, E.; Hall, M.A.; Witten, I.H. The WEKA Workbench. In Data Mining: Practical Machine Learning Tools and Techniques, 4th
ed.; Online Appendix; Morgan Kaufmann: Burlington, MA, USA, 2016.
24. Ropelewska, E. Diversity of Plum Stones Based on Image Texture Parameters and Machine Learning Algorithms. Agronomy
2022, 12, 762.
25. Sabanci, K.; Aslan, M.F.; Ropelewska, E.; Unlersen, M.F. A convolutional neural network-based comparative study for pepper
seed classification: Analysis of selected deep features with support vector machine. J. Food Process Eng. 2022, 45, e13955.
26. Cinar, I.; Koklu, M. Classification of Rice Varieties Using Artificial Intelligence Methods. Int. J. Intell. Syst. Appl. Eng. 2019, 7,
188–194.
27. Ropelewska, E. Distinguishing lacto-fermented and fresh carrot slice images using the Multilayer Perceptron neural network
and other machine learning algorithms from the groups of Functions, Meta, Trees, Lazy, Bayes and Rules. Eur Food Res Technol
2022, 248, 2421–2429.
28. Vaishnav, D.; Rao, B.R. Comparison of machine learning algorithms and fruit classification using orange data mining tool. In
Proceedings of the 2018 3rd International Conference on Inventive Computation Technologies (ICICT), Coimbatore, India, 15–
16 November 2018; pp. 603–607.
29. Koklu, M.; Kursun, R.; Taspinar, Y.S.; Cinar, I. Classification of date fruits into genetic varieties using image analysis. Math.
Probl. Eng. 2021, 2021, 4793293.
30. Ramadan, B.R.; EL-Rify, M.N.A.; Abd El-Hamid, A.A.; Abd El-Majeed, M.H. Effect of some treatments on chemical composition
and quality properties of Saidy date fruit (Phoenix dactylifera L.) during storage. Assiut J. Agric. Sci. 2016, 47, 107–124.
31. Tang, Z.X.; Shic, L.E.; Aleid, S.M. Date fruit: Chemical composition, nutritional and medicinal values, products. J. Sci. Food Agric.
2013, 93, 2351–2361.
32. Juhaimi, F.A.; Ghafoor, K.; Özcan, M.M. Physical and chemical properties, antioxidant activity, total phenol and mineral profile
of seeds of seven different date fruit (Phoenix dactylifera L.) varieties. Int. J Food Sci. Nutr. 2012, 63, 84–89.
Foods 2024, 13, 1602 13 of 13

33. AL Juhaimi, F.; Ghafoor, K.; Özcan, M.M. Physicochemical properties and mineral contents of seven different date fruit (Phoenix
dactylifera L.) varieties growing from Saudi Arabia. Environ. Monit. Assess. 2014, 186, 2165–2170.
34. Hadibi, T.; Mennouche, D.; Boubekri, A.; Arıcı, M.; Wang, Y.; Li, M.; Hassanien, R.H.E.; Shirkole, S.S. Experimental investigation,
performance analysis, and optimization of hot air convective drying of date fruits via response surface methodology. Renew.
Energy 2024, 226, 120404.
35. Sansaniwal, S.K.; Kumar, M.; Kumar Sahdev, R.; Bhutani, V.; Manchanda, H. Toward natural convection solar drying of date
palm fruits (Phoenix dactylifera L.): An experimental study. Environ. Prog. Sustain. Energy 2022, 41, e13862.
36. Al-Awaadh, A.M.; Hassan, B.H.; Ahmed, K.M.A. Hot Air Drying Characteristics of Sukkari Date (Phoenix dactylifera L.) and
Effects of Drying Condition on Fruit Color and Texture. Int. J. Food Eng. 2015, 11, 421–434.
37. Kumar, M.; Shimpy, M.; Sahdev, R.K.; Sansaniwal, S.K.; Bhutani, V.; Manchanda, H. Experimental forced convection green-
house and indirect cabinet drying of date fruits: A comparative study. J. Therm. Anal. Calorim. 2023, 148, 5437–5454.
38. Boubekri, A.; Benmoussa, H.; Mennouche, D. Solar drying kinetics of date palm fruits assuming a step-wise air temperature
change. J. Eng. Sci. Technol. 2009, 4, 292–304.
39. Raihen, M.N.; Akter, S. Prediction modeling using deep learning for the classification of grape-type dried fruits. Int. J. Math.
Comput. Eng. 2024, 2, 1–12.
40. Sağlam, C.; Çetin, N. Machine learning algorithms to estimate drying characteristics of apples slices dried with different meth-
ods. J. Food Process. Preserv. 2022, 46, e16496.
41. Çetin, N.; Sağlam, C. Rapid detection of total phenolics, antioxidant activity and ascorbic acid of dried apples by chemometric
algorithms. Food Biosci. 2022, 47, 101670.
42. Al-Mashhadany, S.A.; Hasan, H.A.; Al-Sammarraie, M.A.J. Using Machine Learning Algorithms to Predict the Sweetness of
Bananas at Different Drying Times. J. Ecol. Eng. 2024, 25, 231–238.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual au-
thor(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.