IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 13, No. 2, June 2024, pp. 2441~2450

ISSN: 2252-8938, DOI: 10.11591/ijai.v13.i2.pp2441-2450  2441

Classification of nutmeg ripeness using artificial intelligence

Imam Bil Qisthi, Hartono Siswono

Department of Electrical Engineering, Faculty of Industrial Technology, Gunadarma University, Jakarta, Indonesia

Article Info ABSTRACT

Article history: Nutmeg seeds can produce a lot of oil if they have optimal maturity. In other
words, they have little moisture content. Based on observations made at one
Received Aug 15, 2023 of the refineries in Sukabumi, farmers do not pay attention to the maturity
Revised Nov 6, 2023 level of nutmeg seeds after drying which can cause a decrease in the quality
Accepted Dec 14, 2023 of nutmeg seeds and the quality of the oil produced. This study aims to make
it easier for nutmeg farmers to classify the maturity of nutmeg seeds. This
study used the convolutional neural network (CNN) method to help with
Keywords: classification problems and several image processing methods. This program
will be run through an Android application. When the application containing
Android this CNN model is run, the camera system will turn on, and the program will
Classifications classify in real-time nutmeg objects into 1 of 3 class labels namely
Convolutional neural network LowQuality, MidQuality, or HighQuality class labels classifying. The results
Image processing will be displayed on the application screen, the results are displayed in the
Nutmeg seeds form of class names and scores. The results of CNN model training accuracy
are 97.92%.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Imam Bil Qisthi
Department of Electrical Engineering, Faculty of Industrial Technology, Gunadarma University
Pesona Kahuripan 5 B7/36, Cilengsi, West Java, Indonesia
Email: [email protected]

1. INTRODUCTION
Nutmeg is one of the plants originating from Indonesia with a high value. The plant with the Latin
name myristica fragrans is classified as a type of spice plant and has the potential benefits of almost all parts
of the nutmeg plant. In addition to nutmeg pulp which can be utilized as candied nutmeg, nutmeg syrup,
pickles, jams, and dodol in the culinary field, myristica fragrans has been widely used for traditional
medicine to treat several diseases, besides myristica fragrans has the potential as an antimicrobial
antioxidant, anti-inflammatory, antiulcer, anticancer, aphrodisiac, and various other activities because it
contains various phytochemicals such as lignans, neolignans, diphenylalkanes, phenylpropanoids, and
terpenoid [1]–[4]. The nutmeg plant parts, namely seeds, mace, and nutmeg oil, also have high economic
value and are export commodities because Indonesia is the largest nutmeg producer in the world, where most
nutmeg plantations in Indonesia are cultivated by community plantations and the rest by large plantations.
According to Indonesian plantation development statistics, nutmeg production in Indonesia in 2021 reached
39,577 tons [5]. Nutmeg essential oil contains up to 80% trimyristin, and trimyristin is a whitening agent for
the skin or whitening agent. Nutmeg, especially nutmeg, is a source of essential oil with high economic value
because this part contains around 20% to 40% fixed oil [6].
Indonesia is the largest producer and exporter of nutmeg on the world market. Sukabumi is one of
the regencies in West Java that has the largest nutmeg plantation, with an area of 1,679 ha. Based on the
results of observations, we made at one of the distillery in the Sukabumi area, nutmeg farmers did not pay
attention to the maturity level of nutmeg seeds after drying which resulted in nutmeg seeds with low maturity
mixed with nutmeg seeds that had an optimal maturity level. This has become a common problem for nutmeg

Journal homepage: http://ijai.iaescore.com

2442  ISSN: 2252-8938

refiners. The oil content in nutmeg plants is most found in nutmeg parts of nutmeg seeds, the oil in nutmeg
seeds will be maximum when the nutmeg seeds have a little moisture content or in other words maximum
drying. The drying process is one way to reduce the moisture content of nutmeg seeds to reach the standard.
According to Jaiyeoba et al. [7], the moisture content will decrease as the maturity level of the seeds
increases, which means that the harvest time or maturity level of the fruit will affect the quality of nutmeg
after drying. Ignorance of farmers about the importance of harvesting and post-harvest processes that can
cause a decrease in the quality of nutmeg.
This research is important to make it easier for nutmeg oil refiners to classify the maturity of nutmeg
seeds. This study uses the convolutional neural network (CNN) method as one of the deep learning
technologies to help with classification problems. This deep learning uses a simpler network scheme with
fewer connections that are computationally much more efficient [8]. CNN was chosen because of research
conducted by Bogar et al. [9]. Seeing the rapid development of mobile technology, this research will utilize
Android mobile technology to facilitate the community, especially nutmeg oil refining entrepreneurs [9].

2. RELATED WORKS
CNN is an algorithm that already exists but is widely developed by several researchers in the field of
deep learning. This study is inseparable from previous research as a reference. Several studies have focused
on the development of CNNs across various objects.
Melinda et al. [9] introduces a mobile phone application capable of differentiating individuals with
autism spectrum disorder (ASD) from those with typical brain signals, based on asynchronous EEG brain
signals. It develops a preprocessing algorithm and utilizes the BCI2000 EEG data signal, making the process
automated using Python. Furthermore, the study employs a deep learning CNN as the output model, deployed
using Python-Flask. This enables the diagnosis of EEG signals for ASD and normal patients to be accessible
across various platforms through a REST API. Bacus and Linsangan [10] classify diseases in papaya leaves
using MobileNet's CNN architecture. The study used 1,394 image data and managed to obtain a high
accuracy average of 91.667%. There are things that need to be underlined in this study, namely that the
dataset used is an imbalanced dataset so that there is the potential for overfitting the model.
Research by Paulson and Ravishankar [11] is aimed at identifying types of herbal plants using
artificial intelligence or deep learning technology. There are 64 types of herbal plants that will be identified
by the CNN model and the visual geometry group (VGG) pre-train model version 16 and VGG version 19.
The results of this study give CNN a good accuracy of about 95.79%, VGG16 of 97.8%, and VGG19 of
9.6%. Research by Roslan et al. [12] purposed to investigate the performance of CNN on dataset of herbal
plants as medicine. The dataset used is a dataset of herbal plants on the Island of Pinang, Malaysia with
original data and augmented additional data. The results of this study testing the model with the original data
resulted in 75% accuracy and testing the model with the original data and additional data resulted in 88%
accuracy. So it can be concluded from this research that the accuracy of the model can increase with the
amount of data.
Wang et al. [13] detecting diseases and pest infections in plants using CNN architecture called ultra-
lightweight efficient network. This architecture consists of 2 part modules, namely the feature extraction
module that adopts residual depth-wise convolution and the qualification module that accepts multi-scale
features enhanced by spatial pyramid pooling layer. As a result, the model with this architecture produces
good results and has a lightweight performance. Wiryana et al. [14] classify store products by using the CNN
algorithm. The test is conducted with 1,050 product images divided into 35 labels and divided into three data,
namely 80% data training, 10% data validation, and 10% data test. The image used is preprocessed with a
size of 256×256 pixels. The data was trained with six convolution layers and an epoch of 50 with an
execution time of 33 minutes so as to achieve an accuracy of 91.37%.
Desai et al. [15] classify flower types by using the CNN algorithm. In this study using CNN
architecture, namely VGG19 as an extractor feature from flower image data. The results of this study
obtained 100% accuracy for training and 91.1% for validation for 17 flower type classes. Dyrmann et al. [16]
classify plant species by using deep convolutional neural network (DCNN) algorithm. The network was
engineered from scratch by being trained and tested on a complete of 10,413 pictures containing 22 weeds
and plant species within the early stages of growth. For these 22 species, this network is in a position to
realize a classification accuracy of 86.2%.
Rathi et al. [17] classify fish species by using DCNN algorithm and image processing methods. The
image processing methods used are Otsu binarization, dilation, and erosion. The results of this study obtained
an accuracy of 96.29%. Liew et al. [18] classify gender by using CNN algorithm. This study used publicly
available datasets namely SUM and AT&T. The input image size used was 32×32. The accuracy results on
each dataset were 98.75% for SUM and 99.38% for AT&T.

Int J Artif Intell, Vol. 13, No. 2, June 2024: 2441-2450

Int J Artif Intell ISSN: 2252-8938  2443

Vishnupriya and Meenakshi [19] classified music genres using neural network (NN) algorithms and
mel frequency cepstral coeficient (MFCC) feature extraction. The dataset in this study covers ten different
genres. The result of this study obtained 76% accuracy for training. Lu et al. [20] classify fruits by using
CNN algorithm. Designed CNN using six layers consisting of convolution layers, pooling layers, and fully
connected layers. The result of this study obtains 91.44% accuracy better than three state-of-the art: Support
vector machine, wavelet entropy, and genetic algorithm. Razali et al. [21] classifying nutrient deficiencies in
oil palms on leaves using the CNN algorithm. The study used 180 datasets and used several CNN
architectures. The results of this study Alexnet became an efficient architecture with few layers.

3. METHOD
The method will explain in detail about the general description of the application and the stages in
preparing data for the model in this study. The stages of preparing image data start from collection until the
data can be used by the model including resizing data, augmentation of image data, and processing of image
data. So that it can be understood for development in the future.

3.1. Overview of the application

The nutmeg seed maturity classification application will be run through a smartphone device after
modeling using CNN. CNN is a deep learning model widely used for image or visual analysis purposes [22].
CNN is a development of multilayer perceptron designed to process image data [23]. The CNN structure
consists of input layer, output layer, and hidden layer. Hidden layers include convolutional layers, pooling
layers, rectified linear unit (ReLU), fully connected layers, and loss layers [24]. A general overview of the
application to be designed will be shown in Figure 1.
Figure 1 describes in general the application we designed, when the application containing the CNN
model is run or opened, the camera system will turn on, the model will read in real-time nutmeg objects.
Then, the model will classify nutmeg seeds into 1 label from 3 class labels namely LowQuality, MidQuality,
or HighQuality. After being classified the results will be displayed on the smartphone screen. The results that
will be displayed are in the form of class name and score.

Figure 1. General overview of the application

3.2. Image datasets

The imagery data used was collected manually using the phone's camera and food photo box. Data is
collected with several provisions in data retrieval to get the best data. The first provision, nutmeg seeds used
are nutmeg seeds with an age of 3 months because nutmeg seeds at that age contain a lot of oil so it is good
for distillation [2]. Second, nutmeg seeds are dried out naturally using sunlight with an average temperature
of 38 °C measured using a thermometer device. Third, nutmeg seeds are dried for 8 hours for 5 days starting
at 08.00 to 16.00. Fourth, nutmeg image data collection was carried out at the end of the drying day, from

Classification of nutmeg ripeness using artificial intelligence (Imam Bil Qisthi)

2444  ISSN: 2252-8938

day 0 (after peeling) to day 5. Fifth, the distance between the camera and the object at the time of image data
collection is 30 cm. These provisions are obtained based on the results of direct observation. The image data
that was successfully collected with the existing provisions was 240 image data. The collected image data
will not be displayed all, only one image data for each class as shown in Figure 2. Figure 2(a) is a label of
LowQuality class, Figure 2(b) is a label of MidQuality class, and Figure 2(c) is a label of HighQuality class.
The data labeling process in this study is based on direct observations with nutmeg seed oil refining business
actors located in Sukabumi, Indonesia.

(a) (b) (c)

Figure 2. Image for (a) LowQuality, (b) MidQuality, and (c) HighQuality

Furthermore, the data will be processed so that it can be used by the model such as resizing,
multiplying images, and removing the background. Image data will be resized to 224×224 size, multiply
image data with augmentation techniques, and remove background using image image data segmentation.
Thus, it is expected that the model can work more efficiently with smaller image sizes but there is still
information that can be learned and will not process all image pixels only segmented pixels.

3.2.1. Resize data image

The data that has been successfully collected will be resized because the CNN model uses 224×224
as input. The image data collected has an original size of 4,032×3,024. An illustration of the image data
resizing process is shown in Figure 3. In Figure 3, the original image data of 4,032×3,024 will be converted
using the resize function to a size of 224×224. The purpose of this stage is to make the model work
efficiently because it only processes images with a size of 224×224 and homogenizes the image size when
there are images that have different sizes. Image size can be converted using the resize function.

Figure 3. Image data resizing process

Furthermore, the image data will be performed in the augmentation process. The general purpose of
this stage is to multiply image data by modifying the image so that it is considered different by the computer.
Augmentation of images in two ways, namely rotating and flipping horizontally so that they are perceived by
the computer as different image.

3.2.2. Image data augmentation

Image data augmentation aims to modify the image so that the modified image will be considered a
different image by the CNN model, thus the image data we have will multiply [25]. Image augmentation was
carried out in this study because the data used was limited, so it was necessary to do augmentation to represent
the data. The image data augmentation process used is rotated with an angle of 359 degrees and flipped
horizontally. The augmented image data will not be displayed all, only one will be displayed as shown in

Int J Artif Intell, Vol. 13, No. 2, June 2024: 2441-2450

Int J Artif Intell ISSN: 2252-8938  2445

Figure 4. Figure 4(a) is the original image before the augmentation process, Figure 4(b) is a horizontal flip
augmentation image, and Figure 4(c) is a rotated augmentation image with an angle of 359 degrees.

(a) (b) (c)

Figure 4. Image for (a) original image, (b) horizontal flip image, and (c) rotate image

After going through the image data augmentation stage, originally the image data amounted to 240
data, now it has become 720 image data. The 720 images data consists of 240 original data, 240 rotate data,
and 240 horizontal flip data. Next, the data will be segmented to separate parts of the image object from the
background and limit the image area to be processed by the model.

3.2.3. Image data segmentation

Data segmentation aims to separate the image object and background parts of the image data to limit
other parts that do not need to be processed, such as the background, to increase efficiency in image data
processing by the model because only the foreground is processed. Image data segmentation, the first process
carried out is image data segmentation using the Otsu segmentation method. Furthermore, the second will be
carried out morphological process based on the results of Otsu segmentation using the erosion method to
obtain binary images. Finally, finding the outermost and largest contours using the region of interest (ROI)
method based on the results of morphological processes or binary images so that the image object is
separated from the background. The segmented image data will not be displayed all, only one will be
displayed as shown in Figure 5. Figure 5(a) is the original image before the segmentation process,
Figure 5(b) is a segmented image using the Otsu method, Figure 5(c) is an inverted image to obtain a binary
image on the erosion method, Figure 5(d) is a binary image of the result of the erosion morphology method,
and Figure 5(e) is the image of finding the outermost using the ROI method.

(a) (b) (c) (d) (e)

Figure 5. Image for (a) original image, (b) segmented image, (c) inverted image, (d) binary image,
and (e) ROI image

3.2.4. Model training and testing

The CNN architecture model used in this research is CNN MobileNet architecture, CNN MobileNet
architecture is used because one of the objectives of this research is to implement the CNN model that has
been made into Android devices. MobileNet architecture is a CNN architecture compatible with android
devices so MobileNet was chosen in this research. After the image dataset is augmented and segmented, the
image dataset will be used to train the model and test the model with a data proportion of 80% training data
or 576 image data from 720 image data with the same amount of data from each nutmeg category and 20%
testing data or 144 image data from 720 image data.

4. RESULTS AND DISCUSSION

4.1. Data processing results
The data processing results include segmentation using Otsu, image morphology segmentation
through erosion, and finding the outermost contour using the ROI method to retrieve image objects.
Furthermore, if the image and background objects have separate image data will be used as datasets for the
CNN model. The results for each class can be seen in Table 1. It appears that some parts of the data are being
Classification of nutmeg ripeness using artificial intelligence (Imam Bil Qisthi)
2446  ISSN: 2252-8938

wasted in Table 1, particularly in the LowQuality category. It seems that the segmentation process is not
perfect, which is causing some object parts to be lost. To improve the results of data processing, it may be
necessary to refine the segmentation process.

Table 1. Data processing results

Classes Result
LowQuality

MidQuality

HighQuality

4.2. Results from creating and training the model

This section will discuss the performance and training results of the CNN classification model
created and trained using the data processing results presented in Table 1. The training model results are
presented as accuracy and loss validation training graphs. These graphs are shown in Figure 6. Figure 6 is a
graph showing that model performance has good results because the training validation accuracy graph
shows the accuracy of each epoch moving close to 1 or 100%, precisely at 0.9792 or 97.92%. Then, the
training and validation loss graph shows a bit of data that the model cannot recognize marked with a
descending graph at each epoch. So, it can be concluded that the model that has been made gets good results.

Figure 6. The training model results

The results of the model's performance in classifying are shown in the confusion matrix shown in
Figure 7. Figure 7 is a confusion matrix that describes the performance of the model. This confusion matrix
can be known as the accuracy value, precision value, recall value, and F1-score value. The confusion matrix
in Figure 7 is then calculated to determine the accuracy value, precision value, recall value, and F1-score
value. These values are used to find out whether the model that has been created is a good model. The
calculation results are shown in Tabel 2.
Based on Table 2, it can be concluded. First, the model has high accuracy for test data, reaching
100% for all classes. This shows that the model can classify data accurately. Secondly, the precision value on
the model for all classes reaches 100%. This indicates that of all the positive predictions made by the model,
all of them are correct, and none of the predictions are wrong. Third, the recall value on the model is also
very good for the "LowQuality" and "MidQuality" classes, with a value of 100%. As for the high class, recall

Int J Artif Intell, Vol. 13, No. 2, June 2024: 2441-2450

Int J Artif Intell ISSN: 2252-8938  2447

reached 92.3%, indicating that most samples included in the "HighQuality" class were identified by the
model. Fourth, all classes have a high F1-score, reaching 100% for the "LowQuality" and "MidQuality"
classes and 95.9% for the "HighQuality" classes. Fifth, the model shows uniform performance and good
performance in all classes, without either class having lower performance compared to the others.

Figure 7. Confusion matrix

Table 2. The results of the calculations confusion matrix

Confusion matrix LowQuality MidQuality HighQuality
True negative 6 13 12
True positive 0 0 0
False positive 0 0 1
False negative 0 0 0
Accuracy 1.0 (100%) 1.0 (100%) 1.0 (100%)
Precision 1.0 (100%) 1.0 (100%) 1.0 (100%)
Recall 1.0 (100%) 1.0 (100%) 0.923 (92.3%)
F1-Score 1.0 (100%) 1.0 (100%) 0.959 (95.9%)

Based on the existing points, it can be concluded that the model has a very good ability to classify
into three different classes, consistent performance and high matrix evaluation values show that this model
can be relied upon in performing classification tasks on the given data. However, it is important to remember
that these results are based on the data that has been provided, and model performance may vary on different
data. This requires further testing using data that is completely separate from the training data to validate the
performance of the model and ensure that the model has good generalization capabilities.

4.3. Model implementation test results

Model implementation test results are an analysis of model test results that have been successfully
implemented into Android applications. Tests are carried out using new nutmeg image data to test whether
the model that has been created and implemented can classify if new data is present. This testing phase is a
very important stage because it determines whether the model can be used according to the purpose of this
study. This stage will display new test data that was correctly classified by the model, and new test data that
was incorrectly classified by the model. Test results of the correct data classified by the model are shown in
Table 3. Table 3 is the test result of new test data with models that have been implemented into the Android
application. The results show that the model can classify correctly. Things that must be underlined in testing
using various camera distances with objects, various backgrounds, various angles, and a lot of data with a
good score.
The results of testing the model are not entirely correct, there are still misclassified data. This is
natural because when viewed from the accuracy itself is not perfect. Test results of data incorrectly classified
by the model are shown in Table 4. Table 4 is the test result of new test data with models that have been
implemented into the Android application. The results showed that the model was wrong in classifying some
new data. This is reasonable because the accuracy of the model itself is not perfect.

Classification of nutmeg ripeness using artificial intelligence (Imam Bil Qisthi)

2448  ISSN: 2252-8938

Table 3. Test results of the correct data classified by the model

Result Actual classes Predict classes Description
LowQuality LowQuality Succeeded with an accuracy 100%

MidQuality MidQuality Succeeded with an accuracy 98%

HighQuality HighQuality Succeeded with an accuracy 100%

Table 4. Test results of the wrong data classified by the model

Result Actual classes Predict classes Description
LowQuality HighQuality Fail

HighQuality LowQuality Fail

LowQuality MidQuality Fail

4.4. Application test results

Application test results are application tests that have been made into several Android devices with
different specifications. The purpose of this test is to assess the extent to which the application can work well
with devices that have limited specifications. The test results are shown in Table 5. Table 5 is the test result
of Android applications that have been made into several Android devices with various specifications.
Performance is a parameter to measure whether the application can run properly on various types of Android
systems, the display is a parameter to measure whether the application has a responsive display in various
screen sizes, and model is a parameter to measure whether, with different camera specifications, the model
can still classify nutmeg.
After conducting several series of tests on the models and applications that have been made, it can
be concluded that the model is quite good at classifying the maturity level of nutmeg seeds. The model can be
implemented into Android applications so that it can make it easier for potential users to classify to minimize

Int J Artif Intell, Vol. 13, No. 2, June 2024: 2441-2450

Int J Artif Intell ISSN: 2252-8938  2449

errors and make time efficiency. However, in addition to the success in making models and implementing
models into Android applications, there are obstacles such as models that are wrong in classifying and do not
rule out the possibility that the model will continue to be developed both structurally and how to process
data.

Table 5. Application test result

Applications
Device Specifications Models
Perform Display
Samsung A33 Main camera 48 MP, screen size 6.4, RAM 8 GB, storage 128 GB, versi Running Good Can classify
version Android 12 data
Samsung galaxy Main camera 48 MP, screen size 6.4, RAM 6 GB, Storage 128 GB, Running Good Can classify
m22 version Android 11 data
Samsung j2 prime Main camera 8 MP, screen size 5.0, RAM 1.5 GB, storage 8 GB, version Running Good Can classify
Android 6.0 data

5. CONCLUSION
Based on the results of the design and testing in this study, it can be concluded that the creation of a
model that can classify the maturity level of nutmeg seeds using artificial intelligence has been successfully
carried out, and the implementation of the model into the android application has been successfully carried
out. The training accuracy result of 97.92% indicates that the CNN model is effective in recognizing the
features contained in the image, this is a pretty good result. The results of testing on models and applications
with new test data also showed good results, the model was able to classify new test data into one of three
existing classes. Then, for testing the application also gets good results, the application can run properly on
several Android devices with different specifications. From this result, it can be developed again such as
improving the results of data processing because there are still some parts of the object wasted due to the
image data segmentation process so it needs to be refined again for further research because the part of the
object must have wasted information. Then there are new test data that are misclassified so that future
research can refine the model structure and multiply the training data so that it can improve the level of
accuracy. Research can be further developed by implementing classification models into devices other than
Android so that it can help in the industrial field.

REFERENCES
[1] V. Nikolic et al., “Chemical composition, antioxidant and antimicrobial activity of nutmeg (myristica fragrans houtt.) seed
essential oil,” Journal of Essential Oil Bearing Plants, vol. 24, no. 2, pp. 218–227, Mar. 2021, doi:
10.1080/0972060X.2021.1907230.
[2] C. R. Kholibrina and A. Aswandi, “The aromatherapy formulation of essential oils in reducing stress and blood pressure on
human,” IOP Conference Series: Earth and Environmental Science, vol. 914, no. 1, pp. 1-7, Nov. 2021, doi: 10.1088/1755-
1315/914/1/012072.
[3] S. Das et al., “Assessment of chemically characterised myristica fragrans essential oil against fungi contaminating stored scented
rice and its mode of action as novel aflatoxin inhibitor,” Natural product research, vol. 34, no. 11, pp. 1611–1615, Jun. 2020, doi:
10.1080/14786419.2018.1519826.
[4] M. T. Ha, N. K. Vu, T. H. Tran, J. A. Kim, M. H. Woo, and B. S. Min, “Phytochemical and pharmacological properties of
myristica fragrans houtt.: an updated review,” Archives of Pharmacal Research, vol. 43, no. 11, pp. 1067–1092, Nov. 2020, doi:
10.1007/s12272-020-01285-4.
[5] E. Kembauw et al., “Cultivation system and marketing chain of nutmeg in East Seram District, Maluku Province, Indonesia,”
International Journal of Multidisciplinary Sciences and Arts, vol. 1, no. 2, pp. 134–139, Jan. 2023, doi:
10.47709/ijmdsa.v1i2.2015.
[6] A. Trifan, G. Zengin, I. Korona-Glowniak, K. Skalicka-Woźniak, and S. V. Luca, “Essential oils and sustainability: in vitro
bioactivity screening of myristica fragrans houtt. post-distillation by-products,” Plants, vol. 12, no. 9, pp. 1-16, Apr. 2023, doi:
10.3390/plants12091741.
[7] K. F. Jaiyeoba, C. A. Ogunlade, O. S. Kwanaki, and O. K. Fadele, “Moisture dependent physical properties of nutmeg (myristica
fragrans) relevant for design of processing machines,” Current Journal of Applied Science and Technology, vol. 39, no. 12, pp.
74–85, Jun. 2020, doi: 10.9734/cjast/2020/v39i1230665.
[8] Y. Wang et al., “A survey on deploying mobile deep learning applications: a systemic and technical perspective,” Digital
Communications and Networks, vol. 8, no. 1, pp. 1–17, Feb. 2022, doi: 10.1016/j.dcan.2021.06.001.
[9] M. Melinda, F. Arnia, A. Yafi, N. A. C. Andryani, and I. K. A. Enriko, “Design and implementation of mobile application for
CNN-based EEG identification of autism spectrum disorder,” International Journal on Advanced Science, Engineering, vol. 14,
no. 1, pp. 57–64, Feb. 2024, doi: 10.18517/ijaseit.14.1.19676.
[10] J. A. Bacus and N. B. Linsangan, “Detection and identification with analysis of carica papaya leaf using android,” Journal of
Advances in Information Technology, vol. 13, no. 2, pp. 162–166, 2022, doi: 10.12720/jait.13.2.162-166.
[11] A. Paulson and S. Ravishankar, “AI based indigenous medicinal plant identification,” in 2020 Advanced Computing and
Communication Technologies for High Performance Applications (ACCTHPA), Jul. 2020, pp. 57–63, doi:
10.1109/ACCTHPA49271.2020.9213224.
[12] N. A. M. Roslan, N. M. Diah, Z. Ibrahim, Y. Munarko, and A. E. Minarno, “Automatic plant recognition using convolutional
neural network on malaysian medicinal herbs: the value of data augmentation,” International Journal of Advances in Intelligent
Classification of nutmeg ripeness using artificial intelligence (Imam Bil Qisthi)
2450  ISSN: 2252-8938

Informatics, vol. 9, no. 1, pp. 136-147, Mar. 2023, doi: 10.26555/ijain.v9i1.1076.

[13] B. Wang, C. Zhang, Y. Li, C. Cao, D. Huang, and Y. Gong, “An ultra-lightweight efficient network for image-based plant disease
and pest infection detection,” Precision Agriculture, vol. 24, no. 5, pp. 1836–1861, 2023, doi: 10.1007/s11119-023-10020-0.
[14] I. M. Wiryana, S. Harmanto, A. Fauzi, I. B. Qisthi, and Z. N. Utami, “Store product classification using convolutional neural
network,” IAES International Journal of Artificial Intelligence, vol. 12, no. 3, pp. 1439–1447, 2023, doi:
10.11591/ijece.v12i3.pp1439-1447.
[15] S. Desai, C. Gode, and P. Fulzele, “Flower image classification using convolutional neural network,” in 2022 First International
Conference on Electrical, Electronics, Information and Communication Technologies (ICEEICT), Feb. 2022, pp. 1–4, doi:
10.1109/ICEEICT53079.2022.9768635.
[16] M. Dyrmann, H. Karstoft, H. S. Midtiby, “Plant species classification using deep convolutional neural network,” Biosystems
engineering, 2016, doi: 10.1016/j.biosystemseng.2016.08.024.
[17] D. Rathi, S. Jain, and S. Indu, “Underwater fish species classification using convolutional neural network and deep learning,” in
2017 Ninth International Conference on Advances in Pattern Recognition (ICAPR), 2017, pp. 1–6, doi:
10.1109/ICAPR.2017.8593044.
[18] S. S. Liew, M. K. -Hani, S. A. Radzi, and R. Bakhteri, “Gender classification: a convolutional neural network approach,” Turkish
Journal of Electrical Engineering and Computer Sciences, vol. 24, pp. 1248–1264, 2016, doi: 10.3906/elk-1311-58.
[19] S. Vishnupriya and K. Meenakshi, “Automatic music genre classification using convolution neural network,” in 2018
International Conference on Computer Communication and Informatics (ICCCI), Jan. 2018, pp. 1–4, doi:
10.1109/ICCCI.2018.8441340.
[20] S. Lu, Z. Lu, S. Aok, and L. Graham, “Fruit classification based on six layer convolutional neural network,” in 2018 IEEE 23rd
International Conference on Digital Signal Processing (DSP), Nov. 2018, pp. 1–5, doi: 10.1109/ICDSP.2018.8631562.
[21] M. I. H. Razali, M. A. Hairuddin, A. H. Jahidin, M. H. M. Som, and M. S. A. M. Ali, “Classification of nutrient deficiency in oil
palms from leaf images using convolutional neural network,” IAES International Journal of Artificial Intelligence, vol. 11, no. 4,
pp. 1314-1322, Dec. 2022, doi: 10.11591/ijai.v11.i4.pp1314-1322.
[22] P. Kalgaonkar and M. El-Sharkawy, “CondenseNeXt: an ultra-efficient deep neural network for embedded systems,” in 2021
IEEE 11th Annual Computing and Communication Workshop and Conference (CCWC), Jan. 2021, pp. 0524–0528, doi:
10.1109/CCWC51732.2021.9375950.
[23] R. M. James, Kusrini, and M. R. Arief, “Classification of x-ray covid-19 image using convolutional neural network,” in 2020 2nd
International Conference on Cybernetics and Intelligent System (ICORIS), Oct. 2020, pp. 1–6, doi:
10.1109/ICORIS50180.2020.9320828.
[24] M. Z. Alom et al., “The history began from AlexNet: a comprehensive survey on deep learning approaches,” arXiv- Computer
Science, pp. 39, 2018.
[25] R. Poojary, R. Raina, and A. Kumar Mondal, “Effect of data-augmentation on fine-tuned CNN model performance,” IAES
International Journal of Artificial Intelligence, vol. 10, no. 1, pp. 84-92, Mar. 2021, doi: 10.11591/ijai.v10.i1.pp84-92.

BIOGRAPHIES OF AUTHORS

Imam Bil Qisthi holds a Bachelor of Engineering (S.T.) in Informatics

Engineering, and is currently pursuing a Master of Engineering (M.T.) in Electrical
Engineering in addition to several certificates and professional skills. He is currently staff at
Gunadarma University. His research areas of interest include artificial intelligent, image
processing, natural language processing (NLP), and data science. He can be contacted at
email: [email protected].

Hartono Siswono is head of graduate program in Electrical Engineering at

Gunadarma University. Graduated from Indiana Institute of Technology, Indiana, USA in
January 1988, with the degree of Bachelor of Science in Engineering. In July 2000, graduated
with the degree of Master of Science in Electrical Engineering from University of Indonesia.
He got the doctoral degree in Electrical Engineering from University of Indonesia in January
2006. Started as an instructor in University of Gunadarma since 1993, and since June 2013 as
head in Department of Electrical Engineering, University of Gunadarma. Since March 2022 as
head in graduate program in Electrical Engineering in University of Gunadarma. When
studying in the United States of America, also work as an assistant for Department of
Electrical Engineering, Indiana Institute of Technology, and also as the president of
Indonesian Students Association in the USA for two periods. He as an author of six books,
which are: digital system design by using XILINX (perancangan sistem digital simulator
XILINX), signal processing system (sistem pemrosesan sinyal), control system (dasar sistem
kontrol), digital signal processing (pengolahan sinyal digital), linear system analysis and
theory (teori dan analisis sistem linier), and analog filter design (perancangan filter analog).
He can be contacted at email: [email protected]

Int J Artif Intell, Vol. 13, No. 2, June 2024: 2441-2450