Indonesian Journal of Electrical Engineering and Computer Science
Not a member yet
9109 research outputs found
Sort by
PRDTinyML: deep learning-based TinyML-based pedestrian detection model in autonomous vehicles for smart cities
Get PDFDetecting pedestrians and cars in smart cities is a major task for autonomous vehicles (AV) to prevent accidents. Occlusion, distortion, and multi-instance pictures make pedestrian and rider detection difficult. Recently, deep learning (DL) systems have shown promise for AV pedestrian identification. The restricted resources of internet of things (IoT) devices have made it difficult to integrate DL with pedestrian detection. Tiny machine learning (TinyML) was used to recognize pedestrians and cyclists in the EuroCity persons (ECP) dataset. After preliminary testing, we propose five microcontroller-deployable lightweight DL models in this study. We applied SqueezeNet, AlexNet, and convolution neural network (CNN) DL models. We also use two pre-trained models, MobileNet-V2 and MobileNet-V3, to determine the optimal size and accuracy model. Quantization aware training (QAT), full integer quantization (FIQ), and dynamic range quantization (DRQ) were used. The CNN model had the shortest size with 0.07 MB using the DRQ approach, followed by SqueezeNet, AlexNet, MobileNet-V2, and MobileNet-V2 with 0.161 MB, 0.69 MB, 1.824 MB, and 1.95 MB, respectively. The MobileNet-V3 model’s DRQ accuracy after optimization was 99.60% for day photos and 98.86% for night images, outperforming other models. The MobileNet-V2 model followed with DRQ accuracy of 99.27% and 98.24% for day and night images
The integration of metaverse technology in healthcare: a comprehensive review and future research directions
Get PDFThe impact of using the metaverse in healthcare is investigated in this research work. Emerging technologies are essential to enhancing medical consultants’ care, especially in developing countries like India. The study filters and reviews the pertinent literature using the scientific procedures and rationales for systematic literature reviews (SPAR-4-SLR) methodology. The initial search yielded 180 articles. Forty-four articles were considered for the study after screening the papers in light of the research questions and relevant literature. The theory-context-characteristics-methodology (TCCM) framework is used in this study to assess future metaverse research trends. This study also used the context, intervention, mechanism, and outcome (CIMO) logic for planning and decision-making. This study examines the development of metaverse research over the past ten years and supports research findings published in peer-reviewed journals. Based on the TCCM framework, recommendations have been made for additional research
Household electric monitoring IoT system
Get PDFIn dense areas in the Philippines, there are recorded cases of power theft or known to be illegally tapping power lines from another household which results to complaints because of increased electricity bills. To address the power theft problems, this work uses internet of things system for household electric monitoring and control. A transmitter and receiver set up is designed to monitor the energy consumption at both ends. When there is discrepancy with the meter reading, an alert system sends notification that there is an illegal wiretapping. The load is monitored through electric meters and the powers measured are compared. These data are being sent wirelessly through a GSM module. The meter readings for both the transmitter and receiver can be viewed in a mobile phone through a web app developed. A minimum of 3W difference between the transmitter and the receiver will mean a discrepancy and notifies illegal wiretapping. Illegal connections are cutoff when an incident of tapping occurs. Based on the results of the test, the household electricity monitoring system through internet of things (IoT) is found to be 100% reliable in detecting and cutting off illegal connections. Additionally, the system is able to compute the monthly power consumption
An intelligent system for job recommendation based on semantic analysis of candidate's resume
Get PDFThe contemporary job market presents significant obstacles to effectively aligning proficient candidates with pertinent employment prospects. The conventional methods of resume screening and job matching frequently require substantial manual effort and are susceptible to subjective biases, resulting in recruiting decisions that are frequently suboptimal. The present study proposes the development of an intelligent job recommendation system that utilises semantic analysis of candidates' resumes and job descriptions sourced from several job portals. The objective of the proposed intelligent system is to enhance and streamline the recruiting process through the automated extraction and analysis of pertinent skills from resumes and job descriptions, utilising natural language processing (NLP) and machine learning (ML) techniques. In addition, web scraping techniques were used to collect job advertisements from several job portals. The developed model exhibits the ability to recommend the most suitable job prospects by computing similarity metrics, such as Euclidean distance, between skill clusters identified in a job advertisement and a specified candidate's resume. The implemented model achieves an accuracy rate of 98.92%. It is anticipated that the integration of an intelligent job recommendation system will augment the recruitment procedure for both job seekers and employers
Investigation on TiO2/graphene as resistance-based gas sensor for volatile organic compound gases detection
Get PDFVolatile organic compound (VOC) gases are usually produced from industrial activities. Short-term exposure to VOC gases can cause dizziness, headaches, nausea, and throat irritation. Years to a long time exposure to VOC gases can cause cancer and system damage in the human body. With the growth of gas sensor technology, a resistance-based gas sensor based on various structures of resistance-based gas sensors using Titanium dioxide/graphene (TiO2/graphene) were investigated as a sensing material for detecting volatile organic compound gases, which are acetone and ethanol. The TiO2/graphene gas sensor was deposited on a Kapton film using a screen printing technique. All TiO2/graphene gas sensors were exposed to acetone and ethanol at room operating temperature. The results revealed that the highest response values to acetone and ethanol were produced by T99_G1_2 and T98_G2_1, respectively. It can be concluded that design 1 generated the most consistent response to acetone, while design 2 generated the most consistent response to ethanol
A hybrid APSO–ANFIS optimization based load shifting technique for demand side management in smart grids
Get PDFCost and performance are considered important parameters to obtain an optimized configuration for smart grids. In this paper, a new optimization approach, based on a hybrid adaptive particle swarm with an adaptive neurofuzzy inference system (ANFIS) algorithm, has been proposed. This approach allows optimizing demand side management (DSM) using the load shifting technique. The impact of the latter on consumer profile, electricity pricing mechanisms, and overall grid performance are illustrated. In this simulation, the focus lies on modeling DSM using a day-ahead load shifting approach as a minimization problem. Simulation experiments have been tested separately on three different demand zones, namely, residential, commercial, and industrial zones. A comparative study of solutions was performed, focusing on both reduced peak demand and operational costs. The obtained results demonstrate that the optimization presented in this article approach outperforms the other approaches by achieving greater savings in the residential and commercial sectors. The study proved a significant reduction in peak demand. In fact, values of 23.76%, 17.61% and 16.5% in peak demand reduction are achieved in the case of residential, commercial, and industrial sectors, respectively. Furthermore, operational cost reductions of 7.52%, 9.6%, and 16.5% are obtained for the three different cases
Optimization of sales by applying e-commerce and digital marketing through social networks
Get PDFCompanies must have a strategy plan to satisfy their users and implement new methods to work with technology since people nowadays are more related to technology avoiding traditional sales and having virtual sales is why it has the objective of optimizing sales in companies by applying e-commerce and digital marketing through social networks. The methodology was carried out with Scrum, which has five stages (planning meeting, sprint backlog, daily meetings, sprint review, and retrospective review) that allows to comply with each established sprint showing as a result a functional project. As a result indicates the solution of each phase of the methodology getting the ecommerce system, with a validation by 7 experts specialized in (realism, integration, adaptability, technology, innovation, functionality, and usability) indicating a total of 93% showing a perfect state of the system and meets the satisfaction for the user and finally indicates the development of digital marketing by the social network Facebook showing a great improvement in their sales reaching up to triple their sales
A hybrid learning model to detect cardiovascular disease from electrocardiogram
Get PDFCardiovascular diseases (CVDs) continue to be the world’s most significant cause of morbidity and mortality. This paper introduces a unique hybrid learning model for CVD detection using advanced deep learning (DL) methods. The proposed method combines the potent feature extraction powers of the EfficientNet pre-trained model with attention mechanisms and graph convolutional networks (GCNs) for improved performance. First, rich representations from cardiovascular electrocardiogram (ECG) data extract using the EfficientNet architecture as a feature extractor. Using a large dataset of cardiovascular ECG images, you can fine-tune the pre-trained EfficientNet model with Pipeline to make it more suitable for disease identification. Including attention techniques that allow the network to focus on informative regions within the input, ECG images enhanced the model’s discriminative capacity. The model can attend to the salient areas selectively linked with CVD path physiology through dynamic attention processes. More accurate predictions result from this attention-based refining, strengthening the model’s ability to identify significant patterns suggestive of cardiovascular problems. GCN aims to link the natural structure in cardiovascular data. It can efficiently capture complex interactions and dependencies among various data pieces by expressing medical data as graphs, where nodes correspond to image regions, and edges imply spatial connections. Combining GCN into the proposed hybrid learning architecture facilitates extracting contextual information from local and global sources, augmenting the model’s accuracy
A comparative study of CNN architectures for the detection of tomato leaf diseases
Get PDFRecent advancements in computer vision and machine learning (ML) have revolutionised various sectors, including precision agriculture (PA). In our study, we focused on detecting tomato leaf diseases (TLD) using deep learning (DL) techniques. Using a convolutional neural network (CNN) model, we developed an agricultural image index to accurately detect TLD. By utilizing available datasets from Kaggle, we trained our model to recognize various TLDs. To determine the most effective one, we compared multiple architectures, including VGG, ResNet, and EfficientNetB1. The obtained results demonstrated a classification accuracy of over 99% on the test set. This approach has allowed us to accelerate and enhance the disease detection process, positively impacting agricultural communities by reducing crop losses and enabling early intervention in case of disease outbreaks. Our study highlights the effectiveness of CNN models in the detection of TLD, paving the way for future applications in PA
Optimizing distance vector-hop localization in wireless sensor networks using the grasshopper optimization algorithm
Get PDFIn scenarios involving mobile sensors within distributed sensor systems, such as those often encountered in wireless sensor networks (WSNs) or the internet of things (IoT), the ability to ascertain the origin of sensor data holds significant importance. Range-free Monte Carlo Localization methods offer an energy-efficient solution that eliminates the need for extra hardware, as they solely rely on the radio hardware already present on sensor nodes. But there are certain disadvantages when implemented, as it occupies more amount of power and some inaccuracies might happen in accessing the data from the sensor node. In this paper, we suggest the grasshopper optimization algorithm (GOA) strategy, which incorporates the distance-vector hop (DVHop) and three-anchor methods. It displays its usefulness in terms of both overall localization accuracy and resistance to hostile attacks or malfunctioning nodes. Nonetheless, the incorporation of dead reckoning based on motion sensor data significantly enhances the precision of location estimates and bolsters the network's robustness against both faulty components and malicious agents