This is an open access article under the CC BY license: 

 

 

Al-Khwarizmi 

Engineering 

Journal 
 

Al-Khwarizmi Engineering Journal, Vol. 19, No. 4, December, (2023) 

P. P. 67- 77 
 

 

Development of an ANN Model for RGB Color Classification using 

the Dataset Extracted from a Fabricated Colorimeter 
 

Shahad A. Mnati*                Furat I. Hussein** 

Ahmed Issa*** 
*, **Department of Mechatronics Engineering/ Al-Khwarizmi College of Engineering/ University of Baghdad, 

Baghdad/ Iraq 

*** Department, Faculty of Engineering and Information Technology/Mechatronics and Medical Devices Engineering 

Al- Azhar University – Gaza/ Gaza 1277/ Palestine 

                            Corresponding Author: *Email: shahed.ali1202a@kecbu.uobaghdad.edu.iq 
**Email: furatnejjar@uobaghdad.edu.iq 

***Email: a.issa@alazhar.edu.ps  

 

(Received 21 May 2023; accepted 11 September 2023) 

https://doi.org/10.22153/kej.2023.09.003 
 

 

Abstract 

 
Codes of red, green, and blue data (RGB) extracted from a lab-fabricated colorimeter device were used to build a 

proposed classifier with the objective of classifying colors of objects based on defined categories of fundamental colors. 
Primary, secondary, and tertiary colors namely red, green, orange, yellow, pink, purple, blue, brown, grey, white, and 

black, were employed in machine learning (ML) by applying an artificial neural network (ANN) algorithm using Python. 

The classifier, which was based on the ANN algorithm, required a definition of the mentioned eleven colors in the form 

of RGB codes in order to acquire the capability of classification. The software's capacity to forecast the color of the code 

that belongs to an object under detection is one of the results of the proposed classifier. The work demanded the collection 

of about 5000 color codes which in turn were subjected to algorithms for training and testing. The open-source platform 

TensorFlow for ML and the open-source neural network library Keras were used to construct the algorithm for the study. 

The results showed an acceptable efficiency of the built classifier represented by an accuracy of 90% which can be 

considered applicable, especially after some improvements in the future to makes it more effective as a trusted 

colorimeter. 

 
Keywords: Colorimeter, RGB classifier, ANN, TensorFlow, ML. 

 
 

1. Introduction 
 
Advanced methods like ML algorithms can be 

employed to address the color classification issue. 

Neural network methods are replacing older 
statistical methods and algorithms in the field of 

machine vision. Due to weak planning, the latter 

encounters numerous challenges that raise 
obstacles when developing an ML model for the 

computer vision system. Executives frequently set 

challenging goals for the data science team during 

the planning process. Neural network techniques, 
as opposed to conventional algorithms, are 

typically quite beneficial for certain challenges [1]. 

One of the simplest non-parametric methods of 
color recognition is color categorization, utilizing 

the K-Nearest Neighbors (KNN) algorithm with 

feature extraction. Most real-world datasets do not 

conform to mathematical theoretical assumptions 
and this will be very useful in practice. 

Classification is widely regarded as one of the most 

extensively studied problems within the domains 
of ML and data mining. Despite its well-

established status in the field of ML, classification 

encounters challenges such as effectively 
managing missing data [2]. The K-mean algorithm 

mailto:shahed.ali1202a@kecbu.uobaghdad.edu.iq
mailto:furatnejjar@uobaghdad.edu.iq
mailto:a.issa@alazhar.edu.ps
https://doi.org/10.22153/kej.2023.09.003


Shahad A. Mnati                                           Al-Khwarizmi Engineering Journal, Vol. 19, No. 4, P.P. 67- 77 (2023) 

 

68 

is another well-known ML approach that can be 
used to extract colors from photographs and 

categorize each image from a collection of images 

based on color space values. Any color space can 
be utilized such as RGB, CYMK, HSV, etc. [3]. 

Kusumo B.S. et al. [4] used RGB image 

processing properties in classifying corn diseases, 
which are one of Indonesia's most significant 

crops. Oriented FAST and rotated BRIEF (ORB), 

scale-invariant feature transform (SIFT), and 

speeded-up robust features were used to extract the 
features (SURF). The categorization was 

performed using Decision Tree (DT), Naive Bayes 

(NB), Support Vector Machines (SVM), and 
Random Forest (RF). Behera S.K. et al. [5] have 

classified disease and the severity of disease in 

oranges using the SVM algorithm with the 
assistance of the K-mean clustering approach and 

fuzzy logic. The suggested framework 

demonstrated an accuracy of 90% in illness 

categorization and severity assessments. Kumari, 
R. S. S. et al. [6] attempted a traditional method of 

classifying fruit that relies on manual operations 

and visual acuity. The SVM classifiers classify 
based on statistical and co-occurrence features 

derived from the wavelet transform. The accuracy 

of classification for the proposed system is 95.3%. 

Panigrahi, K.P. et al. [7] proposed machine 
learning methods such as Decision Tree (DT), 

Random Forest (RF), Support Vector Machine 

(SVM), Naive Bayes (NB), and K-Nearest 
Neighbor (KNN) to be used to diagnose crop 

diseases using plant images. When compared to 

other classification algorithms, the RF algorithm 
obtains the highest accuracy of 79.23%. Hameed, 

N. et al. [8] followed a method of intelligent digital 

diagnosis that improved the classification precision 

of multiple diseases. The diagnostic accuracy is 
96.47% using a Multi-Class Multi-Level (MCML) 

classification algorithm evaluated on 3672 

classified images collected from various sources. 
Nagata, T. et al. [9] classified skin tears using a 

machine learning method to identify the skin layers 

using digital RGB color and subsequently treat the 
wound. The determination of the wound layer 

showed an accuracy of 74% and 71% for the SVM 

and RF respectively. 

The aim of the current study is to employ the 
ANN algorithm in order to build an ML model that 

acts as a classifier and has a good opportunity to 

predict the colors of different objects based on 
datasets extracted from a lab-fabricated 

colorimeter. 

 

 

2. Materials and Methods 

2.1 Dataset Collection 
 
The dataset employed in the study was extracted 

from an RGB colorimeter device (figure 1) 

implemented in the lab using a number of low-cost 
electronics, electrical parts, and other 

complementary components. In this process, the 

fabricated colorimeter was designed with specific 
consideration and high precision to measure the 

RGB values of a number of colored samples in 

order to use these values as input for the proposed 

ANN model. The number of tested samples is 
about 5000 cards of colored paper stamped with 

standard names for wallpaper and paint companies 

in Iraq, namely Jotun, Al-Marjan, Al-Yaqut, and 
Jazeera. The colorimeter was used to record the 

code of each inspected colored card as RGB code 

and save it as a comma-separated values (csv) file. 

 

 
 

Fig. 1. The fabricated colorimeter and standard 

colored cards 

 

 
2.2 RGB and Color Spaces 

 

In essence, a color space is a collection of colors 

that assist in reproducing analog and digital 

representations of color along with physical device 
profiling. Another way to think of color spaces is 

as an abstract mathematical model that enables the 

representation of colors as numbers [10]. The RGB 
system-dependent color space is represented by 24-

bit, with 8 bits allotted to each single channel of R, 

G, and B. As a result, the range of values for each 

channel is 0 to 255. All colors can be represented 



Shahad A. Mnati                                           Al-Khwarizmi Engineering Journal, Vol. 19, No. 4, P.P. 67- 77 (2023) 

 

69 

using various red, green, and blue hues, according 
to this color space paradigm. 

 

2.3 ANN Technology 
 

ANN is essentially a mathematical 
representation of a process that was created 

empirically rather than by employing mass and 

energy balances surrounding the process. A neural 
network is made up of a network of partially 

interconnected processing units, or nodes, 

structured in layers, much like the neurons in the 

brain. They also feature connections between the 
nodes of subsequent layers. The fundamental 

structure of a single neuron or node is within a 

neural network model with inputs, activation 
functions, and a single output. Weights are 

calculated values that represent the relationships 

between nodes [11, 12]. Figure 2 shows a 
schematic diagram of a fundamental neural 

network where 𝑌𝑖 is the output, and W are the 

weights that represent the "strength" of the 

connections between the neurons. Each neuron in 
the hidden layer receives weighted inputs plus bias 

from each neuron in the preceding layer [13]: 

 

Fig. 2. Schematic diagram of fundamental neural 

network [12] 

 
 

𝑍𝑖 = ∑ 𝑋𝑘
𝑗−1𝑊𝑘,𝑖 − 𝑏𝑘

𝑁𝑗−1

𝑘=1

                               … (1) 

where 𝑊𝑘,𝑖is the weight of the link between 

𝑛𝑜𝑑𝑒 and all the nodes in the preceding layers, 𝑏𝑘 

is the bias to the node, 𝑁𝑗−1 is the number of nodes 

in layer 𝑗 − 1, and 𝑋𝑘
𝑗−1

 signifies the input which 

in this case is the (RGB) values from the 𝑘 − 𝑡ℎ 

node in the 𝑗 − 𝑡ℎ layer. 

The summation of equation 1 is fed to the 

activation function, which generates the node's 

output, which is calculated as 𝑌𝑖 = 𝑓(𝑍𝑖) which is 

one of the eleven fundamental color categories 

(red, green, orange, yellow, pink, purple, blue, 

brown, grey, white, or black). The activation 

function for each layer in the neural network (NN) 
model is a rectified linear unit (ReLU) which has 

been applied in the current work. The activation 

functions are used to introduce non-linearity in 
order to assist the NN in learning complex data 

patterns.  
 

2.4 Performance Metrics 

 
Various metrics are employed for the estimation 

of an ANN performance and quality. Performance 

metrics were considered in this study for evaluating 
the performance of the ML model. Different 

metrics are used in evaluating the performance of 

the model such as accuracy, precision, recall, and 

F1score [14, 15]: 

2.4.1 Accuracy (Acc): The accuracy of a model 

determines how well it classifies the target 

variable. The accuracy of the selected model 
is given as follows [16]: 

𝐴𝑐𝑐 =  
𝑇𝑃 + 𝑇𝑁

𝑇𝑃 + 𝑇𝑁 + 𝐹𝑃 + 𝐹𝑁
                        … (2) 

where TP is the total positive which refers to the 

number of accurately labeled positive samples, TN 

is total negative which refers to the number of 

negative samples that have been accurately 

classified, FP is the false positive which refers to 
the misclassification of negative samples as 

positive, and FN is the false negative which refers 

to the misclassification of positive samples as 
negative. 

2.4.2 Precision (Prec) is the proportion of 

correctly predicted positives to the total 

number of positive observations predicted. 
A low rate of false positives is correlated 

with precision [16]: 

𝑃𝑟𝑒𝑐 =
𝑇𝑃

𝑇𝑃 + 𝐹𝑃
                                       … (3)  

2.4.3 Recall (Rec): is the proportion of positively 
predicted observations to the total number of 

observations in the actual class. In general, 

recall values above 0.5 are considered 
acceptable [16]: 

𝑅𝑒𝑐 =  
𝑇𝑃

𝑇𝑃 + 𝐹𝑁
                                                 … (4) 

2.4.4 F1 Score (F1S): is the harmonic mean of 

precision and recall, it is a more accurate 
measurement than precision [16]: 

𝐹1S = 2 ×  
𝑃𝑟𝑒𝑐 −  𝑅𝑒𝑐

𝑃𝑟𝑒𝑐 +  𝑅𝑒𝑐
                          … (5) 

where F1-Score is the most effective 

measurement when the target's classification is 
imbalanced. 



Shahad A. Mnati                                           Al-Khwarizmi Engineering Journal, Vol. 19, No. 4, P.P. 67- 77 (2023) 

 

70 

2.4.5 Confusion matrix: is used to summarize the 
performance of the ANN algorithm. To 

discriminate between the RGB classes 

(black, white, …, etc.), the actual colors are 
assigned to positive and negative. However, 

the prediction colors are assigned to true for 

and false for no damage [17] as shown in 
Figure 3. 

The primary goal of the model is to optimize 

the true positives and true negatives, while 

minimizing the false positives and false 
negatives to achieve optimal performance. The 

total number of predictions is equal to the 

summation of TP, TN, FP and FN. 
 

 
 

Fig. 3. confusion matrix for a binary classifier. 

 

 

2.5 Methodology Development for ANN 

 
In short, an ANN model consists of five layers 

stacked sequentially. Each layer is composed of a 
number of neurons. Each neuron of the first layer 

takes the input feature (the red, green, and blue 

values), multiplies it by its initial weight, calculates 

the total number of each neuron together with the 
bias, and then passes it through the activation 

function to the next layer of neurons, giving the 

result again as an input to the next layer.  Figure 4 
shows the segments of the proposed ANN as 

follows: 

 The initial layer consists of three neurons, which 

collectively possess a total of 12 parameters. 
These parameters include 9 weights and three 

biases. 

The second layer, consisting of 24 neurons, 

contains a total of 96 parameters. These parameters 
are comprised of 72 weights and 24 biases. 

 

2.6 Analytical Framework 
 

Python v2.13.0 was used to build, train, and 

validate the model using TensorFlow v2.13.0 and 
Keras libraries. TensorFlow uses a single data flow 

diagram to represent all machine learning 

computations and states.  
The experimental data collected from the 

colorimeter device is employed as an input for the 

ML algorithm. The ANN model has five layers. It 

is a densely connected neural network in the dense 
layer, also known as a fully connected layer, and is 

a fundamental component of neural networks. It is 

characterized by a deep interconnection structure, 
wherein each neuron within the dense layer 

receives input from every neuron in the preceding 

layer. The dense layer is employed in this model, 
making it a widely utilized component. The dense 

layer conducts matrix-vector multiplication in the 

underlying process. The values employed within 

the matrix are, in fact, parameters that possess the 
capability to be trained and adjusted through the 

utilization of backpropagation.  

 
 

Fig. 4. The proposed ANN configuration. 



Shahad A. Mnati                                           Al-Khwarizmi Engineering Journal, Vol. 19, No. 4, P.P. 67- 77 (2023) 

 

71 

 

Fig. 5. The methodology of the proposed ANN. 

 

 

 

Fig. 6. Confusion matrix correlation before training. 

 

 



Shahad A. Mnati                                           Al-Khwarizmi Engineering Journal, Vol. 19, No. 4, P.P. 67- 77 (2023) 

 

72 

 

3. Results and Discussion 

 

Results are obtained through a number of 

sequential steps in order to achieve and present the 

color classification. In the first step, the dataset is 

obtained from the measurement of the colored 
cards using the fabricated colorimeter and by 

recording the RGB codes manually. After that, 

preprocessing is applied in order to label and 
modify the null data. The training and validation 

phase is the subsequent step which is started by 

adjusting the classification settings of the 

algorithm. The prediction phase comes in order to 
confirm the accuracy of the trained model. 

Eventually, the performance metrics for the 

algorithm are applied with a view to evaluate 
results. Figure 6 shows the manner of data 

correlation before the training phase. 

The graphical correlation of the RGB data is 
presented in this section. 

It is important to mention that the dataset was 

split into training data, testing data, and the setup 

of the ANN layers which requires the use of proper 
neural network settings to get better results. The 

confusion matrix is used to evaluate the 

performance of a classifier using Python software. 
Actual labels are represented by rows, while the 

predicted labels are represented by columns. To get 

better classification results, the ANN algorithm 
was adjusted at certain settings such as the number 

of epochs = 5001, verbose=0, batch_size = 2048, 

and validation_split = 0.2. The relationship 

between training and validation versus the no. 
epochs are shown in figure 6. Here, the ‘Epochs’ 

means the number of times at which the model 

iterates over the entire x and y data while 
training. Batch size is the number of samples after 

which the parameters are updated while 

training. Verbose is used for displaying the 

training progress. 
 

Fig. 7. The accuracy of the training process of the 

ANN model. 

Moreover, the loss of the ANN model is shown 

in figure 6 which indicates that the accuracy is 
improving during the training phase. Also, it can be 

seen that both training and validation accuracies 

are matched at 90% which means that the 
classification algorithm is efficient. 

 

 
 

Fig. 8. The loss of the training process of the ANN 

model. 

 

 

In addition to the training and validation of the 

data, the built and saved ANN model was used in 

the prediction phase. Figure 9a shows the 

visualization of the actual values and the predicted 
values based on the ANN model training data. The 

visualization of the actual values and predicted 

values of the tested data after using the ANN model 
showed an acceptable match as seen in figure 9b. 

 

 
(a) 

 
 

(b) 

Fig. 9. The prediction of the trained model and b) 

tested model. 

 

 



Shahad A. Mnati                                           Al-Khwarizmi Engineering Journal, Vol. 19, No. 4, P.P. 67- 77 (2023) 

 

73 

Figure 10 shows the confusion matrix of the 
trained data for the labeled color. Looking at the 

first value (158) gives a piece of information that 

the model predicts the presence of 158 unique 
shades of red. As seen in the figure, the first pink 

value in the sixth row is 16, which means there are 

16 instances of pink incorrectly identified as red. 

Table 1 tabulates the values of performance 
measures for the trained data. It can be seen that the 

ML model correctly predicts the color green 96% 

of the time, as well as other colors with different 
percentages, with an accuracy of 89% for all eleven 

classes. 
 

 

Table 1, 

The performance measures the trained model. 

 

 

 

 

Fig. 10.  Confusion matrix correlation after training. 

 

 
Figure 11 shows the confusion matrix of the 

tested data for labeled colors. It is well known that 

this model predicts the existence of thirty-six 

unique shades represented by the first value of red 
(36). The first pink value in the sixth row is six, 

which means that there were 6 instances in which 

pink was incorrectly identified as red. Table 2 

tabulates the values of the performance measures 

of the tested data. It can be seen that the ML model 

correctly predicts the color green 94% of the time, 
as well as other colors with different percentages, 

and has an accuracy of 90% for all eleven classes. 

 Precision Recall F1score Support 

Red 0.79 0.83 0.81 190 

Green 0.96 0.95 0.95 1168 

Blue 0.94 0.93 0.93 885 

Yellow 0.87 0.80 0.83 225 

Orange 0.81 0.83 0.82 167 

Pink 0.89 0.83 0.86 451 

Purple 0.82 0.86 0.84 450 

Brown 0.84 0.91 0.87 311 
Grey 0.73 0.80 0.76 144 

Black 0.88 0.88 0.88 16 

White 0.70 0.58 0.64 24 

     

Accuracy   0.89 4004 

Macro avg 0.84 0.84 0.84 4004 

Weighted 

avg 
0.89 0.89 0.89 4004 



Shahad A. Mnati                                           Al-Khwarizmi Engineering Journal, Vol. 19, No. 4, P.P. 67- 77 (2023) 

 

74 

The general evaluation of the trained model is 
good with acceptable performance represented by 

an accuracy of 89% which may be improved by 

working with more epochs. The tested data showed 
an accuracy of 90% and better performance metrics 

in the colors green, blue, black, white, and purple. 

 

 
 

Fig. 11.  Confusion matrix correlation after testing. 

 

 

Table 2, 

The performance measures the tested model. 

 

4. Conclusion 

 
In this work, a proposed ANN model offers a 

non-parametric, user-friendly, comprehensive 

alternative to traditional statistical methods used in 

classifying RGB colors. About 5000 samples of 
RGB colors were utilized to create the dataset by 

applying measurements to a number of standard 

colored cards through a lab-designed and 
implemented colorimeter. The dataset has been 

preprocessed to eliminate some unidentified 

samples that are considered null data. The 
established neural network has been well set to 

provide a better classification of ANN settings. The 

whole amount of data was split into 80% trained 

data and 20% tested data. The data has been trained 
and validated with a high level of accuracy and 

performance. The results exhibit that the 

classification of unknown color grades can be 
applied with an accuracy of 90% which can be 

 Precision Recall F1score Support 

Red 0.70 0.95 0.81 22 

Green 0.94 0.96 0.95 150 

Blue 0.96 0.93 0.94 109 

Yellow 0.83 0.68 0.75 28 

Orange 0.81 0.72 0.76 18 

Pink 0.85 0.87 0.86 47 

Purple 0.90 0.89 0.90 64 

Brown 0.90 0.79 0.84 34 

Grey 0.80 0.91 0.85 22 

Black 1.00 1.00 1.00 7 

White 1.00 1.00 1.00 4 

     
Accuracy   0.90 505 

Macro avg 0.88 0.88 0.88 505 

Weighted 

avg 
0.90 0.90 0.90 

505 



Shahad A. Mnati                                           Al-Khwarizmi Engineering Journal, Vol. 19, No. 4, P.P. 67- 77 (2023) 

 

75 

extended to more than 11 classes of colors and 
different datasets. The proposed model can be 

utilized as a reference in choosing and identifying 

the proper color in many applications such as car 
painting, by linking the colorimeter with the color 

mixers in order to mix the three primary RGB 

colors into any grade of the eleven previously 
mentioned colors. 

 

 

References 

 
[1] J. Sainui and P. Pattanasatean, “Color 

Classification based on Pixel Intensity 

Values,” 2018 19th IEEE/ACIS International 

Conference on Software Engineering, 
Artificial Intelligence, Networking and 

Parallel/Distributed Computing (SNPD), Jun. 

2018, doi: 10.1109/snpd.2018.8441141. 
[2] A. A. Soofi and A. M. Awan, “Classification 

techniques in Machine Learning: Applications 

and issues,” Journal of Basic and Applied 

Sciences, vol. 13, pp. 459–465, Jan. 2017, doi: 
10.6000/1927-5129.2017.13.76. 

[3] S. Lassri, E. H. Benlahmar, and A. Tragha, “A 

review of Machine Learning Algorithms for 

web page Classification,” 2018 IEEE 5th 

International Congress on Information 

Science and Technology (CiSt), Oct. 2018, 

doi: 10.1109/cist.2018.8596420. 

[4] B. S. Kusumo, A. Heryana, O. Mahendra, and 

H. F. Pardede, “Machine Learning-based for 

Automatic Detection of Corn-Plant Diseases 

Using Image Processing,” 2018 International 

Conference on Computer, Control, 

Informatics and Its Applications (IC3INA), 

Nov. 2018, doi: 

10.1109/ic3ina.2018.8629507. 

[5] S. K. Behera, L. Jena, A. K. Rath, and P. K. 

Sethy, “Disease Classification and Grading of 

Orange Using Machine Learning and Fuzzy 

Logic,” 2018 International Conference on 

Communication and Signal Processing 

(ICCSP), Apr. 2018, doi: 

10.1109/iccsp.2018.8524415. 

[6] R. S. S. Kumari and V. Gomathy, “Fruit 

Classification using Statistical Features in 

SVM Classifier,” International Conference on 

Electrical Energy Systems (ICEES), Feb. 

2018, doi: 10.1109/icees.2018.8442331. 

[7] K. P. Panigrahi, H. Das, A. K. Sahoo, and S. 

C. Moharana, “Maize leaf disease detection 

and classification using machine learning 

algorithms,” in Advances in intelligent 

systems and computing, 2020, pp. 659–669. 

doi: 10.1007/978-981-15-2414-1_66. 

[8] N. Hameed, A. M. Shabut, M. Ghosh, and M. 

A. Hossain, “Multi-class multi-level 

classification algorithm for skin lesions 

classification using machine learning 

techniques,” Expert Systems With 

Applications, vol. 141, p. 112961, Mar. 2020, 

doi: 10.1016/j.eswa.2019.112961. 

[9] T. Nagata, S. Noyori, H. Noguchi, G. 

Nakagami, A. Kitamura, and H. Sanada, “Skin 

tear classification using machine learning 

from digital RGB image,” Journal of Tissue 

Viability, vol. 30, no. 4, pp. 588–593, Nov. 

2021, doi: 10.1016/j.jtv.2021.01.004.  

[10] A. Mary, Z. Bilal, and Z. Saad, “Face 

Recognition and Emotion Recognition from 

Facial Expression Using Deep Learning 

Neural Network,” ResearchGate, Nov. 2020, 

[Online]. Available: 

https://www.researchgate.net/publication/346

007138_Face_Recognition_and_Emotion_Re

cognition_from_Facial_Expression_Using_D

eep_Learning_Neural_Network 

[11] A. H. Mary, A. H. Miry, T. Kara, and M. H. 

Miry, “Nonlinear state feedback controller 

combined with RBF for nonlinear 

underactuated overhead crane system,” 

Journal of Engineering Research 9.3A (2021). 

https://doi.org/10.36909/jer.v9i3A.9159 

[12] E. Arce-Medina and J. I. Paz-Paredes, 

“Artificial neural network modeling 

techniques applied to the hydrodesulfurization 

process,” Mathematical and Computer 

Modelling, vol. 49, no. 1–2, pp. 207–214, Jan. 

2009, doi: 10.1016/j.mcm.2008.05.010. 

[13] J. Maitre, K. Bouchard, and L. P. Bédard, 

“Mineral grains recognition using computer 

vision and machine learning,” Computers & 

Geosciences, vol. 130, pp. 84–93, Sep. 2019, 

doi: 10.1016/j.cageo.2019.05.009. 

[14] C. Kirankaya and L. G. Aykut, “Training of 

artificial neural networks with the multi-

population based artifical bee colony 

algorithm,” Network: Computation in Neural 

Systems, vol. 33, no. 1–2, pp. 124–142, Apr. 

2022, doi: 10.1080/0954898x.2022.2062472. 

[15] D. Hemavathi and H. Srimathi, “Retraction 

Note to: Effective feature selection technique 

in an integrated environment using enhanced 

principal component analysis,” Journal of 

https://www.researchgate.net/publication/346007138_Face_Recognition_and_Emotion_Recognition_from_Facial_Expression_Using_Deep_Learning_Neural_Network
https://www.researchgate.net/publication/346007138_Face_Recognition_and_Emotion_Recognition_from_Facial_Expression_Using_Deep_Learning_Neural_Network
https://www.researchgate.net/publication/346007138_Face_Recognition_and_Emotion_Recognition_from_Facial_Expression_Using_Deep_Learning_Neural_Network
https://www.researchgate.net/publication/346007138_Face_Recognition_and_Emotion_Recognition_from_Facial_Expression_Using_Deep_Learning_Neural_Network


Shahad A. Mnati                                           Al-Khwarizmi Engineering Journal, Vol. 19, No. 4, P.P. 67- 77 (2023) 

 

76 

Ambient Intelligence and Humanized 

Computing, vol. 14, no. S1, p. 535, Jun. 2022, 

doi: 10.1007/s12652-022-04239-4. 

[16] P. V. Chandrika and K. Srinivasan, 

“Predicting stock market movements using 

artificial neural networks,” Universal Journal 

of Accounting and Finance, vol. 9, no. 3, pp. 

405–410, Jun. 2021, doi: 

10.13189/ujaf.2021.090315. 

[17] J. Meiyazhagan, S. Sudharsan, A. Venkatasen, 

and M. Senthilvelan, “Prediction of 

occurrence of extreme events using machine 

learning,” European Physical Journal Plus, 

vol. 137, no. 1, Dec. 2021, doi: 

10.1140/epjp/s13360-021-02249-3. 

 



 (2023) 67-77، صفحة 4، العدد19المجلد  الخوارزمي الهندسيةجلة مشهد علي مناتي                                                                

 

77 

 RGBلوان اللتصنيف أ (ANN)الشبكة العصبونية االصطناعية تطوير نموذج 

 مصنعمجموعة البيانات المستخرجة من مقياس ألوان  باستخدام
 

 **فرات ابراهيم حسين                        *شهد علي مناتي

 ***حمد عيسىأ
الميكاترونكس/ كلية هندسة الخوارزمي/ جامعة بغداد/ بغداد/ العراق*,** قسم هندسة   

 ***قسم هندسة الميكاترونكس واالجهزة الطبية/ مؤسسة الهندسة وتكنولوجيا المعلومات/ جامعة االزهر/ غزة/ فلسطين

shahed.ali1202a@kecbu.uobaghdad.edu.iq :البريد االلكتروني* 

furatnejjar@uobaghdad.edu.iq :البريد االلكتروني** 
a.issa@alazhar.edu.ps :البريد االلكتروني*** 

  

 

 الخالصة

 
من جهاز قياس األلوان الُمصنع في المختبر لبناء ُمصنف مقترح لتصنيف  أخوذة( المRGBزرق )خضر واالحمر واالااللوان اال تم استخدام قيم بيانات

الدراسة وهي األحمر واألخضر  د اجراءعنبناًء على فئات محددة من األلوان األساسية. تم أخذ األلوان األساسية والثانوية والثالثية في االعتبار  جسامألوان اال

( من خالل تطبيق خوارزمية الشبكة العصبونية MLوالبرتقالي واألصفر والوردي واألرجواني واألزرق والبني والرمادي واألبيض واألسود في التعلم اآللي )

 كتسابمن أجل ا RGBشكل أكواد  علىالمذكورة تعريف األلوان األحد عشر  ANN(. تطلب المصنف الذي اعتمد على خوارزمية ANNاالصطناعية )

. تطلب فحص والتحققالبرنامج الذي يعتمد على المصنف المقترح القدرة على التنبؤ بلون الكود الذي ينتمي إلى كائن قيد ال مخرجاتالقدرة على التصنيف. 

ومكتبة  MLلـ  TensorFlowقيمة لونية تم إخضاعها بدورها لخوارزميات التدريب واالختبار. تم استخدام المنصة مفتوحة المصدر  5000العمل جمع حوالي 

 ٪ والتي يمكن اعتبارها90قدرها إلنشاء خوارزمية الدراسة. أظهرت النتائج كفاءة مقبولة للمصنف المبني ممثلة بدقة Kerasالشبكة العصبية مفتوحة المصدر 

 قابلة للتطبيق خاصة بعد بعض التحسينات في األعمال المستقبلية لتكون أكثر فاعلية كمقياس ألوان موثوق به.

 

 

 

mailto:shahed.ali1202a@kecbu.uobaghdad.edu.iq*البريد
mailto:shahed.ali1202a@kecbu.uobaghdad.edu.iq*البريد
mailto:furatnejjar@uobaghdad.edu.iq**البريد
mailto:furatnejjar@uobaghdad.edu.iq**البريد
mailto:a.issa@alazhar.edu.ps***البريد
mailto:a.issa@alazhar.edu.ps***البريد

	2. Materials and Methods
	Fig. 1. The fabricated colorimeter and standard colored cards
	2.2 RGB and Color Spaces
	2.3 ANN Technology
	ANN is essentially a mathematical representation of a process that was created empirically rather than by employing mass and energy balances surrounding the process. A neural network is made up of a network of partially interconnected processing units...
	where ,𝑊-𝑘,𝑖.is the weight of the link between 𝑛𝑜𝑑𝑒 and all the nodes in the preceding layers, ,𝑏-𝑘. is the bias to the node, ,𝑁-𝑗−1. is the number of nodes in layer 𝑗−1, and ,𝑋-𝑘-𝑗−1. signifies the input which in this case is the (RGB)...
	2.4 Performance Metrics
	Various metrics are employed for the estimation of an ANN performance and quality. Performance metrics were considered in this study for evaluating the performance of the ML model. Different metrics are used in evaluating the performance of the model ...
	The second layer, consisting of 24 neurons, contains a total of 96 parameters. These parameters are comprised of 72 weights and 24 biases.
	The experimental data collected from the colorimeter device is employed as an input for the ML algorithm. The ANN model has five layers. It is a densely connected neural network in the dense layer, also known as a fully connected layer, and is a funda...
	Fig. 6. Confusion matrix correlation before training.
	3. Results and Discussion
	Fig. 9. The prediction of the trained model and b) tested model.
	Fig. 10.  Confusion matrix correlation after training.
	4. Conclusion
	In this work, a proposed ANN model offers a non-parametric, user-friendly, comprehensive alternative to traditional statistical methods used in classifying RGB colors. About 5000 samples of RGB colors were utilized to create the dataset by applying me...
	References
	[1] J. Sainui and P. Pattanasatean, “Color Classification based on Pixel Intensity Values,” 2018 19th IEEE/ACIS International Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD), Jun. 2018,...
	[2] A. A. Soofi and A. M. Awan, “Classification techniques in Machine Learning: Applications and issues,” Journal of Basic and Applied Sciences, vol. 13, pp. 459–465, Jan. 2017, doi: 10.6000/1927-5129.2017.13.76.