Indonesian Journal of Electrical Engineering and Computer Science
Not a member yet
9109 research outputs found
Sort by
Recognizing geographical locations using a GAN-based text-to-image approach
Get PDFGenerating photo-realistic images that align with the text descriptions is the goal of the text-to-image generation (T2I) model. They can assist in visualizing the descriptions thanks to advancements in machine learning algorithms. Using text as a source, generative adversarial networks (GANs) can generate a series of pictures that serve as descriptions. Recent GANs have allowed oldest T2I models to achieve remarkable gains. However, they have some limitations. The main target of this study is to address these limitations to enhance the text-to-image generation models to enhance location services. To produce high-quality photos utilizing a multi-step approach, we build an attentional generating network called AttnGAN. The fine-grained image-text matching loss needed to train the AttnGAN’s generator is computed using our multimodal similarity model. With an inception score of 4.81 on the PatternNet dataset, our AttnGAN model achieves an impressive R-precision value of 70.61 percent. Because the PatternNet dataset comprises photographs, we’ve added verbal descriptions to each one to make it a text-based dataset instead. Many experiments have shown that AttnGAN’s proposed attention procedures, which are critical for text-to-image production in complex circumstances, are effective
Value group classifier model for ethical decision-making
Get PDFDecision-makers refer to ethics or moral philosophy during times of ethical dilemma. Dilemmas are situations of inner conflict, which require a methodical approach. Diversity in viewpoints on moral decisions ensures there cannot be a fixed solution for ethical dilemmas as in the case of numerical problems. Existing ethical and sustainable decision models for businesses are not automated because of a lack of a comprehensive list of dilemmas. To resolve this gap, an AI model was trained to classify all dilemmas into three value groups by using a support vector classifier (SVC). The model provided scaffolding to the ethical decision-maker by suggesting relevant human values applicable to the dilemma. The design works on the ethical theory of stakeholder management, which includes sustainable business goals. The study was conducted with 30 students and 30 adults to identify their dilemmas. The dilemma dataset was used to train an ethical decision-support tool, called the value group classification (VGC) model. The model achieved a score of 0.52 on performance. The VGC model overcomes the black-box biases of similar machine-learning models by allowing human autonomy in ethical decisions
Proposed strategies for lightning performance improvement of 35 kV distribution lines in Vietnam
Get PDFLightning strikes frequently cause power outages on 35 kV distribution lines in Vietnam. These lines was normally protected with a sparse density of surge arrester and typically do not make use of shield wire. On the other hand, a single rod, that is frequently used as the grounding electrode, is not suitable for some region with high value of soil resistivity. Therefore, it is of particular of interest to investigate the solutions to protect the medium voltage lines from the back-flashover due to lightning strikes by analyzing the impact of the grounding impedance, the surge arrester density and the installation of the shield wire. The finite element method and the electromagnetic transient program EMTP-RV are combined in this work to propose techniques for increasing the critical flashover current of 35 kV lines in Hanoi city. The obtained results showed that the installation of shield wires combined with high density of surge arrester and reduced grounding impedance value can achive a very significant improvement of the overall lightning performance of the line
Authenticated image encryption using robust chaotic maps and enhanced advanced encryption standard
Get PDFThe ability of advanced encryption standard (AES) algorithm to protect information systems has given cryptography a new dimension. Recent encryption approaches to enhance randomness include the use of chaotic algorithms, which provide resistance to differential attacks. We have proposed the application of robust chaotic maps in the block cipher to design a secure authenticated encryption scheme to get advantages of both. The chaotic sequence is generated using hyperbolic tangent map and added to input image initially to increase randomness. The basic 256-bit AES key is generated using the robust Renyi modulo map. An additional 128-bit key enhances security. Instead of static values used in AES, dynamic initialization vector (IV), different for every image will be generated. The results are mathematically verified using various security parameters. The algorithm provides lower values of peak signal-to-noise ratio (PSNR) (7.81 to 9.10 dB) for encrypted images and higher dissimilarities between input and encrypted image histograms. Thus, it is highly resistant to statistical attacks. The experimental results and their comparison prove the superiority of our proposed cryptosystem against statistical, differential and brute-force attacks. Thus, the novel multi-chaotic AES-GCM (galois/counter mode) algorithm can be used for color image encryption in military and industrial applications demanding high data security and authentication
BanSpEmo: a Bangla audio dataset for speech emotion recognition and its baseline evaluation
Get PDFSpeech interfaces provide a natural and comfortable way for humans to communicate with machines. Recognizing emotions from acoustic signals is essential in audio and speech processing. Detection of emotion in speech is critical to the next generation of human-computer interaction (HCI) fields. However, a lack of large-scale datasets has hampered the progress of relevant research. In this study, we prepare BANSpEmo, a demanding Bangla speech emotion dataset consisting of 792 audio recordings totaling more than 1 hour and 23 minutes. The recordings feature 22 native speakers and each speaker uttered two sets of sentences representing six emotions: disgust, happiness, anger, sadness, surprise, and fear. The dataset consists of 12 Bangla sentences, each expressed in these six emotions. Furthermore, a series of investigations are carried out to assess the baseline performance of the support vector machine (SVM), logistic regression (LR), and multinomial Naive Bayes models on the BANSpEmo dataset presented in this study. The studies found that SVM performed best on this dataset, with an accuracy of 87.18%
Effective autism spectrum disorder sensory and behavior data collection using internet of things
Get PDFWireless body area networks (WBANs) connected with wearable internet of things (WIoT) offer useful features including sensory information collection, analysis, and transmission for continuous behavior monitoring of autism spectrum disorder (ASD) patients. Due to users’ mobility and time-driven sensed data, data collection becomes very difficult. The current approach employs cluster-based multi-objective path-optimized data collection mechanisms that have experienced hotspot issues leading to loss of energy and coverage problems near the base stations. This work presents the high energy and reliable sensory and behavior data collection (HERSBDC) mechanism to address the research difficulties. To ensure network coverage, the HERSBDC initially provides a new uneven clustering mechanism. Next, multi-objective-based cluster head (CH) selection metrics are proposed. The final step is the creation of a multi-objective routing path to gather vital ASD data more reliably and energy-efficiently. Comparing the proposed HERSBDC algorithm to the low energy adaptive cluster-hierarchy (LEACH)-based, and distributed energy-efficient clustering and routing (DECR) methods, the simulation results demonstrate that the HERSBDC mechanism achieves a much better lifetime by 62.28% and 11.89%, the delivery ratio by 15.04% and 9.51%, with minimal delay by 52.65%, and 9.65%, and routing overhead by 32.05%, and 42.65%, respectively
SMOTE tree-based autoencoder multi-stage detection for man-in-the-middle in SCADA
Get PDFSecurity incidents targeting supervisory control and data acquisition (SCADA) infrastructure are increasing, which can lead to disasters such as pipeline fires or even lost of lives. Man-in-the-middle (MITM) attacks represent a significant threat to the security and reliability of SCADA. Detecting MITM attacks on the Modbus SCADA networks is the objective of this work. In addition, this work introduces SMOTE tree-based autoencoder multi-stage detection (STAM) using the Electra dataset. This work proposes a four-stage approach involving data preprocessing, data balancing, an autoencoder, and tree classification for anomaly detection and multi-class classification. In terms of attack identification, the proposed model performs with highest precision, detection rate/recall, and F1 score. In particular, the model achieves an F1 score of 100% for anomaly detection and an F1 score of 99.37% for multi-class classification, which is preeminence to other models. Moreover, the enhanced performance of multi-class classification with STAM on minority attack classes (replay and read) has shown similar characteristics in features and a reduced number of misclassifications in these classes
Anomaly based detection in time series data on IoT systems using statistical models
Get PDFThe internet of things (IoT) has become a real revolution that represents technological innovation in all domains, it becomes more and more integrated into human activities starting from personal needs, to professional eras including industry, logistics, healthcare. Yet this technology didn’t only bring advantages to all industries, it also creates new challenges mostly security-related, due to the specifications of the IoT environment: its heterogeneity, the restricted resources, and the continuous enormous sensitive data generated and exchanged on the IoT ecosystem. In this paper, we propose a study of statistical models (autoregression, moving-average, autoregression-movingaverage, autoregression integrated movingaverage, seasonal autoregression integrated movingaverage) to build an anomaly-based detection model for times series data, to detect abnormal behavior that can be explained by a sensor failure, or a compromised sensor. The proposed anomaly-based detection relies on defining the data behavior and creating a profile on the assumed normal state, the created profile will be used as knowledge to predict future values to which real records will be compared, any deviation from the predicted data will be considered as abnormal, that indicates an anomaly has occurred on the sensor or the exchanged data. The proposed approach will help improve accuracy, reliability, trustworthiness, and data integrity
Improving unmanned armoured mobile robot navigation accuracy using Kalman filter
Get PDFThailand's remarkable economic progress has earned it recognition as a development success story, but terrorism remains a significant threat, particularly in the southern provinces. To effectively combat terrorism, intelligent systems like military robots are being increasingly utilized worldwide, including the development of vision-based robots and unmanned military equipment. The Thai government aligns with this trend by aiming to manufacture military robots, including armored mobile robot (AMR) equipped with cameras for peacekeeping and surveillance purposes. The development of an automatic tank simulation robot control system integrates global position system (GPS) technology, digital compasses, and lidar sensors to enhance efficiency, directional control, and obstacle avoidance. The study investigates the implementation of Kalman filters to enhance the precision of navigation in AMRs used in military contexts. The suggested system combines GPS, Lidar sensors, and digital compasses, utilizing advanced sensor fusion algorithms to improve directional control and obstacle avoidance. The system's usefulness is demonstrated through field tests and simulations, especially in complicated contexts where conventional approaches may face difficulties. The findings help to the progress of military robotics by providing a strong answer for improving navigation accuracy in real-world situations
A review of convolutional neural networks for classifying power quality problems using Keras API
Get PDFThe major causes of electric power quality (PQ) problems are mainly due to the increased utilization of nonlinear loads, capacitor and load switching events, transformer energization, and occurrence of assorted faults at the distribution corridor. The problems often introduce harmonics and other waveform anomalies like voltage sags, voltage swells and interruptions along the power systems. A timely classification of such problems is important in understanding their impact on costly power system economy. The paper explores comprehensive review of PQ issues, operational concept of convolutional neural network (CNN) and its utilization in solving PQ problems. Novel deep learning (DL) approach using variant of DenseNet CNN technique in Keras API platform is deployed to extract the features of, and classify PQ problems. The proposed technique improves classification performance with an accuracy of 99.96%. It shows remarkable improvement over the traditional techniques in the literature which were 73.53% to 99.92% accurate for a period from 2018 to 2023. The most promising part of the method is the improvement shown in the classification performance when compared with that obtained in the literature. The technique can also be applied in real time to cater for real PQ problems