1. Introduction

Figure 1. Steps in the process of classifying meat texture images.

Figure 1. Steps in the process of classifying meat texture images.

2. Materials and Methods

2.6. k-Nearest Neighbor (k-NN)

The results indicate that the k-nearest neighbour algorithm has the highest accuracy among algorithms for categorizing and comparing k-nearest neighbour patterns based on accuracy [48]. The k-nearest neighbours (k-NN) method is employed to process data [49]. In classification tasks, the class majority of K's nearest neighbours are used by k-NN to derive the class label of an input data point. When utilizing the k-NN approach for classification, the new data's class is determined by comparing it to most of the K nearest neighbours' courses in the training dataset. The K-Nearest Neighbour approach is used to classify things using learning data near the object and the number of nearest neighbours, or the k value. Using the Euclidian distance method, the following formula is typically used to determine a neighbour's proximity or distance.

$$d\left(x,y\right)=\sqrt{{\sum }_{j=1}^n{\left(x_j-y_j\right)}^2}$$

(1)

With:

$d\left(x,y\right)$	: euclidien distance between vector x and y
$d$	: distance testing data to training data
$x_j$	: testing data -j, with j = 1,2,..., n
$y_j$	: training data -j, with j = 1,2,...., n
$n$	: amount of feature.

The distance between each data point in the training dataset and the data point to be predicted is measured as part of the k-NN computation. The following are the primary steps in k-NN calculations:

• The distance metric used to calculate the proximity between data points should be chosen. Euclidean distance is the distance metric that is employed.
• Using the Euclidean distance metric, determine the distance to each data point in the training dataset and the distance for each test data point.
• After computing the distance, find the k-nearest neighbours; see the test data points' k-nearest neighbours based on the most negligible distance value. Sorting the calculated distances and choosing the lowest K value will do this.
• Predicting a class (classification) As the class prediction for the test data point, if the task involves classification, ascertain the majority class of the k-nearest neighbours.

Figure 2. Classification of texture images of meat types using the k-Nearest Neighbor (k-NN) approach.

Figure 2. Classification of texture images of meat types using the k-Nearest Neighbor (k-NN) approach.

2.6.1. Performance Evaluation of Classification Results

The confusion matrix is a valuable tool for assessing classification models' performance in predicting the proper class. The confusion matrix consists of four primary components: False Positive (FP), True Negative (TN), True Positive (TP), and False Negative (FN). As a result of classifying texture photos of several meat varieties, Table 1 displays the classification performance in Confusion matrix form. There are four options based on Table 1 of the confusion matrix: a true positive (TP) is an instance that is positive and classed as positive; a false negative (FN) is an instance that is grouped as unfavourable. A true negative (TN) is an instance that has been declared hostile; a false positive (FP) is an instance that has been declared positive.True Positive (TP): The number of positive observations correctly predicted by the model.

• True Negative (TN): The number of negative observations correctly predicted by the model.
• False Positive (FP): The number of negative observations incorrectly predicted as positive by the model (Type I error).
• False Negative (FN): The number of positive observations incorrectly predicted as negative by the model (Type II error).

2.6.2. Validation

The possibility that a picture is considered to be of good quality and that the image is indeed of good quality is known as sensitivity. Equation 2 can be used to compute sensitivity.

$$\mathit{S}\mathit{e}\mathit{n}\mathit{s}\mathit{i}\mathit{t}\mathit{i}\mathit{v}\mathit{i}\mathit{t}\mathit{y}=\frac{\mathit{T}\mathit{P}}{\mathit{T}\mathit{P}+\mathit{F}\mathit{N}}$$

(2)

Specificity is the likelihood that a picture is perceived as being of low quality and actually of low quality. Equation 3 is used to calculate specificity.

$$\mathit{S}\mathit{p}\mathit{e}\mathit{c}\mathit{i}\mathit{f}\mathit{i}\mathit{c}\mathit{i}\mathit{t}\mathit{y}=\frac{\mathit{T}\mathit{N}}{\mathit{T}\mathit{N}+\mathit{F}\mathit{P}}$$

(3)

By dividing the number of classifications by the number of correct classifications, as shown in Equation 4, one can describe the accuracy of the classification success.

$$\mathit{A}\mathit{c}\mathit{c}\mathit{u}\mathit{r}\mathit{a}\mathit{c}\mathit{y}=\frac{\mathit{T}\mathit{P}+\mathit{T}\mathit{N}}{\mathit{T}\mathit{P}+\mathit{T}\mathit{N}+\mathit{F}\mathit{P}+\mathit{F}\mathit{N}}$$

(4)

2.7. Wavelet Haar

The Discrete Wavelet Transform (DWT), which is based on the Haar wavelet basis function, was used to reduce the dimensionality of the data. The algorithm coefficients and prediction time consumption are better with DWT (wavelet basis function and number of transformation layers are Haar-4, respectively) [50]. One kind of wavelet function utilized in wavelet analysis is the Haar wavelet. The wavelet function is a mathematical tool for simultaneously analyzing and representing texture images of different meat types or data in the time and frequency domains. The Dutch mathematician Alfréd Haar, who first presented this function in 1910, is honoured with the Haar wavelet.

The Haar wavelet is one of the simplest wavelets to comprehend and use because of its discrete and straightforward character. The Haar wavelet function is a local segment function frequently used in data compression, image processing, and image analysis because it can quickly identify variations in texture photographs of different types of meat. The image is separated into smaller intervals for the Haar wavelet transform, and each interval is then subjected to the Haar wavelet. This procedure enables depicting pictures on distinct signal components at varying resolutions. Since Haar wavelets' primary benefit is their ability to deliver high-frequency information effectively, they are frequently employed in applications where quickly identifying edges or changes in the signal is crucial. An approximation coefficient and details-based mathematical formula can be used to explain the Haar wavelet transform. For a discrete signal $\mathit{x}\left[\mathit{n}\right]$ with length $\mathit{N},$ approximation coefficient $\mathit{A}\left[\mathit{k}\right]$ and detail coefficients $\mathit{D}\left[\mathit{k}\right]$ at level k can be calculated as follows:

• Approximation Coefficient:

  $$A\left[k\right]=\frac{x\left[2k\right]+x[2k+1]}{\sqrt{2}}$$

  (2)
• Detailed Coefficients:

  $$D\left[k\right]=\frac{x\left[2k\right]-x[2k+1]}{\sqrt{2}}$$

  (3)

This process can be repeated at each transformation level by replacing $\mathit{x}\left[\mathit{n}\right]$ with $\mathit{A}\left[\mathit{k}\right]$ for the next level. Basic steps to perform the Haar wavelet transform:

• Intervals for texture photos of different kinds of meat are divided into smaller intervals. A convolution operation on neighbouring intervals or the average of two successive values can accomplish this.
• Once the intervals have been separated, compute the approximation coefficient (A) and the detail coefficient (D). The ap-proximation coefficient represents the finer details or low-frequency components in a meat type's texture image. Detail coefficients represent high-frequency elements or information that is coarser or changes more quickly.
• Normalize the coefficient values to suit the requirements of a particular use case. Scale adjustments or weight assignments could be part of this process.

Figure 3. Approximation Haar wavelet of the image of fresh beef types and histogram.

Figure 3. Approximation Haar wavelet of the image of fresh beef types and histogram.

2.8. Gray Level Co-Occurrence Matrix (GLCM)

Selecting GLCM features is essential for improving sample differentiation. [51]. The Gray Level Co-Occurrence Matrix (GLCM) is a texture analysis tool used in digital image processing. GLCM looks at the spatial connection between grayscale image pixel intensities to extract information about image patterns and textures. GLCM measures the frequency at which pairs of pixels with a given intensity appear together at a given distance and direction. By calculating these matrices and analyzing the texture of meat texture photos, numerous statistics can be produced, such as energy, contrast, correlation, homogeneity, and entropy. For a GLCM element C (i, j) with horizontal direction and distance 1, the general formula is:

$$C\left(i,j\right)={\sum }_{m=1}^M{\sum }_{n=1}^N\delta (I\left(m,n\right)=idanI\left(m,n+1=j\right)$$

(4)

Where:

• $I\left(m,n\right)$ is the image's pixel intensity at coordinates (m, n).
• $\delta \left(.\right)$ is a delta function that returns one if the statement in parentheses is accurate and 0 otherwise.
• M is the number of image rows.
• N is the number of image columns.

Basic GLCM computation for the horizontal direction at a distance of one from the grayscale image that follows:

$$\left[\begin{array}{cc}1,& 1,\end{array}\begin{array}{cc}2,& 3\end{array}\right]$$

$$\left[\begin{array}{cc}2,& 2,\end{array}3\begin{array}{cc},& 4\end{array}\right]$$

$$\left[\begin{array}{cc}1,& 1,\end{array}\begin{array}{cc}2,& 2\end{array}\right]$$

$$\left[3\begin{array}{cc},& 4,\end{array}4\begin{array}{cc},& 4\end{array}\right]$$

The calculation steps are as follows:

• Select horizontal direction and distance 1.
• Count the pairs of pixels that appear together:
  • Scan the image to identify pairs of pixels with matching intensities.
  • Pixel pairs that appear together are: (1, 1), (1, 2), (2, 3), (3, 4), (2, 2), (3, 3), (4, 4 ), (1, 1), (1, 2), (2, 3), (3, 4), (2, 2), (3, 3), (4, 4).
• Create GLCM Matrix:
  • Count the occurrences of pixel pairs and insert them into the GLCM matrix.

    $$\left[\begin{array}{cc}2,& 3,\end{array}\begin{array}{cc}2,& 1\end{array}\right]$$

    $$\left[\begin{array}{cc}1,& 2,\end{array}2\begin{array}{cc},& 3\end{array}\right]$$

    $$\left[\begin{array}{cc}0,& 1,\end{array}\begin{array}{cc}2,& 2\end{array}\right]$$

    $$\left[1\begin{array}{cc},& 0,\end{array}0\begin{array}{cc},& 3\end{array}\right]$$
• GLCM Matrix Normalization:
  • Normalize the matrix to obtain a probability distribution.
  $$\left[\begin{array}{cc}0.1,& 0.15,\end{array}0.1\begin{array}{cc},& 0.05\end{array}\right]$$

  $$\left[\begin{array}{cc}0.05,& 0.1,\end{array}0.1\begin{array}{cc},& 0.15\end{array}\right]$$

  $$\left[\begin{array}{cc}0,& 0.05,\end{array}0.1\begin{array}{cc},& 0.1\end{array}\right]$$

  $$\left[0.05\begin{array}{cc},& 0,\end{array}0\begin{array}{cc},& 0.15\end{array}\right]$$

Figure 4. GLCM horizontal of the image of fresh beef types and histogram.

Figure 4. GLCM horizontal of the image of fresh beef types and histogram.

For a GLCM element C (i, j) with vertical direction and distance 1, the general formula will depend on the definition of the GLCM matrix and the specific direction used. Assume we are working with grayscale images with intensity levels between 0 and $L-1,$where L is the number of intensity levels. The general formula for calculating GLCM elements C (i, j) with vertical direction and distance 1 is:

$$C\left(i,j\right)={\sum }_{m=1}^{M-1}{\sum }_{n=1}^N\delta (I\left(m,n\right)=idanI\left(m+1,n=j\right)$$

(5)

Where:

• $I\left(m,n\right)$ is the image's pixel intensity at coordinates (m, n).
• $\delta \left(.\right)$ is a delta function that returns one if the statement in parentheses is accurate and 0 otherwise.
• M is the number of image rows.
• N is the number of image columns.

Simple GLCM calculation for the vertical direction with a distance of 1 from the following grayscale image:

$$\left[\begin{array}{cc}1,& 1,\end{array}\begin{array}{cc}2,& 3\end{array}\right]$$

$$\left[\begin{array}{cc}2,& 2,\end{array}3\begin{array}{cc},& 4\end{array}\right]$$

$$\left[\begin{array}{cc}1,& 1,\end{array}\begin{array}{cc}2,& 2\end{array}\right]$$

$$\left[3\begin{array}{cc},& 4,\end{array}4\begin{array}{cc},& 4\end{array}\right]$$

The calculation steps are as follows:

• Select horizontal direction and distance 1.
• Count the pairs of pixels that appear together:
  • Scan the image to identify pairs of pixels with matching intensities.
  • Pixel pairs that appear together are: (1, 1), (2, 2), (1, 1), (3, 4), (2, 2), (1, 1), (4, 4), (2, 2), (2, 2), (3, 3), (4, 4), (2, 2), (2, 2), (4, 4).
• Create GLCM Matrix:
  • Count the occurrences of pixel pairs and insert them into the GLCM matrix.
  $$\left[3\begin{array}{cc},& 1,\end{array}3\begin{array}{cc},& 0\end{array}\right]$$

  $$\left[\begin{array}{cc}0,& 4,\end{array}0\begin{array}{cc},& 2\end{array}\right]$$

  $$\left[\begin{array}{cc}2,& 1,\end{array}\begin{array}{cc}2,& 1\end{array}\right]$$

  $$\left[0\begin{array}{cc},& 2,\end{array}0\begin{array}{cc},& 2\end{array}\right]$$
• GLCM Matrix Normalization:
  • Normalize the matrix to obtain a probability distribution.
  $$\left[\begin{array}{cc}0.1875,& 0.625,\end{array}0.1875\begin{array}{cc},& 0\end{array}\right]$$

  $$\left[\begin{array}{cc}0,& 0.25,\end{array}0\begin{array}{cc},& 0.125\end{array}\right]$$

  $$\left[\begin{array}{cc}0.125,& 0.0625,\end{array}0.125\begin{array}{cc},& 0.0625\end{array}\right]$$

  $$\left[0\begin{array}{cc},& 0.125,\end{array}0\begin{array}{cc},& 0.125\end{array}\right]$$

Figure 5. GLCM vertical of the image of fresh beef types and histogram.

Figure 5. GLCM vertical of the image of fresh beef types and histogram.

For GLCM elements C (i, j) with diagonal direction and distance 1, the general formula can be formulated as follows:

$$C\left(i,j\right)={\sum }_{m=1}^{M-1}{\sum }_{n=1}^{N-1}\delta (I\left(m,n\right)=idanI\left(m+1,n+1=j\right)$$

(6)

Where:

• $I\left(m,n\right)$ is the image's pixel intensity at coordinates (m, n).
• $\delta \left(.\right)$ is a delta function that returns one if the statement in parentheses is accurate and 0 otherwise.
• M is the number of image rows.
• N is the number of image columns.

Simple GLCM calculation for the diagonal direction with a distance of 1 from the following grayscale image:

$$\left[\begin{array}{cc}1,& 1,\end{array}\begin{array}{cc}2,& 3\end{array}\right]$$

$$\left[\begin{array}{cc}2,& 2,\end{array}3\begin{array}{cc},& 4\end{array}\right]$$

$$\left[\begin{array}{cc}1,& 1,\end{array}\begin{array}{cc}2,& 2\end{array}\right]$$

$$\left[3\begin{array}{cc},& 4,\end{array}4\begin{array}{cc},& 4\end{array}\right]$$

The calculation steps are as follows:

• Select horizontal direction and distance 1.
• Count the pairs of pixels that appear together:
  • Scan the image to identify pairs of pixels with matching intensities.
  • Pixel pairs that appear together are: (1, 1), (1, 1), (2, 2), (3, 3), (2, 2), (3, 3), (4, 4), (1, 1), (1, 1), (2, 2), (3, 3), (2, 2), (3, 3), (4, 4).
• Create GLCM Matrix:
  • Count the occurrences of pixel pairs and insert them into the GLCM matrix.
  $$\left[3\begin{array}{cc},& 0,\end{array}0\begin{array}{cc},& 0\end{array}\right]$$

  $$\left[\begin{array}{cc}0,& 4,\end{array}0\begin{array}{cc},& 0\end{array}\right]$$

  $$\left[\begin{array}{cc}0,& 0,\end{array}\begin{array}{cc}3,& 0\end{array}\right]$$

  $$\left[0\begin{array}{cc},& 0,\end{array}0\begin{array}{cc},& 3\end{array}\right]$$
• GLCM Matrix Normalization:
  • Normalize the matrix to obtain a probability distribution.
  $$\left[\begin{array}{cc}0.375,& 0,\end{array}0\begin{array}{cc},& 0\end{array}\right]$$

  $$\left[\begin{array}{cc}0,& 0.5,\end{array}0\begin{array}{cc},& 0\end{array}\right]$$

  $$\left[\begin{array}{cc}0,& 0,\end{array}0.375\begin{array}{cc},& 0\end{array}\right]$$

  $$\left[0\begin{array}{cc},& 0,\end{array}0\begin{array}{cc},& 0.375\end{array}\right]$$

Figure 6. GLCM diagonal of the image of fresh beef types and histogram.

Figure 6. GLCM diagonal of the image of fresh beef types and histogram.

A clear explanation of how the Gray Level Co-Occurrence Matrix (GLCM) matrix is utilized to explain the relationships between pixels in meat texture photos may be found in your analysis of the GLCM computation stages. Here are a few more points:

• Direction and Distance Selection
  The selection of direction and distance holds significance as it influences the measurement of the relationship between pixels. This has an impact on the texture information that is taken from the picture.
• GLCM Matrix Calculation
  Counting the occurrences of particular pairs of pixel intensities at specific distances and directions is necessary to calculate a GLCM matrix. The end product is a matrix displaying the frequency of particular pairs of pixels occurring together.
• Matrix Normalization
  Matrix normalisation is applied to determine the probability distribution of pixel pair appearances. This probability distribution can calculate the likelihood that a pair of pixels will appear about the image's overall size.
• Feature Extraction from Matrix Normalization
  After normalisation, numerous texture properties, including energy, contrast, correlation, homogeneity, and entropy, can be retrieved from the matrix. Every element offers distinct details regarding the textural characteristics of the picture
• Texture Statistics and Characteristics
  Energy calculations can see the degree to which pixel intensity approaches a given value. The difference in the intensities of neighbouring pixels is referred to as contrast. The degree of correlation between pixel brightness within a specific distance and direction is reflected in correlation. The degree of uniformity in pixel intensity distribution across the image is known as homogeneity. The degree of uncertainty in the pixel intensity distribution is measured by entropy.

Following these procedures creates a set of attributes that can be utilized to identify or describe meat texture photographs more accurately. This method has several uses in pattern recognition and image analysis, particularly in image processing for meat type identification and classification.

Figure 7. Texture image classification of types of meat using k-NN, Wavelet Haar, and GLCM approaches.

Figure 7. Texture image classification of types of meat using k-NN, Wavelet Haar, and GLCM approaches.

The k-NN, Wavelet Haar, and GLCM methodologies can be used to identify several stages and procedures in the classification of textural photographs of different varieties of meat. First, the colour photos of the various beef types were transformed to grayscale as the first step in the processing procedure. Image decomposition utilizing Haar wavelets with smoothing and reduction on lines and columns constitutes the second feature extraction stage using GLCM and Haar wavelets. Creates four images: HH (detail components), LH, HL, and LL (approximation). Each detailed element's average, standard deviation and tendency are computed statistically on deconstructed images. For every deconstructed image, GLCM normalization produces a normalized GLCM. GLCM analysis, or GLCM, to extract features or textural properties, including energy, contrast, correlation, homogeneity, and entropy, is the third feature extraction method from GLCM. Sorting texture photos of different kinds of meat into fresh, frozen, and rotten categories is the fourth classification procedure. Class or label predictions are made using Euclidean and confusion matrices. The final method involves assessing performance using evaluation metrics, i.e., determining how effectively the model can predict the proper class on test data using measures like accuracy. Its accuracy is how closely a prediction resembles the actual or expected value. The six classification outcomes are based on picture classification. Precisely, fresh, frozen, and rotting are the three categories into which different meat varieties' texture picture categorization findings are divided. Using this method, the model can comprehend and identify patterns or characteristics that distinguish different meat varieties' textures in pictures. Metrics like accuracy are used in performance evaluation to show how effectively the model predicts the proper class on test data. Using the classification results, meat varieties can subsequently be categorized according to their conditions.

3. Results and Discussion

3.1. Experiment Results

The size of each fresh, frozen, and rotten meat class was confirmed by analyzing 750 meat texture samples from five different types of goat, horse, pork, and beef images. The pixels in each meat image are unique. The absence of scaling when taking meat samples results in variations in pixel size between images. Especially in high-resolution photography, scaling can result in visual distortion, reduced image quality, and increased computational effort [52,53]. Experiments were conducted to ensure that different pixel sizes for each type and meat quality were carefully studied to provide consistency and accuracy in test results and avoid distortion or conflict due to variations in image quality. Based on pixel intensity and pixel frequency, it is simulated that the sample size and histogram distribution in class (a) datasets are, on average, more significant and more dominant than in class (b) and class (c) datasets. The frequency and intensity distribution of pixels in an image, as well as the histogram distribution, determine the image's appearance, and this is the main focus of the analysis performed by various algorithms.

Table 2. An illustration of the histogram distribution and the meat texture experiment findings for classes (a) fresh meat, class (b) frozen meat, and class (c) rotten meat. There are 50 samples of fresh meat, 50 samples of frozen meat, and 50 samples of rotten meat for each type of meat in each class.

Table 2. An illustration of the histogram distribution and the meat texture experiment findings for classes (a) fresh meat, class (b) frozen meat, and class (c) rotten meat. There are 50 samples of fresh meat, 50 samples of frozen meat, and 50 samples of rotten meat for each type of meat in each class.

This research K-NN was created using the Google Colab platform. A GPU, or graphics processing unit, was used in each experiment. Two approaches are usually taken. In the first stage, two techniques—GLCM and the Haar wavelet algorithm—are used to extract features used to identify meat images. In the second approach, meat images are determined directly from tracked meat features (feature extraction) using k-NN, which has been trained on various image domains. The feature set is then fed into a k-NN model to see if the classification performance is sufficient to separate texture images (a) from texture images (b) and texture images (c).

3.2. Feature Selection Result

Five different types of meat images, beef, buffalo, goat, horse, and pork, were used to test the feature selection findings. Each type of meat image was classified into three classes: fresh, frozen, and rotten. Each meat is described in Table 3 using GLCM metrics, including homogeneity, contrast, correlation, energy, and entropy. It's crucial to remember that every flesh texture image has the same pixel size, which is 500x500 [56,57]. The choice to employ uniform pixel sizes across all flesh texture images may have been made with an eye toward optimizing computer power. It can lower the computing load and expedite the data analysis process by employing a consistent pixel size. This choice has a number of drawbacks, though, including a lack of spatial resolution and difficulty accurately capturing any scale variations that might exist among the photos. The texture of each meat image can still be learned a great deal by using GLCM metrics, which include contrast, correlation, energy, homogeneity, and entropy. When comparing the textural properties of fresh, frozen, and rotten meat from different kinds of animals under observation, this GLCM measure can be helpful. Thus, despite limitations in spatial resolution, GLCM analysis can still provide valuable insights into meat texture recognition and classification. Therefore, GLCM analysis can still offer valuable insights into the detection and classification of flesh texture despite its limitations in spatial resolution.

Table 3 illustrates how well all feature sets distinguish between the three classes (a), (b), and (c). For fresh meat images, the average ranges from 51.52% to 183.21%, while for frozen meat images, the average ranges from 78.25% to 185.75%. Rotten meat images, on the other hand, have an average range of 34.62% to 115.79%.

The picture of fresh meat demonstrates how fresh pigs have a less diverse texture than other meats. In contrast, higher averages for all recorded GLCM measurements suggest that the intensity variation of fresh goat meat was more significant. Significant variations in the intensity and texture structure of fresh meat images from various meat varieties were shown by this experiment. These differences are crucial for categorizing and identifying pictures and understanding the textural characteristics of various meat kinds.

In the meantime, the frozen meat image reveals that the frozen pork texture image has the lowest average. This indicates that the texture of frozen pig has less variation in intensity than that of frozen goat meat. Frozen goat meat, on the other hand, displayed more notable intensity differences. This report provides a summary of the textural characteristics of frozen goat flesh. This information can be helpful in both learning more about the differences between different slices of meat and identifying or classifying images of frozen beef.

However, the picture of the rotting meat demonstrates that, in most cases, the texture difference between rotten pork and rotten beef is less severe. The intensity difference was more significant for the decaying flesh. This study advances our knowledge of the textural characteristics of pictures of rotting beef. This information can distinguish between various states of meat and identify or classify the stages of rotten meat [58,59].

Table 4 presents the validation results for all aspects of class (a), (b), and (c) using k-NN classification. It is expected that (A) sensitivity, (B) specificity, (C) accuracy, and (D) Matthews correlation function well. For fresh meat images, the coefficients have percentages ranging from (A): 97.959% to 100%; (B): 96.078% to 100%; (C): 97% to 99%; and (D): 94.019% to 98.02%. Meanwhile, images of frozen beef with percentages ranging between 96.078% to 100% in (A), 97.959% to 100% in (B), 97% to 99% in (C), and 94.019% to 98.02% in ( D)). Meanwhile, the percentage of rotten meat images is as follows: (A): 92.308% to 96%; (B): 95.833% to 97.917%; (C): 94% to 96%; and (D): 88.07% to 92.074%) [60,61]. Findings from analyses of (A), (B), (C), and (D) fresh, frozen, and rotten meat images offer essential information about how well k-NN classification models can categorize meat states. Based on the texture image.

Fresh Meat: The model can correctly detect fresh meat images for both categories, as evidenced by (A) having the most significant proportion and being more dominant in fresh goat and fresh horse meat texture images. (B) have different quality percentages. With the highest accuracy images for fresh meat texture, the model can correctly differentiate fresh beef from frozen or rotten beef. This shows how well the model can differentiate fresh meat images from other settings. C) The model can classify fresh meat images correctly; fresh beef and fresh horse meat texture images are the most accurate. This shows how well the model can recognize fresh meat images and differentiate them from other conditions. Images of fresh beef and fresh horse meat (D) are more common in texture images and have the highest percentage (D). This shows a good agreement between the feature model and the actual classes in the fresh meat environment for the classification results. On the other hand, the percentage (D) in the texture image of fresh buffalo meat shows worse performance.

Frozen Meat: The tendency of the model to accurately recognize frozen pork images is demonstrated by the fact that the model has (A) the highest percentage of predominance over frozen buffalo meat texture images. On the other hand, the texture image of frozen buffalo meat has (A) the lowest rate, indicating a problem with identifying frozen buffalo meat. The decreasing tendency of the model to correctly classify frozen buffalo meat photos is shown by (B), which displays the lowest percentage of all frozen buffalo meat texture images. In contrast, the texture images of frozen beef and frozen horse meat received the most significant percentage (B), indicating that the model performs better in accurately classifying these images. (C) The frozen buffalo meat texture image has the lowest accuracy, while the frozen pork texture image has the best dominant accuracy. This shows that the model classifies frozen pork images more accurately than buffalo meat images. However, this algorithm performed very well when classifying frozen meat images. When comparing the texture images of frozen goat meat and frozen pork, the percentage of frozen beef with (D) is the highest. This shows how well and accurately the algorithm can classify texture images of frozen mutton and pork. On the other hand, (D) the percentage of frozen buffalo meat texture images shows worse performance.

Rotting Meat: shows the model's skill in identifying images of rotting goat meat, with (A) the highest percentage being more dominant in texture images. However, in the picture, the texture of rotten buffalo meat has (A) the lowest percentage, so it is challenging to distinguish rotten buffalo meat. (B) has the lowest rate for texture images of rotten buffalo meat and rotten horse meat; this shows the difficulty of effectively recognizing the images of these two types of Meat. However, the texture images of rotten beef and rotten pork have (B) the highest percentage, which shows that the model is more accurate in classifying images of rotten Meat and rotten pork. (C) The texture image of rotten goat meat, rotten pork and rotten buffalo meat has the highest accuracy, while the texture image of rotten beef has the lowest accuracy. This shows that for the model to accurately classify images of rotting buffalo meat, the model must be more precise. However, the images of rotten beef, rotten goat meat, and rotten pork can be recognized correctly through this experiment. But overall, this system works well in categorizing cases of rotting flesh. (D) texture description of rotten pork, rotten goat meat, and rotten beef. Rotten Meat is more common and has the best percentage (D). The model performs well in classifying images of rotting flesh; however, the lowest rate of images was associated with the texture of rotting horse meat (D).

Table 5 shows how well the three models compare in terms of comparison results of performance measurements of the k-nearest neighbor (k-NN) algorithm, the Haar Wavelet algorithm, and the Gray Level Co-occurrence Matrix (GLCM) for the three classes (fresh, frozen, and frozen). Frozen). Rotten). (A) k-NN, (B) Wavelet Haar and (C) GLCM assumed. First, for the fresh meat texture image class, (A): 97% to 99%; (B): 89.25% to 89.96%; and (C): 51.52% to 183.21%. The percentage of frozen meat in both images is 97% to 99% in (A), 87.56% to 88.25% in (B), and 78.25% to 185.75% in (C), third, with the percentage ranges from 94%. Up to 96% in (A), 86.26% to 87.97% in (B), and 34.62% to 115.79% in (C) for the rotting flesh image.

The performance results of algorithms (A), (B), and (C) on fresh, frozen, and rotten meat images provide valuable insight into the performance of algorithm models in determining the condition of Meat based on texture images. Fresh Meat First: (A) The k-NN algorithm gives the best results on texture images of fresh goats, horses and cattle. It also has the most significant standout score (99%). This shows how well the k-NN algorithm performs when classifying fresh meat images. Regarding the texture image of fresh buffalo meat, k-NN has the worst performance, with a score of 97%. Meanwhile, (B) the texture image of fresh pork shows that the performance of the Haar Wavelet algorithm is worse (89.25%) than k-NN. However, when it comes to fresh beef texture images, our algorithm performs well (89.96%). Meanwhile, for the texture image of fresh pork (C), the GLCM method gave inconsistent results, with the lowest application percentage (51.52%). However, GLCM data can still be used for classification. However, GLCM has the most significant proportion of fresh goat meat texture images (183.21%). Of the three algorithms studied, the GLCM, Wavelet Haar, and k-NN models performed the best in categorizing fresh meat texture images into meat categories. This shows that the k-NN technique is more suitable for applications that organize texture images of Fresh Meat.

Second Frozen Meat: (A) Using the k-NN method, the texture image of frozen pork gives the most prominent value (99%), while the texture image of frozen buffalo meat gives the lowest result (97%). This shows how well the k-NN algorithm classifies frozen meat images. However, compared with k-NN, the Haar Wavelet algorithm performs worst (87.56%) on frozen beef texture images. However, the algorithm continues to produce good results. 88.25 per cent) is the highest performance in frozen beef texture images. Based on the GLCM algorithm, the texture image of frozen pork (C) has the lowest value, 78.25%. However, GLCM results can still be used to classify frozen meat images. After three rounds of testing, k-NN maintained its top ranking in frozen meat texture image classification, with Wavelet Haar and GLCM ranking second and third. This shows that applications that require frozen meat texture image classification are more suitable for using the k-NN technique. GLCM was the best model in performance testing, with an accuracy of 185.75%.

Third, Rotten Meat: (A) has the lowest percentage (94%), although the texture images of rotten beef, rotten goat, and rotten pig have the highest k-NN values (96%). This demonstrates how effectively the k-NN algorithm classifies pictures of rotting flesh. While (B) k-NN outperforms the Haar Wavelet algorithm (86.26%), there is an image of decaying buffalo flesh. However, the algorithm continues to yield good results. Conversely, (C) The GLCM approach yields erratic results; the textural image of rotten pork has the lowest performance % (34.62). However, GLCM data can still be used to categorize images of rotting flesh. However, the rotten beef texture image had the highest performance percentage (115.79%). After testing three different algorithms, k-NN, GLCM, and Wavelet, Haar continued to produce the best classification results for images containing rotting meat textures. This shows that when classifying rotting meat texture images, the k-NN approach is more suitable.

Table 6. Classification accuracy based on meat texture analysis.

Table 6. Classification accuracy based on meat texture analysis.

Author	Structure	Texture Analysis Method (Features)	Method	Accuracy (%)
Yudhana, Anton Umar, Rusydi Saputra, Sabarudin[36]	Fish	RGB colors and GLCM features	k-NN	94%
Don Africa, Aaron M Claire Alberto, Stephanie T Evan Tan, Travis Y[62]	Beef and pork	Skewness, Kurtosis, Mean, and Std Deviation	k-NN	98.6%
Wijaya, Dedy Rahman Sarno, Riyanarto Zulaika, Enny[63]	beef	Regression results (black: actual, blue: prediction, red: prediction with error	Discrete Wavelets Transform and Long Short-Term Memory (DWTLSTM) dan k-nearest neighbour (k-NN)	85,05%
Kiswanto, Hadiyanto, and Eko Sediyon[2]	Beef, buffalo, goat, horse and pork	RGB, GLCM and HSV	Haar wave algorithm	76.72%
Ayaz, Hamail Ahmad, Muhammad Mazzara, Manuel Sohaib, Ahmed[64]	Meat	HSI	k-NN	82%

5. Conclusions

References

1. S. Ayu Aisah, A. Hanifa Setyaningrum, L. Kesuma Wardhani, and R. Bahaweres, “Identifying Pork Raw-Meat Based on Color and Texture Extraction Using Support Vector Machine,” 2020 8th Int. Conf. Cyber IT Serv. Manag. CITSM 2020, no. c, 2020. [CrossRef]
2. and E. S. Kiswanto, Hadiyanto, Modification of the Haar Wavelet Algorithm for Texture Identification of Types of Meat Using Machine. Springer Nature Singapore, 2024. [CrossRef]
3. H. H. H, S. Madenda, and S. Widiyanto, “Comparison of Beef Marbling Segmentation by Experts towards Computational Techniques by Using Jaccard , Dice and Cosine Comparison of Beef Marbling Segmentation by Experts towards Computational Techniques by Using Jaccard , Dice and Cosine,” vol. 12, no. 13, pp. 623–627, 2021.
4. H. H. Handayani and A. F. N. Masruriyah, “Determination of beef marbling based on fat percentage for meat quality,” Int. J. Psychosoc. Rehabil., vol. 24, no. 1, pp. 8394–8401, 2020. [CrossRef]
5. K. Adi, S. Pujiyanto, O. D. Nurhayati, and A. Pamungkas, “Beef quality identification using color analysis and k-nearest neighbor classification,” Proc. - 2015 4th Int. Conf. Instrumentation, Commun. Inf. Technol. Biomed. Eng. ICICI-BME 2015, pp. 180–184, 2016. [CrossRef]
6. S. M. K. Uddin, M. A. M. Hossain, Z. Z. Chowdhury, and M. R. Bin Johan, “Short targeting multiplex PCR assay to detect and discriminate beef, buffalo, chicken, duck, goat, sheep and pork DNA in food products,” Food Addit. Contam. - Part A Chem. Anal. Control. Expo. Risk Assess., vol. 38, no. 8, pp. 1273–1288, 2021. [CrossRef]
7. V. Sciences et al., “Review Article Identiication of Diferent Animal Species in Meat and Meat Products: Trends and Advances,” vol. 3, no. 6, pp. 334–346, 2015.
8. L. S. Dalsecco, R. M. Palhares, P. C. Oliveira, L. V. Teixeira, M. G. Drummond, and D. A. A. de Oliveira, “A Fast and Reliable Real-Time PCR Method for Detection of Ten Animal Species in Meat Products,” J. Food Sci., vol. 83, no. 2, pp. 258–265, 2018. [CrossRef]
9. Y. C. Li et al., “Comparative review and the recent progress in detection technologies of meat product adulteration,” Compr. Rev. Food Sci. Food Saf., vol. 19, no. 4, pp. 2256–2296, 2020. [CrossRef]
10. M. A. M. Hossain et al., “Quantitative Tetraplex Real-Time Polymerase Chain Reaction Assay with TaqMan Probes Discriminates Cattle, Buffalo, and Porcine Materials in Food Chain,” J. Agric. Food Chem., vol. 65, no. 19, pp. 3975–3985, 2017. [CrossRef]
11. F. M. Albkosh, M. S. Hitam, W. N. J. H. Wan Yussof, A. A. K. Abdul Hamid, and R. Ali, “Optimization of discrete wavelet transform features using artificial bee colony algorithm for texture image classification,” Int. J. Electr. Comput. Eng., vol. 9, no. 6, pp. 5253–5262, 2019. [CrossRef]
12. X. Sun et al., “Predicting beef tenderness using color and multispectral image texture features,” Meat Sci., vol. 92, no. 4, pp. 386–393, 2012. [CrossRef]
13. P. Jackman, D. W. Sun, P. Allen, N. A. Valous, F. Mendoza, and P. Ward, “Identification of important image features for pork and turkey ham classification using colour and wavelet texture features and genetic selection,” Meat Sci., vol. 84, no. 4, pp. 711–717, 2010. [CrossRef]
14. M. M. Ávila et al., “Magnetic Resonance Imaging, texture analysis and regression techniques to non-destructively predict the quality characteristics of meat pieces,” Eng. Appl. Artif. Intell., vol. 82, no. March, pp. 110–125, 2019. [CrossRef]
15. L. Chen, “Accepted manuscript to appear in IJWMIP Accepted Manuscript International Journal of Wavelets, Multiresolution and Information Processing,” 2017. [CrossRef]
16. R. Wijaya, “Noise filtering framework for electronic nose signals: An application for beef quality monitoring,” Comput. Electron. Agric., vol. 157, pp. 305–321, 2019. [CrossRef]
17. M. Kishore, A. Issac, N. Minhas, and B. Sarkar, “Image processing based method to assess fi sh quality and freshness,” J. Food Eng., vol. 177, pp. 50–58, 2016. [CrossRef]
18. H. Pu, A. Xie, D. W. Sun, M. Kamruzzaman, and J. Ma, “Application of Wavelet Analysis to Spectral Data for Categorization of Lamb Muscles,” Food Bioprocess Technol., vol. 8, no. 1, pp. 1–16, 2015. [CrossRef]
19. M. Masoumi, M. Marcoux, L. Maignel, and C. Pomar, “Weight prediction of pork cuts and tissue composition using spectral graph wavelet,” J. Food Eng., vol. 299, no. January, p. 110501, 2021. [CrossRef]
20. R. Wijaya, R. Sarno, E. Zulaika, and S. I. Sabila, “Development of mobile electronic nose for beef quality monitoring,” Procedia Comput. Sci., vol. 124, pp. 728–735, 2017. [CrossRef]
21. K. Song, S. hui Wang, D. Yang, and T. yu Shi, “Combination of spectral and image information from hyperspectral imaging for the prediction and visualization of the total volatile basic nitrogen content in cooked beef,” J. Food Meas. Charact., vol. 15, no. 5, pp. 4006–4020, 2021. [CrossRef]
22. I.G. Sujana and E. Putra, “Classification of Tuna Meat Grade Quality Based on Color Space Using Wavelet and k-Nearest Neighbor Algorithm,” no. November, pp. 2–4, 2023.
23. V. Amin, D. Wilson, G. Rouse, and S. Udpa, “Multiresoi , Utional Texture Analysis for Ultrasound Tissue,” vol. 14, pp. 201–215, 1998.
24. Y. Tao, “Wavelet-based adaptive thresholding method for image segmentation,” Opt. Eng., vol. 40, no. 5, p. 868, 2001. [CrossRef]
25. M. K. Dutta, A. Issac, N. Minhas, and B. Sarkar, “Image processing based method to assess fish quality and freshness,” J. Food Eng., vol. 177, pp. 50–58, 2016. [CrossRef]
26. N. D. Kim, V. Amin, D. Wilson, G. Rouse, and S. Udpa, “Ultrasound image texture analysis for characterizing intramuscular fat content of live beef cattle,” Ultrason. Imaging, vol. 20, no. 3, pp. 191–205, 1998. [CrossRef]
27. R. A. Asmara et al., “Classification of pork and beef meat images using extraction of color and texture feature by Grey Level Co-Occurrence Matrix method,” IOP Conf. Ser. Mater. Sci. Eng., vol. 434, no. 1, 2018. [CrossRef]
28. S. Widiyanto, “Texture Feature Extraction Based On GLCM and DWT for Beef Tenderness Classification,” 2018 Third Int. Conf. Informatics Comput., pp. 1–4.
29. R. A. Asmara et al., “Chicken meat freshness identification using colors and textures feature,” 2018 Jt. 7th Int. Conf. Informatics, Electron. Vis. 2nd Int. Conf. Imaging, Vis. Pattern Recognition, ICIEV-IVPR 2018, pp. 93–98, 2019. [CrossRef]
30. M. M. Santoni, D. I. Sensuse, A. M. Arymurthy, and M. I. Fanany, “Cattle Race Classification Using Gray Level Co-occurrence Matrix Convolutional Neural Networks,” Procedia Comput. Sci., vol. 59, no. Iccsci, pp. 493–502, 2015. [CrossRef]
31. M. E.-H. . H. M. E. . H. M. . & N. M. Ibrahim, “Muzzle feature extraction based on gray level co-occurrence matrix,” Isr. J. Vet. Med., vol. 01, pp. 16–24, 2016.
32. R. Farinda, Z. Firmansyah, C. Sulton, I. G. P. S. Wijaya, and F. Bimantoro, “Beef Quality Classification based on Texture and Color Features using SVM Beef Quality Classification based on Texture and Color Features using SVM Classifier,” no. September, 2018. [CrossRef]
33. S. Agustin and R. Dijaya, “Beef Image Classification using K-Nearest Neighbor Algorithm for Identification Quality and Freshness,” J. Phys. Conf. Ser., vol. 1179, no. 1, 2019. [CrossRef]
34. B. Jia, W. Wang, S. C. Yoon, H. Zhuang, and Y. F. Li, “Using a combination of spectral and textural data to measure water-holding capacity in fresh chicken breast fillets,” Appl. Sci., vol. 8, no. 3, 2018. [CrossRef]
35. S. Suhadi, P. Dina Atika, S. Sugiyatno, A. Panogari, R. Trias Handayanto, and H. Herlawati, “Mobile-based fish quality detection system using k-nearest neighbors method,” 2020 5th Int. Conf. Informatics Comput. ICIC 2020, 2020. [CrossRef]
36. A.Yudhana, R. Umar, and S. Saputra, “Fish Freshness Identification Using Machine Learning: Performance Comparison of k-NN and Naïve Bayes Classifier,” J. Comput. Sci. Eng., vol. 16, no. 3, pp. 153–164, 2022. [CrossRef]
37. H. Jiang, W. Yuan, Y. Ru, Q. Chen, J. Wang, and H. Zhou, “Feasibility of identifying the authenticity of fresh and cooked mutton kebabs using visible and near-infrared hyperspectral imaging,” Spectrochim. Acta - Part A Mol. Biomol. Spectrosc., vol. 282, no. July, 2022. [CrossRef]
38. A.P. A. da C. Barbon et al., “Development of a flexible Computer Vision System for marbling classification,” Comput. Electron. Agric., vol. 142, no. November, pp. 536–544, 2017. [CrossRef]
39. S. I. Sabilla, R. Sarno, K. Triyana, and K. Hayashi, “Deep learning in a sensor array system based on the distribution of volatile compounds from meat cuts using GC–MS analysis,” Sens. Bio-Sensing Res., vol. 29, no. July, 2020. [CrossRef]
40. J. A. Matera et al., “Discrimination of Brazilian artisanal and inspected pork sausages: Application of unsupervised, linear and non-linear supervised chemometric methods,” Food Res. Int., vol. 64, pp. 380–386, 2014. [CrossRef]
41. T. Gaber, A. Tharwat, A. E. Hassanien, and V. Snasel, “Biometric cattle identification approach based on Weber’s Local Descriptor and AdaBoost classifier,” Comput. Electron. Agric., vol. 122, pp. 55–66, 2016. [CrossRef]
42. Y. Xiong et al., “Non-Destructive Detection of Chicken Freshness Based on Electronic Nose Technology and Transfer Learning,” Agric., vol. 13, no. 2, 2023. [CrossRef]
43. A. González-Mohino, T. Pérez-Palacios, T. Antequera, J. Ruiz-Carrascal, L. S. Olegario, and S. Grassi, “Monitoring the processing of dry fermented sausages with a portable NIRS device,” Foods, vol. 9, no. 9, pp. 1–12, 2020. [CrossRef]
44. A. Varghese, S. Jain, M. Jawahar, and A. A. Prince, “Auto-pore segmentation of digital microscopic leather images for species identification,” Eng. Appl. Artif. Intell., vol. 126, no. PC, p. 107049, 2023. [CrossRef]
45. V. Fernández-Ibáñez, T. Fearn, A. Soldado, and B. de la Roza-Delgado, “Development and validation of near infrared microscopy spectral libraries of ingredients in animal feed as a first step to adopting traceability and authenticity as guarantors of food safety,” Food Chem., vol. 121, no. 3, pp. 871–877, 2010. [CrossRef]
46. S. Kumar, S. K. Singh, R. Singh, and A. K. Singh, “Animal biometrics: Techniques and applications,” Anim. Biometrics Tech. Appl., pp. 1–243, 2018. [CrossRef]
47. T. Chen, X. Qi, M. Chen, and B. Chen, “Gas Chromatography-Ion Mobility Spectrometry Detection of Odor Fingerprint as Markers of Rapeseed Oil Refined Grade,” J. Anal. Methods Chem., vol. 2019, 2019. [CrossRef]
48. Najam ul Hasan, N. Ejaz, W. Ejaz, and H. S. Kim, “Meat and fish freshness inspection system based on odor sensing,” Sensors (Switzerland), vol. 12, no. 11, pp. 15542–15557, 2012. [CrossRef]
49. A.V. Shik et al., “Rapid Testing of Irradissation Dose in Beef and Potatoes by Reaction-Based Optical Sensing Technique,” J. Food Compos. Anal., vol. 127, no. August 2023, p. 105946, 2023. [CrossRef]
50. Z. S. Jiang, Rongchang, Rongchang Jiang, Jingxin Shen, Xinran Li, Rui Gao, Qinghe Zhao, “Detection and recognition of veterinary drug residues in beef using hyperspectral discrete wavelet transform and deep learning,” Int. J. Agric. Biol. Eng., vol. 15, no. 1, pp. 224–232, 2022. [CrossRef]
51. C. Malegori, L. Franzetti, R. Guidetti, E. Casiraghi, and R. Rossi, “GLCM , an image analysis technique for early detection of bio fi lm,” J. Food Eng., vol. 185, pp. 48–55, 2016. [CrossRef]
52. P. P. S. Parsania and D. P. V.Virparia, “A Review: Image Interpolation Techniques for Image Scaling,” Int. J. Innov. Res. Comput. Commun. Eng., vol. 02, no. 12, pp. 7409–7414, 2015. [CrossRef]
53. Fedorov, T. Utkina, O. Nechyporenko, and Y. Korpan, “Development of technique for face detection in image based on binarization, scaling and segmentation methods,” Eastern-European J. Enterp. Technol., vol. 1, no. 9–103, pp. 23–31, 2020. [CrossRef]
54. C. DeCarli, D. G. M. Murphy, D. Teichberg, G. Campbell, and G. S. Sobering, “Local histogram correction of MRI spatially dependent image pixel intensity nonuniformity,” J. Magn. Reson. Imaging, vol. 6, no. 3, pp. 519–528, 1996. [CrossRef]
55. S. H. Bae and M. Kim, “A novel SSIM index for image quality assessment using a new luminance adaptation effect model in pixel intensity domain,” 2015 Vis. Commun. Image Process. VCIP 2015, pp. 5–8, 2015. [CrossRef]
56. J. J. Ã and W. C. Ro, “Data processing and image reconstruction methods for pixel detectors The first transmission radiogram was measured by,” vol. 576, pp. 223–234, 2007. [CrossRef]
57. H. Wu and Z. Li, “Scale Issues in Remote Sensing: A Review on Analysis, Processing and Modeling,” pp. 1768–1793, 2009. [CrossRef]
58. Y. Yang, W. Wang, H. Zhuang, S. C. Yoon, and H. Jiang, “Fusion of spectra and texture data of hyperspectral imaging for the prediction of the water-holding capacity of fresh chicken breast filets,” Appl. Sci., vol. 8, no. 4, 2018. [CrossRef]
59. B. Park, M. Kise, W. R. Windham, K. C. Lawrence, and S. C. Yoon, “Textural analysis of hyperspectral images for improving contaminant detection accuracy,” Sens. Instrum. Food Qual. Saf., vol. 2, no. 3, pp. 208–214, 2008. [CrossRef]
60. N. Ali, D. Neagu, and P. Trundle, “Evaluation of k - nearest neighbour classifier performance for heterogeneous data sets,” SN Appl. Sci., no. September, 2019. [CrossRef]
61. A.J. Gallego, J. R. Rico-juan, and J. J. Valero-mas, “Efficient k -nearest neighbor search based on clustering and adaptive k values,” Pattern Recognit., vol. 122, p. 108356, 2022. [CrossRef]
62. A.M. Don Africa, S. T. Claire Alberto, and T. Y. Evan Tan, “Development of a portable electronic device for the detection and indication of fireworks and firecrackers for security personnel,” Indones. J. Electr. Eng. Comput. Sci., vol. 19, no. 3, pp. 1194–1203, 2020. [CrossRef]
63. D.R. Wijaya, R. Sarno, and E. Zulaika, “DWTLSTM for electronic nose signal processing in beef quality monitoring,” Sensors Actuators, B Chem., vol. 326, no. September 2020, p. 128931, 2021. [CrossRef]
64. H. Ayaz, M. Ahmad, M. Mazzara, and A. Sohaib, “Hyperspectral imaging for minced meat classification using nonlinear deep features,” Appl. Sci., vol. 10, no. 21, pp. 1–13, 2020. [CrossRef]