Jurnal Nasional Teknik Elektro
Not a member yet
359 research outputs found
Sort by
Integrating YOLOv7 with FixMatch for Enhancing Vehicle Detection Performance in Mixed Traffic Environments
Full text linkA major challenge in the development of object detection technology is the significant reliance on large labeled datasets, which requires substantial time and memory for manual annotation—especially in complex, mixed traffic environments with varied vehicle types, congestion levels, and unpredictable motion patterns. This study addresses this issue by integrating the semi-supervised learning technique, FixMatch, into the YOLOv7 object detection model, utilizing 4000 transportation-related datasets. The FixMatch technique enables the model to detect unlabeled objects effectively through strong and weak augmentation methods. In this study, the detected objects in the mixed traffic environment include public transportation, pedicabs, cars, motorcycles, and trucks. This study achieved an impressive 97.5% detection accuracy by leveraging unlabeled data, demonstrating the model's efficiency and effectiveness in identifying vehicles under diverse traffic conditions. Consequently, integrating the FixMatch method into YOLOv7 provides a practical and efficient solution for object detection in situations where collecting labeled data is challenging, such as in dynamic and highly variable traffic environments
Analisis Perbandingan Quality of Service VSAT IP dan VSAT Star Telkomsat
No full textTechnological developments and the increase in internet users, estimated to reach 185 million in early 2024, have driven the need for reliable communication network services, especially for underdeveloped, frontier, and outermost (3T) areas in Indonesia, where internet coverage is still minimal. Based on these problems, telecommunications service providers need to prepare appropriate designs to provide optimal service to customers by providing a quality network. Telkomsat offers solutions through satellite-based services, especially very small aperture terminals (VSAT). Quality of service (QoS) testing is carried out by measuring throughput, packet loss, delay, and jitter using transmission control protocol (TCP) and user datagram protocol (UDP) data to compare the performance of VSAT IP and VSAT Star. The results show that VSAT Star is superior in throughput and delay, with an average throughput of 8.79 Mbps and an average delay of 12.32 ms. This is better than VSAT IP which only produces 6.43 Mbps and 83.94 ms. Both services have the same average packet loss of 0%. However, VSAT IP is more stable in terms of jitter with an average value of 0.36 ms compared to VSAT Star which produces 1.05 ms. In the ping test to the public domain (Google.com), VSAT Star showed an excellent average value of 38.55 ms compared to 584.05 ms for VSAT IP. Overall, VSAT Star has greater potential because of its advantages such as auto point, larger bandwidth, and lower delay
Performance Enhancement of Elephant Herding Optimization Algorithm Using Modified Update Operators
Full text linkThis research paper presents a modified version of the Elephant Herding Optimization (EHO) algorithm, referred to as the Modified Elephant Herding Optimization (MEHO) algorithm, to enhance its global performance. The focus of this study lies in improving the balance between exploration and exploitation within the algorithm through the modification of two key operators: the matriarch updating operator and the separation updating operator. By reframing the equations governing these operators, the proposed modifications aim to enhance the algorithm’s ability to discover optimal global solutions. The MEHO algorithm is implemented in the MATLAB environment, utilizing MATLAB R2019a. To assess its efficacy, the algorithm is subjected to rigorous testing on various standard benchmark functions. Comparative evaluations are conducted against the original EHO algorithm, as well as other established optimization algorithms, namely the Improved Elephant Herding Optimization (IEHO) algorithm, Particle Swarm Optimization (PSO) algorithm, and Biogeography-Based Optimization (BBO) algorithm. The evaluation metrics primarily focus on the algorithms’ capacity to produce the best global solution for the tested functions. The proposed MEHO algorithm outperformed the other algorithms on 75% of the tested functions, and 62.5% under two specific test scenarios. The findings highlight the effectiveness of the proposed modification in enhancing the global performance of the Elephant Herding Optimization algorithm. Overall, this work contributes to the field of optimization algorithms by presenting a refined version of the EHO algorithm that exhibits improved global search capabilities
Perbandingan Performa Metode Maximum Power Point Tracking Human Psychology Optimization (HPO), Artificial Bee Colony (ABC) dan Fuzzy Logic Controller (FLC) pada Flyback Converter Kondisi Parsial Shading
No full textMaximum Power Point Tracking (MPPT) is a method to track the power point of an energy source with the intention to generate maximum power. The surface of the Solar Panel has the possibility of being blocked when it receives sunlight. The barrier can be in the shape of shadows of objects that are nearby solar panels. The problem causes the power generated to be not optimal and makes more than one MPPT peak on the characteristics of P-V. This paper compares several methods of MPPT such as Human Psychology Optimization (HPO), Artificial Bee Colony (ABC), and Fuzzy logic Controller (FLC) under partial shading conditions, the comparison of three method by simulation. This algorithm hooks up to a flyback converter to provide MPP. From the results of MPPT accuracy in partial shading situations, the ABC and HPO approach methods can achieve GMPP with more than 82.22 % accuracy. For convergence, ABC needs extra time to discover GMPP. From the results, the Fuzzy approach can track however nevertheless trapped on LMPP.Maximum Power Point Tracking (MPPT) adalah metode untuk melacak titik daya dari sumber energi dengan tujuan untuk menghasilkan daya maksimum, yang kemungkinan besar terhalang di permukaan panel surya. Penghalang itu bisa berupa bayangan benda-benda yang berada di dekat panel surya. Permasalahan tersebut menyebabkan daya yang dihasilkan menjadi tidak optimal dan membuat lebih dari satu puncak MPPT pada karakteristik P-V. Makalah ini membandingkan beberapa metode MPPT seperti Human Psychology Optimization (HPO), Fuzzy logic Controller (FLC), dan Artificial Bee Colony (ABC) pada kondisi partial shading, perbandingan ketiga metode tersebut secara simulasi. Algoritma ini menghubungkan ke konverter flyback untuk menyediakan MPP. Dari hasil akurasi MPPT pada situasi partial shading, metode pendekatan ABC dan HPO dapat mencapai GMPP dengan akurasi lebih dari 82%. Untuk konvergensi, ABC membutuhkan waktu ekstra untuk menemukan GMPP. Dari hasil pendekatan Fuzzy dapat melacak namun tetap terjebak pada LMPP
An Embedded Convolutional Neural Network for Maze Classification and Navigation
Full text linkTraditionally, the maze solving robots employ ultrasonic sensors to detect the maze walls around the robot. The robot is able to transverse along the maze omnidirectionally measured depth. However, this approach only perceives the presence of the objects without recognizing the type of these objects. Therefore, computer vision has become more popular for classification purpose in robot applications. In this study, a maze solving robot is equipped with a camera to recognize the types of obstacles in a maze. The types of obstacles are classified as: intersection, dead end, T junction, finish zone, start zone, straight path, T–junction, left turn, and right turn. Convolutional neural network, consisting of four convolution layers, three pooling layers, and three fully-connected layers, is employed to train the robot using a total of 24,000 images to recognize the obstacles. Jetson Nano development kit is used to implement the trained model and navigate the robot. The results show an average training accuracy of 82% with a training time of 30 minutes 15 seconds. As for the testing, the lowest accuracy is 90% for the T-junction with the computational time being 500 milliseconds for each frame. Therefore, the convolutional neural network is adequate to serve as classifier and navigate a maze solving robot
Performance Comparison of FBMC-OQAM and CP-OFDM Using AWGN Channel
Full text linkThe 5G NR network planning covers the types of use scenarios and applications that include Enhanced Mobile Broadband (eMBB), Ultra-Reliable and Low Latency Communications (URLLC), and Massive Machine Type Communications (MTC). Regarding multicarrier modulation schemes, Orthogonal Frequency Division Multiplexing (OFDM) has become the most popular choice in previous technology, so OFDM is a strong candidate for its 5G NR technology application. However, OFDM has disadvantages such as higher PAPR and decreased bandwidth efficiency due to the addition of CP. These weaknesses can be overcome by the FBMC modulation scheme with Offset Quadrature Amplitude Modulation (OQAM) as a more efficient CP replacement for its implementation in 5G NR. This study analyzed the development of OQAM in Filter Bank multicarrier (FBMC) and compared it with using Cyclic Prefix (CP) based on OFDM using the AWGN channel. The first step of this research is to present an overview of the modulation scheme used. Next, compare the performance of FBMC-OQAM and CP-OFDM by analyzing several Bit Error Rate (BER) simulation results against the SNR value when both systems use the same simulation parameters. Based on the test results of each BER, both methods have different values, almost 2 dB for the same BER results. It indicates that the FBMC-OQAM system reached the BER value of 10-4 at SNR 15 dB. The CP-OFDM system, meanwhile, was able to achieve a BER value of 10-4 at SNR 17 dB. These results indicate that the FBMC-OQAM system is superior to CP-OFDM based on the BER values
Desain dan Implementasi Sistem Sensor untuk Lokalisasi pada Autonomous Robot IVANA di Area Gedung
No full textOne of the popular studies recently is about social robots that have been implemented in several public areas such as offices. The robot is an employee or worker assistant robot in the Telkom Surabaya Institute of Technology building to help carry out the work of delivering packages to the destination according to the tasks given. The problem that often occurs is an error in the robot's localization system causing the robot's movement to the target point to experience a position error. This research contributes to the comparative evaluation of 2 localization methods on mobile robots, namely the first is the use of a rotary encoder sensor and the second is the use of sensor fusion based on the extended Kalman filter implemented on the robot prototype. This study aims to develop a sensor system that is adapted to the design of the robot and the environment in which the robot is tested and to find out the comparison of the two methods. The use of extended Kalman filter-based sensor fusion can provide more accurate results in robot localization, especially when moving on complex paths. Sensor fusion enables the combination of several sensors such as rotary encoders and IMU (Inertial Measurement Unit) sensors to provide more complete and accurate information about the position and orientation of the robot. In this study, sensor fusion successfully reduced the localization error of the robot to 0.63 m when moving straight and 0.29 m when moving on a complex path, compared to the use of a single sensor which resulted in a larger error of 0.89 m. Based on the study that has been conducted, it can be considered as a potential solution in the development of other social robots to improve the accuracy and performance of the robots when performing certain tasks in the future.Salah satu penelitian populer baru baru ini adalah mengenai robot sosial yang telah diimplementasikan di beberapa area publik seperti halnya perkantoran. Robot IVANA adalah robot asisten karyawan atau pekerja di gedung Institut Teknologi Telkom Surabaya untuk membantu melakukan pekerjaan mengantarkan paket ke tempat yang dituju sesuai dengan tugas yang diberikan. Permasalahan yang masing sering terjadi adalah kesalahan dalam sistem lokalisasi robot membuat gerakan robot menuju titik target mengalami kesalahan posisi. Penelitian ini memberikan kontribusi pada evaluasi perbandingan 2 metode lokalisasi pada robot bergerak yaitu yang pertama adalah penggunanaan satu sensor rotary encoder dan yang kedua adalah penggunaan sensor fusion berbasis extended Kalman filter yang diimplementasikan pada prototype robot . Studi ini bertujuan untuk mengembangkan sistem sensor yang disesuaikan dengan rancang bangun robot dan lingkungan tempat robot diujikan serta mengetahui perbandingan kedua metode. Penggunaan sensor fusion berbasis extended Kalman filter dapat memberikan hasil yang lebih akurat dalam melakukan lokalisasi pada robot, terutama saat bergerak pada lintasan yang kompleks. Sensor fusion memungkinkan gabungan beberapa sensor seperti sensor rotary encoder dan sensor IMU (Inertial Measurement Unit) untuk memberikan informasi yang lebih lengkap dan akurat tentang posisi dan orientasi robot. Dalam penelitian ini, sensor fusion telah berhasil mengurangi kesalahan lokalisasi pada Robot IVANA menjadi 0,63 m saat bergerak lurus dan 0,29 m saat bergerak pada lintasan yang kompleks, dibandingkan dengan penggunaan sensor tunggal yang menghasilkan kesalahan yang lebih besar 0,89 m.Berdasarkan studi, ke depan dapat menjadi pertimbangan solusi pada pengembangan robot sosial lainnya untuk meningkatkan akurasi dan kinerja robot saat melakukan tugas-tugas tertentu
Klasifikasi Multikelas Infark Miokard Berdasarkan Sinyal Phonokardiogram dengan Ensemble Learning
No full textMyocardial infarction (MI) is a serious cardiovascular disease with a high mortality rate worldwide. Early detection and consistent treatment can significantly reduce mortality from cardiovascular diseases. However, there is a need for efficient models that can enable the early detection of heart disease without relying on trained clinical experts. MI studies using phonocardiogram (PCG) signals and implementing ensemble learning models are still relatively scarce, often resulting in poor accuracy and low detection rates. This study aims to implement an ensemble learning model for the classification of MI using PCG signals into different classes. In this stage of research, several classification algorithms, including Random Forest and Logistic Regression, serve as basic models for ensemble learning, utilizing features extracted from audio signals. Evaluation of the model's performance reveals that the stacking model achieves an accuracy of 96%. These results demonstrate that our system can appropriately and accurately classify MI within PCG data. We believe that the findings of this study will enhance the diagnosis and treatment of heart attacks, making them more effective and accurate.Infark Miokard (MI) adalah penyakit kardiovaskular yang serius dengan tingkat kematian yang tinggi di seluruh dunia. MI disebabkan oleh berkurangnya atau terhentinya aliran darah ke sebagian miokardium, yang menyebabkan kekurangan oksigen di dalam miokardium. Pasokan yang tidak memadai ke miokardium dapat menyebabkan kematian kardiomiosit dan nekrosis. Deteksi dini dan pengobatan yang konsisten dapat mengurangi kematian dini akibat penyakit kardiovaskular. Namun, model yang efisien diperlukan untuk deteksi dini penyakit jantung tanpa memerlukan ahli klinis yang terlatih. Penelitian MI menggunakan sinyal PCG yang mengimplementasikan model pembelajaran ensembel masih jarang dilakukan dengan akurasi yang buruk dan tingkat deteksi yang rendah. Penelitian ini bertujuan untuk mengembangkan implementasi model pembelajaran ensemble untuk mengklasifikasikan sinyal phonocardiogram (PCG) ke dalam beberapa kelas. Pada tahap penelitian ini, beberapa algoritma klasifikasi seperti SVM, Random Forest, XGBoost dan Neural Networks digunakan sebagai model dasar untuk pembelajaran ensembel berdasarkan fitur yang diekstrak dari sinyal audio. Setelah mengevaluasi kinerja model, hasilnya menunjukkan bahwa model stacking memiliki akurasi sebesar 94%. Hasil ini menunjukkan bahwa sistem kami dapat mengklasifikasikan MI dalam data PCG dengan tepat dan dengan akurasi yang tinggi. Diharapkan hasil penelitian ini akan meningkatkan diagnosis dan pengobatan serangan jantung secara lebih efektif dan akurat
Model Berbasis Transfer Learning untuk Deteksi Tumor Otak pada Citra MRI
No full textBrain tumors are life-threatening medical conditions characterized by abnormal cell proliferation in or near the brain. Early detection is crucial for successful treatment. However, the scarcity of labelled brain tumor datasets and the tendency of convolutional neural networks (CNNs) to overfit on small datasets have made it challenging to train accurate deep learning models for brain tumor detection. Transfer learning is a machine learning technique that allows a model trained on one task to be reused for a different task. This approach is effective in brain tumor detection as it allows CNNs to be trained on larger datasets and generalize better to new data. In this research, we propose a transfer learning approach using the Xception model to detect four types of brain tumors: meningioma, pituitary, glioma, and no tumor (healthy brain). The performance of our model was evaluated on two datasets, demonstrating a sensitivity of 98.07%, specificity of 97.83%, accuracy of 98.15%, precision of 98.07%, and f1-score of 98.07%. Additionally, we developed a user-friendly prototype application for easy access to the Xception model for brain tumor detection. The prototype was evaluated on a separate dataset, and the results showed a sensitivity of 95.30%, specificity of 96.07%, accuracy of 95.30%, precision of 95.31%, and f1-score of 95.27%. These results suggest that the Xception model is a promising approach for brain tumor detection. The prototype application provides a convenient and easy-to-use way for clinical practitioners and radiologists to access the model. We believe the model and prototype generated from this research will be valuable tools for diagnosing, quantifying, and monitoring brain tumors.Tumor otak adalah kondisi medis parah yang dapat berakibat fatal. Tumor ini dibedakan dari proliferasi sel yang tidak teratur dan berlebihan di dalam atau di dekat otak. Diagnosis dan deteksi dini tumor otak harus dipertimbangkan karena sangat penting untuk keberhasilan terapi. Namun, kelangkaan dataset tumor otak yang berlabel dan kecenderungan jaringan saraf konvolusi (convolutional neural network/CNN) untuk melakukan overfitting pada dataset yang kecil telah mempersulit dalam melatih model deep learning yang akurat untuk mendeteksi tumor otak. Transfer learning adalah teknik pembelajaran mesin yang memungkinkan model yang dilatih untuk satu tugas dapat digunakan kembali untuk tugas yang berbeda. Teknik ini efektif untuk deteksi tumor otak, karena memungkinkan CNN untuk dilatih pada set data yang lebih besar dan menggeneralisasi lebih baik untuk data yang tidak terlihat. Dalam penelitian ini, kami mengusulkan pendekatan pembelajaran transfer menggunakan model Xception untuk mendeteksi empat jenis tumor otak: meningioma, hipofisis, glioma, dan tidak ada tumor (otak sehat). Kami mengevaluasi kinerja model kami pada dua dataset, dan hasilnya menunjukkan bahwa model kami mencapai sensitivitas 98,07%, spesifisitas 97,83%, akurasi 98,15%, presisi 98,07%, dan f1-score 98,07%. Kami juga mengembangkan prototipe aplikasi yang memungkinkan pengguna untuk mengakses model Xception untuk deteksi tumor otak dengan mudah. Prototipe ini dievaluasi pada dataset terpisah, dan hasilnya menunjukkan sensitivitas 95,30%, spesifisitas 96,07%, akurasi 95,30%, presisi 95,31%, dan f1-skor 95,27%. Hasil ini menunjukkan bahwa model Xception adalah pendekatan yang menjanjikan untuk deteksi tumor otak. Aplikasi prototipe ini menyediakan cara yang nyaman dan mudah digunakan bagi para praktisi klinis dan ahli radiologi untuk mengakses model tersebut. Kami percaya bahwa model dan prototipe yang dihasilkan dari penelitian ini akan menjadi alat yang berharga untuk mendiagnosis, mengukur, dan memantau tumor otak
Audible Obstacle Warning System for Visually Impaired Person Based on Image Processing
Full text linkTo be able to do their daily activities, a visually impaired person needs a guidance device to help him/her walk including to avoid obstacles on their way to the destination. The quick and clear instruction is given to the user is the most challenging problem to be solved. The visually impaired person should have simple guidance about the obstruction in front of him/her. Most guidance devices use simple sounds to give the warning without information about which direction the user should go. In this paper, an obstacle warning system based on image processing methods was developed. A guidance device for visually impaired persons using a single-board computer based on an image-processing algorithm has been designed. The main sensor of the guidance device is a NoIR camera. The distance measurement approximation model was developed with errors up to 4.3%. The test found that the proposed system can detect obstruction in the form of a person, the device also detects the stairs. The best detection obtains when the object position is less than 300 cm in front of the user. The stair detection was carried out by using the Hough line transform method. The output of the system is the sound of direction that can be heard through the headset