Taiwan Association of Engineering and Technology Innovation: E-Journals
Not a member yet
887 research outputs found
Sort by
A Novel CNN-FRTAM Model for Enhanced Detection of Epileptic Seizures in EEG Signals Utilizing Deep Learning and Continuous Wavelet Transform
Full text linkThis study presents a novel approach to enhancing electroencephalography (EEG) signal classification for improved epileptic seizure detection. Traditional techniques for seizure detection often rely on manual analysis and suffer from high computational burdens and bias. Integrating deep learning with the continuous wavelet transform (CWT) is proposed to address these limitations for effective feature extraction. The CNN-FRTAM model, which integrates a convolutional neural network (CNN) and a frequency-region temporal attention mechanism (FRTAM), employs rigorous pre-processing of EEG signals to optimize performance. Extensive evaluation on a diverse dataset revealed an accuracy of 99.80% in multi-class classification and 99.90% in binary classification between Normal and Abnormal states. The CNN-FRTAM model significantly outperformed traditional architectures such as InceptionV3, VGG19, and ResNet50, demonstrating its potential for effective real-time applications in clinical epileptic seizure management. By opening up new avenues for accurate seizure detection, this work contributes to improving patient outcomes in epilepsy care
Surface Defect Detection of Aluminum Plates Using Improved Faster R-CNN and ResNet-50
Full text linkThis study aims to enhance the accuracy of surface defect detection in aluminum profiles. To address low recognition accuracy caused by irregular defect sizes and the coexistence of multiple defects, an inspection system integrating Faster Region-based Convolutional Neural Network (Faster R-CNN) and Residual Networks (ResNet-50) is proposed. After data enhancement and preprocessing of the aluminum profile image dataset from the Tianchi platform, ResNet-50 is used to extract deep features, and the Region Proposal Network (RPN) within Faster R-CNN is applied to generate candidate regions for classification and localization. Experimental evaluations demonstrate that the proposed model identifies all defect types with over 93% accuracy, while the error rate remains below 2%. Compared to the You Only Look Once Version 4 (YOLOv4) model, it exhibits greater performance in detecting surface defects. This advancement may lead to increased productivity and quality control in the manufacturing of aluminum profiles
Diffractive Efficiency Prediction of Surface Relief Grating Waveguide Using Artificial Neural Network
Full text linkThis study aims to develop lightweight and comfortable wearable devices using surface-relief grating, which can be designed to meet different diffraction conditions. However, extensive calculations must be performed to obtain the impact of the variation in the structural dimensions. The finite element method is used to solve the diffractive efficiency and then replaced by trained artificial neural networks with a single hidden layer containing 25 neurons. By using raw data with geometric parameters as the features, the performance of the network is investigated with different numbers of raw data; in addition, the regression analysis shows a high R-value of approximately 0.999. The predicted results are compared with those calculated from the simulation. The diffraction efficiency tendencies vary with the different geometric parameters, which show a high level of agreement between the predicted and calculated data; this confirms that the proposed method supports and reduces the burden of extensive calculations
Enhanced Electrocardiogram Arrhythmia Diagnosis with Deep Learning and Selective Attention Mechanism
Full text linkThe study aims to improve the diagnosis of arrhythmia in cardiovascular disease management. A novel approach using a deep convolutional network combined with a selective attention mechanism is proposed for electrocardiogram signal classification. The deep convolutional network extracts relevant features directly from raw electrocardiogram signals, while the selective attention mechanism focuses on the most critical regions of the signals and suppresses irrelevant or noisy components. This method achieves an accuracy of 99.70% in multi-class arrhythmia classification and 99.85% in binary classification, significantly outperforming traditional classification algorithms. Furthermore, the selective attention mechanism improves the localization of critical electrocardiogram segments, offering valuable insights for clinicians and aiding in the diagnosis process. This enhanced approach increases diagnostic accuracy and provides a clearer understanding of the electrocardiogram signals, which is crucial for effective patient management in cardiovascular diseases
FEGAN: A High-Performance Font Enhancement Network for Text CAPTCHA Preprocessing
Full text linkThis study aims to address performance deficiencies in CAPTCHA preprocessing methods that impede the accurate recognition of text CAPTCHAs, which are crucial for identifying security vulnerabilities. To improve CAPTCHA preprocessing methods, a similar font is initially searched and acquired by manually removing obstructing pixels from a target CAPTCHA and retaining the font part. Using the found font, a pseudo-dataset is generated containing a large number of clean and dirty pairs to train to the proposed supervised Font Enhancement Generative Adversarial Network (FEGAN), which is designed to effectively eliminate non-font-related interferences and preserve the font outlines. Test results show that FEGAN can improve the recognizer’s accuracy by approximately 16% to 50% on the M-CAPTCHA dataset (a publicly available dataset on Kaggle) and 5% to 35% on the P-CAPTCHA dataset (generated using the Python ImageCaptcha package), substantially outperforming the Multiview-filtering-based preprocessing approach
Using Webpage Comparison Method for Automated Web Application Testing with Reinforcement Learning
Full text linkWeb application testing often uses crawlers to explore the application under test (AUT) and identify potential vulnerabilities. For dynamically generated pages, crawlers must provide test inputs for web forms. A previous tool combines a web crawler with a reinforcement learning agent, which uses code coverage to guide the crawler in filling web forms. This paper aims to improve the applicability of web application testing by using webpage comparison techniques instead of code coverage and source code access, thereby enhancing the handling of multiple web forms on a single page. Experimental results show that this approach explores more pages, reaches greater crawling depths, and achieves better code coverage than the original method. It also interacts more efficiently with multiple web forms and outperforms a random-action Monkey on new, untrained web applications. Therefore, this approach is promising for automated web application testing
Dynamic Processing 2D Maps Method for Robot’s Trajectory Planning
Full text linkUnlike common usage of programmed maps, highly robust maps that mimic reality have rarely been tested for path-planning problems with a variety of search algorithms. Meanwhile, utilizing real-like maps might direct studies toward the image processing field and can be time-consuming. Therefore, this study aims to propose a method to effectively and quickly read and process 2D maps in such a way that search algorithms can recognize them. Simulations are conducted on two maps to show the merit of the proposed method. In all simulations, the proposed method successfully read and processed maps in an average time of 1.5043 seconds. Moreover, the search algorithm, which is a probabilistic roadmap can quickly recognize the maps and plan feasible paths from starting points to target points
Investigation of Heat Transfer Characteristics and Electrical Conductivities in NaCl, KCl, and NaNO3 Solutions
Full text linkThis study aims to optimize heat exchanger systems by investigating the effects of water-soluble salts (NaCl, KCl, and NaNO3) on heat transfer rates and electrical conductivity. Experiments are conducted using plate-type (PHE) and double-pipe (DPHE) heat exchangers. The heat transfer coefficient ranged from 1.5–6.5 kW/m²K in PHE and 3.5–20 kW/m²K in DPHE. NaCl achieves the highest heat transfer rates, followed by KCl and NaNO3, all outperforming pure water. Electrical conductivity peaks at 1 MHz, decreasing afterward, with NaCl and KCl showing higher conductivity than NaNO3. Conductivity increases with temperature, peaking at 70°C, and is more sensitive to temperature for KCl and NaCl. This dual-focus study correlates thermal and electrical properties, illustrating how variations in salt type, concentration, and temperature influence ion behavior, which plays a critical role in optimizing industrial heat transfer and electrical conductivity processes
Finite Element Analysis of Ti-6Al-4V Lattice Cubic Scaffolds for Mandibular Bone Implant Applications
Full text linkThis study evaluates the compressive strength of a cubic lattice scaffold made from Titanium alloy (Ti-6Al-4V) for mandibular bone implants. Scaffold designs with pore sizes ranging from 800 µm to 1000 µm were analyzed using finite element analysis under compressive forces of up to 800 N. Pore sizes of 800 µm and 850 µm achieved a safety factor greater than 1.4, indicating their suitability for both dynamic and static loading. Planned production with bound metal deposition, maintaining a density below 35%, emphasizes material efficiency and cost-effectiveness. Results indicate that 800 µm and 850 µm pore sizes offer optimal strength and safety, suggesting effective mandibular implant integration. Further research on cyclic load testing and osseointegration is recommended
Factors Affecting the Mechanical Properties of Precast Concrete Infill Walls
Full text linkPrecast concrete infill walls are widely applied to enhance the lateral stiffness and seismic performance of reinforced concrete frames. This study aims to establish a quantitative understanding of how key design parameters influence the mechanical behavior of precast concrete infill wall systems. To achieve this objective, nonlinear finite element analyses validated against ATENA-based experimental results were conducted to examine the effects of wall aspect ratio, thickness, and tie reinforcement configuration on system-level stiffness, strength, and ductility. Results show that decreasing the aspect ratio from 0.67 to 0.47 increases lateral stiffness by approximately 15-20% but reduces ductility by about 10%. Increasing wall thickness from 100 mm to 200 mm enhances peak load capacity by up to 30% while shifting damage from the infill wall to the frame. Denser wall-column ties improve residual load capacity by 18-25%, whereas wider wall–beam tie spacing slightly reduces ductility without significantly affecting peak load