DSpace collection: 期刊論文
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/822
The collection's search engineSearch the Channels
https://tkuir.lib.tku.edu.tw/dspace/simple-search
Recognition of Glaucomatous Fundus Images Using Machine Learning Methods Based on Optic Nerve Head Topographic Features
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125325
title: Recognition of Glaucomatous Fundus Images Using Machine Learning Methods Based on Optic Nerve Head Topographic FeaturesEfficiet Encrypted Image Retrieval and Management Mechanisms
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125324
title: Efficiet Encrypted Image Retrieval and Management MechanismsEffect of Hypoalbuminemia on Mortality in Cirrhotic Patients with Spontaneous Bacterial Peritonitis
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125248
title: Effect of Hypoalbuminemia on Mortality in Cirrhotic Patients with Spontaneous Bacterial Peritonitis abstract: Objectives:
The impact of hypoalbuminemia on the short-term and long-term mortality of cirrhotic patients with spontaneous bacterial peritonitis (SBP), both with and without renal function impairment, remains insufficiently elucidated based on population-based data.
Materials and Methods:
We retrieved data from Taiwan’s National Health Insurance Database encompassing 14,583 hospitalized patients diagnosed with both cirrhosis and SBP during the period from January 1, 2010, to December 31, 2013. Prognostic factors influencing 30-day and 3-year survival were computed. Furthermore, the impact of hypoalbuminemia on the mortality rate among SBP patients, with or without concurrent renal function impairment, was also assessed.
Results:
The 30-day mortality rates for patients with SBP, comparing those with hypoalbuminemia and those without, were 18.3% and 29.4%, respectively (P < 0.001). Similarly, the 3-year mortality rates for SBP patients with hypoalbuminemia and those without were 73.7% and 85.8%, respectively (P < 0.001). Cox proportional hazard regression analysis, adjusted for patients’ gender, age, and comorbid conditions, substantiated that individuals with hypoalbuminemia exhibit an inferior 30-day survival (hazard ratio [HR]: 1.62, 95% confidence interval [CI]: 1.51–1.74, P < 0.001) and reduced 3-year survival (HR: 1.57, 95% CI: 1.50–1.63, P < 0.001) in comparison to those lacking hypoalbuminemia. Among SBP patients with renal function impairment, those presenting hypoalbuminemia also experienced diminished 30-day survival (HR: 1.81, 95% CI 1.57–2.07, P < 0.001) as well as reduced 3-year survival (HR: 1.70, 95% CI 1.54–1.87, P < 0.001). Likewise, in SBP patients without renal function impairment, the presence of hypoalbuminemia was associated with poorer 30-day survival (HR: 1.54, 95% CI 1.42–1.67, P < 0.001) and 3-year survival (HR: 1.53, 95% CI 1.46–1.60, P < 0.001).
Conclusion:
Among cirrhotic patients with SBP, the presence of hypoalbuminemia predicts inferior short-term and long-term outcomes, regardless of renal function.
<br>Prognostic Factors of Cirrhotic Patients with Invasive Fungal Infections
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125247
title: Prognostic Factors of Cirrhotic Patients with Invasive Fungal Infections abstract: Fungal infection (FI) is a life-threatening condition in cirrhotic patients. However, a population-based study is required to determine the short-term mortality of these patients. The Taiwan National Health Insurance Database was used to enroll 1214 cirrhotic patients with FIs who were hospitalized between January 1, 2010 and December 31, 2013. Among them, 165 were diagnosed with invasive FIs. The overall 30-day and 90-day mortality rates for patients with invasive FIs were 25.7% and 49.9%, respectively (P < .001). After adjusting for sex, age, and other comorbidities, the following 90-day mortality prognostic factors were statistically different: renal function impairment (hazard ratio = 1.98, 95% confidence interval = 1.05-3.70, P = .034), concurrent with bacterial infections (hazard ratio = 1.75, 95% CI = 1.07-2.88, P = .027). Half of the cirrhotic patients died within 90-daysdue to invasive FIs, highlighting the importance of renal function impairment and concurrent with bacterial infections as an important prognostic factor.
<br>Effects of Taiwan's COVID-19 Alert Levels on the Physical Activity Behaviors and Psychological Distress of Community-dwelling Older Adults
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125246
title: Effects of Taiwan's COVID-19 Alert Levels on the Physical Activity Behaviors and Psychological Distress of Community-dwelling Older Adults abstract: Background
The Taiwanese government implemented stringent preventative health measures to curb the spread of COVID-19. However, these measures negatively affected the physical activity behaviors and psychological distress of individuals. In this study, we investigated the effects of Taiwan’s COVID-19 alert–based restrictions on the physical activity behaviors and psychological distress of community-dwelling older adults.
Methods
In this longitudinal study, 500 community-dwelling older adults were randomly sampled from a health promotion center in Taiwan. Telephone interviews were conducted between May 11, 2021, and August 17, 2021, which coincided with the Level 3 alert period when group physical activities were prohibited. Telephone interviews were again conducted between June 20, 2022, and July 4, 2022, after the alert level was reduced to Level 2 but group physical activities were prohibited period. Through the telephone interviews, data regarding the participants’ physical activity behaviors (type and amount) and 5-item Brief Symptom Rating Scale (BSRS-5) scores were collected. Moreover, data regarding physical activity behaviors were collected from the records of our previous health promotion programs, which were conducted before the national alert period. The obtained data were analyzed.
Results
The alert levels influenced physical activity behaviors. Because of strict regulations, physical activity amount decreased during the Level 3 alert period and did not recover rapidly during the Level 2 alert period. Instead of engaging in group exercises (e.g., calisthenics and qigong), the older adults chose to exercise alone (e.g., strolling, brisk walking, and biking). Our findings indicate that the COVID-19 alert level has a significant influence on the amount of physical activity for participants (p < 0.05, partial η2 = 0.256), with pairwise comparisons showing that the physical activity amount decreased significantly across the three time periods (p < 0.05). The psychological distress of the participants did not appear to change during the regulation period. Although the participants' overall BSRS-5 score was slightly lower during the Level 2 alert period compared to the Level 3 alert period, the difference was not statistically significant (p = 0.264, Cohen's d = 0.08) based on a paired t-test. However, the levels of anxiety (p = 0.003, Cohen's d = 0.23) and inferiority (p = 0.034, Cohen's d = 0.159) were considerably higher during the Level 2 alert period than during the Level 3 alert period.
Conclusions
Our findings indicate that Taiwan’s COVID-19 alert levels influenced the physical activity behaviors and psychological distress of community-dwelling older adults. Time is required for older adults to regain their prior status after their physical activity behaviors and psychological distress were affected by national regulations.
<br>偏常態加速破壞衰變模型之貝氏方法
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125245
title: 偏常態加速破壞衰變模型之貝氏方法 abstract: 高可靠度產品使用傳統的加速壽命試驗技巧,在給定合理測試時間內,往往無法獲得足夠的失效資料,使得產品壽命推估上造成困難。加速衰變試驗(accelerated degradation test, ADT)測量產品隨著時間衰變的品質特徵值(quality characteristics, QC),藉由這些品質特徵值衰變訊息,可提供較精準的產品壽命推估。某些特定實驗的樣本量測,需經過破壞測試樣本,才可測得其品質特徵值,此加速衰變資料的實驗,稱之加速破壞衰變試驗(accelerated destructive degradation test, ADDT)。針對聚合物材料(polymer material)之ADDT資料,所建構出的非線性偏常態ADDT衰變模型,本文以貝氏方法(Bayesian method),提高估計的精確度,進而有效地估得產品壽命,與描述聚合物材料的衰變路徑。最後以模擬分析方式,與傳統的最大概似估計法(maximum likelihood estimator, MLE)進行比較,來探討所提出模型之參數估計的精準性。
<br>MASPP and MWASP: Multi-Head Self-Attention Based Modules for UNet Network in Melon Spot Segmentation
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125172
title: MASPP and MWASP: Multi-Head Self-Attention Based Modules for UNet Network in Melon Spot SegmentationSV2-SQL: A Text-to-SQL Transformation Mechanism Based on BERT Models for Slot Filling, Value Extraction and Verification
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125171
title: SV2-SQL: A Text-to-SQL Transformation Mechanism Based on BERT Models for Slot Filling, Value Extraction and Verification abstract: Information retrieval from databases is challenging for a non-SQL-domain expert. Some previous studies have provided solutions for translating the natural language to SQL instruction, aiming to access the information in the database directly. However, most solutions are in English Natural Language. In addition, the accuracies of the existing works still need to be improved. This work presents a mechanism called SV2-SQL, based on the pre-trained BERT. The proposed SV2-SQL mainly consists of multiple deep-learning models, including select-where slot filling model (SWSF-model), value extraction model (VE-model), and verification (V-model). The SWSF-model handles the classification tasks for those fields that appear in the “Select” and “Where” clauses, and the VE-model extracts the values for the “Where” clause from the input. The V-model sorts out the unwanted candidates from two previous models and leaves only the ones with the highest possibility. The proposed SV2-SQL also includes an algorithm for the inference process and allows the three models to be cooperative. Experimental results show that the proposed SV2-SQL outperforms the existing studies in terms of precision, accuracy, and recall.
<br>A Recharging Mechanism for Improving Data Quality and Maintaining Sustainable Lifetime in WRSNs
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125170
title: A Recharging Mechanism for Improving Data Quality and Maintaining Sustainable Lifetime in WRSNs abstract: The wireless recharging technology of wireless sensor networks (WSNs) has received much attention in recent years. In literature, many studies assumed that the sensors are static and aimed to increase the number of recharged sensors and reduce the traveling path of the mobile charger. Since the given WSNs consisted of a large number of sensors that were usually deployed in a wide monitoring region, it is time-consuming for the mobile charger to visit all sensors and recharge them. This article considers the mobile wireless rechargeable sensor networks (MWRSNs) where all sensors are assumed to be mobile, and hence, they can adopt mobility to reduce the overhead of the mobile charger in terms of traveling cost. The proposed recharging mechanism initially determines the number and the locations of recharging stations. Then, the proposed algorithm further exploits the cooperation opportunities between the sensors and the mobile charger to improve recharging efficiency and data quality. In addition, the proposed algorithm further finds the bottleneck location, which has the lowest data quality and then relocates the sensors according to their workloads, aiming to maximize the data quality by breaking the bottleneck. The experimental results reveal that the proposed algorithm outperforms the related studies in terms of recharging efficiency and data quality.
<br>JCF: Joint Coarse and Fine-Grained Similarity Comparison for Plagiarism Detection Based on NLP
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125169
title: JCF: Joint Coarse and Fine-Grained Similarity Comparison for Plagiarism Detection Based on NLP abstract: Document similarity recognition is one of the most important problems in natural language processing. This paper proposes a plagiarism comparison mechanism called JCF. Initially, the TF–IDF scheme is applied to build a bag of words as the representation of the common features of all documents. Then, the plagiarism comparison is carried out in a coarse-grained manner, which speeds up the similarity comparison. Finally, the most similar documents can then be compared in detail based on a fine-grained approach. In addition, the JCF detects plagiarism at both syntax level and semantic-like level. To prevent the distortion of similarity comparison, this paper further develops a similarity restoration approach such that the proposed JCF can obtain both advantages of quickness and accuracy. Performance studies confirm that the proposed JCF outperforms existing studies in terms of precision, recall and F1 score.
<br>RLR: Joint Reinforcement Learning and Attraction Reward for Mobile Charger in Wireless Rechargeable Sensor Networks
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125168
title: RLR: Joint Reinforcement Learning and Attraction Reward for Mobile Charger in Wireless Rechargeable Sensor Networks abstract: Advances in wireless charging technology give great new opportunities for extending the lifetime of a wireless sensor network (WSN) which is an important infrastructure of IoT. However, the existing greedy algorithms lacked learning from the experiences of energy dissipation trends. Unlike the existing studies, this article proposes a reinforcement learning approach, called reinforcement learning recharging (RLR), for mobile charger to learn the trends of WSNs, including the energy consumption of the sensors, the recharging cost as well as the coverage benefit, aiming to maximize the coverage contribution of the recharged WSN. The proposed RLR mainly consists of three modules, including sensor energy management (SEM), charger location update (CLP), and charger reinforcement learning (CRL) modules. In the SEM module, each sensor manages its energy and calculates its threshold for the recharging request in a distributed manner. The CLP module adopts the quorum system to ensure effective communication between sensors and the mobile charger. Meanwhile, the CRL module employs attraction rewards to reflect the coverage benefit and penalties of waiting time raised due to charger movement and recharging other sensors. As a result, the charger accumulates the learning experiences from the Q -Table such that it is able to execute the appropriate actions of charging or moving in a manner of state management. Performance results show that the proposed RLR outperforms the existing recharging mechanisms in terms of charging waiting time of sensors, the energy usage efficiency of the mobile charger, as well as the coverage contribution of the given sensor network.
<br>WLARS: Work Load Aware Recharge Scheduling Mechanism for Improving Surveillance Quality in Wireless Rechargeable Sensor Networks
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125167
title: WLARS: Work Load Aware Recharge Scheduling Mechanism for Improving Surveillance Quality in Wireless Rechargeable Sensor Networks abstract: With radio frequency (RF), the mobile charger (MC) can wirelessly transmit energy to the sensor nodes in the network. The wireless energy transfer enhances the lifetime of sensor nodes in wireless rechargeable sensor networks (WRSNs). Most of the studies improved the efficiency of MC or perpetuated the lifetime of WRSNs. At the same time, none of the studies focused on considering the spatial and temporal surveillance qualities (STSQs) of each sensor recharged. Therefore, this article proposed an efficient energy recharge scheduling for MC, aiming to maximize the STSQ of the given network. Initially, sensor routing load (RL) is considered to partition the network to distribute the charging load of the MCs evenly. Second, the busy level of the MC is considered to send the recharging requests to the MC. Based on the busy level of MC, the sensors adjust the sensing rate frequency to manage their energy. Finally, cooperation between the neighboring MCs is proposed to reduce the waiting time for recharging requested sensors. The simulation results present that the proposed work yields the literature in terms of STSQ, traveling distance, average waiting time (AWT), and energy usage efficiency.
<br>JECP: Joint Energy Conservation and Collision Avoidance for Path Construction in Bluetooth Networks
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125166
title: JECP: Joint Energy Conservation and Collision Avoidance for Path Construction in Bluetooth Networks abstract: With the growing elderly population of the world, the number of elderly people who live alone is continuously growing. Providing them with high-quality healthcare has become a major concern nowadays. In order to detect the elderly people’s sudden danger of living alone, this article proposes a path construction mechanism, called JECP, which aims to conserve energy and avoid collision for the emergent sensors in Bluetooth mesh networks. In case that the emergent sensors are deployed in the same room, the proposed JECP constructs disjoint paths to avoid collisions, which are triggered by the same emergent event at the same time. The other case is that the emergent sensors are deployed in different rooms. The probability that these emergent sensors are triggered by the same event is very small. Hence, the proposed JECP constructs the best emergent path that shares the general sensors as much as possible to save energy consumption. The experimental results show that the proposed JECP outperforms the existing mechanisms in terms of transmission delay and energy consumption for the emergent sensors in Bluetooth low-energy (BLE) networks.
<br>CSCP: Energy Charging Mechanism for Surveillance Quality, Network Connectivity and Perpetual Lifetime in WRSNs
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125165
title: CSCP: Energy Charging Mechanism for Surveillance Quality, Network Connectivity and Perpetual Lifetime in WRSNs abstract: In recent years, wireless charging techniques applied to sensor networks have been widely studied. The sensors in most studies are considered to be equally important, and the main purpose is to maintain the working sensors as many as possible. However, different sensors have different contributions to the surveillance application, especially for network connectivity and surveillance quality. The sensor closer to the base station has a larger contribution to network connectivity since its failure can block more data transmissions. On the other hand, a sensor in a sparse area has a larger contribution to surveillance quality because few or no neighboring sensors can execute the sensing operation instead of the sensor if it is energy exhaustion. This paper proposes an energy recharging mechanism, called CSCP, which adopts the mobile charger to recharge the sensors according to the recharging requests. The proposed CSCP considers the contribution of each grid in terms of network connectivity, surveillance quality as well as path cost, aiming to recharge the grids where the sensors can maximize the surveillance quality. Performance research reveals that the proposed CSCP has better performance than the related studies in the surveillance quality, the number of working sensors, and recharging efficiency.
<br>Intelligent Performance Evaluation in Rowing Sport Using a Graph-Matching Network
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125164
title: Intelligent Performance Evaluation in Rowing Sport Using a Graph-Matching Network abstract: Rowing competitions require consistent rowing strokes among crew members to achieve optimal performance. However, existing motion analysis techniques often rely on wearable sensors, leading to challenges in sporter inconvenience. The aim of our work is to use a graph-matching network to analyze the similarity in rowers’ rowing posture and further pair rowers to improve the performance of their rowing team. This study proposed a novel video-based performance analysis system to analyze paired rowers using a graph-matching network. The proposed system first detected human joint points, as acquired from the OpenPose system, and then the graph embedding model and graph-matching network model were applied to analyze similarities in rowing postures between paired rowers. When analyzing the postures of the paired rowers, the proposed system detected the same starting point of their rowing postures to achieve more accurate pairing results. Finally, variations in the similarities were displayed using the proposed time-period similarity processing. The experimental results show that the proposed time-period similarity processing of the 2D graph-embedding model (GEM) had the best pairing results.
<br>Video Based Basketball Shooting Prediction and Pose Suggestion System
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125163
title: Video Based Basketball Shooting Prediction and Pose Suggestion System abstract: Video based motion analysis, which aims to acquire the whole posture data by simple camera and without placing sensors on the body parts, has become the major analysis method in the sport domain. However, most video based motion analysis approaches either work only for some specific domain action recognition, or suffer from low prediction rates for practical applications in the sport domain. This paper presents an effective system to predict basketball shooting and to suggest corrected postures, as based on video based motion analysis with the OpenPose system. Given a basketball shooting video sequences, the proposed system first detects the human joint points acquired from the OpenPose system, and then, the video frames of the shooting period are detected by two important features of the shooting process. Basketball shooting is predicted using the adopted trajectory curves matching method and the K-nearest neighbor classification method. Finally, the wrong shooting posture is corrected and suggested based on the pix2pix conditional GAN (cGAN) model. Experimental results show that our approach can effectively estimate shooting results with high accuracy.
<br>Detection of coronary lesions in Kawasaki disease by Scaled-YOLOv4 with HarDNet backbone
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125162
title: Detection of coronary lesions in Kawasaki disease by Scaled-YOLOv4 with HarDNet backbone abstract: Introduction: Kawasaki disease (KD) may increase the risk of myocardial infarction or sudden death. In children, delayed KD diagnosis and treatment can increase coronary lesions (CLs) incidence by 25% and mortality by approximately 1%. This study focuses on the use of deep learning algorithm-based KD detection from cardiac ultrasound images.
Methods: Specifically, object detection for the identification of coronary artery dilatation and brightness of left and right coronary artery is proposed and different AI algorithms were compared. In infants and young children, a dilated coronary artery is only 1-2 mm in diameter than a normal one, and its ultrasound images demonstrate a large amount of noise background-this can be a considerable challenge for image recognition. This study proposes a framework, named Scaled-YOLOv4-HarDNet, integrating the recent Scaled-YOLOv4 but with the CSPDarkNet backbone replaced by the CSPHarDNet framework.
Results: The experimental result demonstrated that the mean average precision (mAP) of Scaled-YOLOv4-HarDNet was 72.63%, higher than that of Scaled YOLOv4 and YOLOv5 (70.05% and 69.79% respectively). In addition, it could detect small objects significantly better than Scaled-YOLOv4 and YOLOv5.
Conclusions: Scaled-YOLOv4-HarDNet may aid physicians in detecting KD and determining the treatment approach. Because relatively few artificial intelligence solutions about images for KD detection have been reported thus far, this paper is expected to make a substantial academic and clinical contribution.
<br>On the Identifiability of Artificial Financial Time Series
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125089
title: On the Identifiability of Artificial Financial Time Series abstract: Financial time series are often considered to be difficult to model and unlikely to predict. In this study, we assume that financial time series are based on a stochastic series generated by a Markov decision process. Based on this assumption, we investigate two problems related to the identification of the price time series of financial instruments. We try to distinguish the real price-volume time series from the artificial one. First, we investigate whether there is any machine learning model that can distinguish between real price-volume time series and those with time horizon reversed. Then, we investigate whether there is any machine learning model that can distinguish the price-volume time series from the real one when they are subjected to random manipulations of different proportions. The data we use are the daily prices and trading volumes of six U.S. stocks and one crypto-currency BTC/USD. We apply Long-Short Term Memory (LSTM) as the main machine learning model for the binary classification due to its success in fitting time series data. Based on the experimental results, we give positive answers to the above two questions. Our results also partially support the conjecture that the dynamics of a financial time series are driven by an underlying Markov decision processes.
<br>Intelligent Performance Evaluation in Rowing Sport Using a Graph-Matching Network
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125045
title: Intelligent Performance Evaluation in Rowing Sport Using a Graph-Matching Network abstract: Rowing competitions require consistent rowing strokes among crew members to achieve optimal performance. However, existing motion analysis techniques often rely on wearable sensors, leading to challenges in sporter inconvenience. The aim of our work is to use a graph-matching network to analyze the similarity in rowers’ rowing posture and further pair rowers to improve the performance of their rowing team. This study proposed a novel video-based performance analysis system to analyze paired rowers using a graph-matching network. The proposed system first detected human joint points, as acquired from the OpenPose system, and then the graph embedding model and graph-matching network model were applied to analyze similarities in rowing postures between paired rowers. When analyzing the postures of the paired rowers, the proposed system detected the same starting point of their rowing postures to achieve more accurate pairing results. Finally, variations in the similarities were displayed using the proposed time-period similarity processing. The experimental results show that the proposed time-period similarity processing of the 2D graph-embedding model (GEM) had the best pairing results.
<br>Effective Document Image Rectification via a Deep Learning Framework
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125032
title: Effective Document Image Rectification via a Deep Learning FrameworkUnsupervised Domain Adaptation Deep Network Based on Discriminative Class-Wise MMD
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/125031
title: Unsupervised Domain Adaptation Deep Network Based on Discriminative Class-Wise MMD abstract: General learning algorithms trained on a specific dataset often have difficulty generalizing effectively across different domains. In traditional pattern recognition, a classifier is typically trained on one dataset and then tested on another, assuming both datasets follow the same distribution. This assumption poses difficulty for the solution to be applied in real-world scenarios. The challenge of making a robust generalization from data originated from diverse sources is called the domain adaptation problem. Many studies have suggested solutions for mapping samples from two domains into a shared feature space and aligning their distributions. To achieve distribution alignment, minimizing the maximum mean discrepancy (MMD) between the feature distributions of the two domains has been proven effective. However, this alignment of features between two domains ignores the essential class-wise alignment, which is crucial for adaptation. To address the issue, this study introduced a discriminative, class-wise deep kernel-based MMD technique for unsupervised domain adaptation. Experimental findings demonstrated that the proposed approach not only aligns the data distribution of each class in both source and target domains, but it also enhances the adaptation outcomes.
<br>Effectively Learn How to Learn: A Novel Few-Shot Learning with Meta-Gradient Memory
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/124787
title: Effectively Learn How to Learn: A Novel Few-Shot Learning with Meta-Gradient MemoryA Novel Multipath QUIC Protocol with Minimized Flow Complete Time for Internet Content Distribution
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/124786
title: A Novel Multipath QUIC Protocol with Minimized Flow Complete Time for Internet Content Distribution abstract: The rapid growth of network services and applications has led to an exponential increase in data flows on the internet. Given the dynamic nature of data traffic in the realm of internet content distribution, traditional TCP/IP network systems often struggle to guarantee reliable network resource utilization and management. The recent advancement of the Quick UDP Internet Connect (QUIC) protocol equips media transfer applications with essential features, including structured flow controlled streams, quick connection establishment, and seamless network path migration. These features are vital for ensuring the efficiency and reliability of network performance and resource utilization, especially when network hosts transmit data flows over end-to-end paths between two endpoints. QUIC greatly improves media transfer performance by reducing both connection setup time and transmission latency. However, it is still constrained by the limitations of single-path bandwidth capacity and its variability. To address this inherent limitation, recent research has delved into the concept of multipath QUIC, which utilizes multiple network paths to transmit data flows concurrently. The benefits of multipath QUIC are twofold: it boosts the overall bandwidth capacity and mitigates flow congestion issues that might plague individual paths. However, many previous studies have depended on basic scheduling policies, like round-robin or shortest-time-first, to distribute data transmission across multiple paths. These policies often overlook the subtle characteristics of network paths, leading to increased link congestion and transmission costs. In this paper, we introduce a novel multipath QUIC strategy aimed at minimizing flow completion time while taking into account both path delay and packet loss rate. Experimental results demonstrate the superiority of our proposed method compared to standard QUIC, Lowest-RTT-First (LRF) QUIC, and Pluginized QUIC schemes. The relative performance underscores the efficacy of our design in achieving efficient and reliable data transfer in real-world scenarios using the Mininet simulator.
<br>Arbitrary Style Transfer System with Split-and-Transform Scheme
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/124674
title: Arbitrary Style Transfer System with Split-and-Transform SchemeGroup Formation by Group Joining and Opinion Updates via Multi-Agent Online Gradient Ascent
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/124668
title: Group Formation by Group Joining and Opinion Updates via Multi-Agent Online Gradient Ascent abstract: This article aims to exemplify best-response dynamics and multi-agent online learning by group formation. This extended abstract provides a summary of the full
paper in IEEE Computational Intelligence Magazine on the special issue AI-eXplained (AI-X). The full paper
includes interactive components to facilitate interested readers to grasp the idea of pure-strategy Nash equilibria and how the system of strategic agents
converges to a stable state by the decentralized online gradient ascent with and without regularization.
<br>IoT-interfaced solid-contact ion-selective electrodes for cyber-monitoring of element-specific nutrient information in hydroponics
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/124565
title: IoT-interfaced solid-contact ion-selective electrodes for cyber-monitoring of element-specific nutrient information in hydroponics abstract: This study aims to monitor element-specific nutrient information during hydroponic cultivation by IoT-interfaced miniaturized ion sensors. Because of size, cost, and manufacturing advantages, solid-contact ion-selective electrodes (SCISEs) were fabricated as an ion sensor array and interfaced with wireless embedded-systems to construct an IoT nutrient sensor system (IoNSS) for the first time. The entire IoNSS framework was composed of (i) a nutrient solution sampling and sensing module with SCISEs controlled by an Arduino Due® microcontroller, (ii) a Wio Terminal® microcontroller for automated procedure setting, data recording, and wireless transmission, (iii) a private cloud server (a Network Attached Storage equipped with Node-RED® and MongoDB®) for data management, and (iv) MQTT webpage-based interactive interfaces. In experiments, we found that potentiometric signal resolution and noise of the Arduino-interfaced SCISEs were significantly improved and approached to instrumental DAQ-like quality by additional delta-sigma ADC (ADS1115®) chip conditioning. This facilitated cost-effective harvest of precise and high-quality IoT ion sensor data. Before on-site applications, each SCISE was two-point calibrated in multiple-ion solutions and was checked with the fixed interference method. The ion concentration measurements were also compared with those of commercial ISEs and ion chromatography. To test the system’s feasibility, the IoNSS was applied to cyber-monitoring of K+, NO3-, and NH4+ concentrations during two-week hydroponic cultivation of arugula (E. vesicaria) in an indoor plant factory (in Northern Taiwan) with a modified Cornell solution and an outdoor greenhouse (in Southern Taiwan) with a modified Yamasaki solution, respectively. It was demonstrated that the IoNSS was capable of real-time observation of the crop’s nitrate/ammonium utilization and nutrient solution’s EC-element dependency. Besides, the web interfaces successfully reported growing condition-dependent ion signals in a simultaneous and remote manner. To sum up, this work achieves novel cyber-monitoring of element-specific nutrient by IoT-interfaced SCISEs and paves a promising way for intelligent hydroponic management.
<br>自動化骨頭年齡分析系統
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/124451
title: 自動化骨頭年齡分析系統 abstract: 在現今社會中,身高是許多人衡量完美的條件之一,其重點莫過於看骨頭之生長板有無癒合,是否還會增長。一般生長板會依年齡成長而癒合,癒合完全後骨頭會停止增長。在骨齡判斷中,最有名為GP法,由Greulich及Pyle等人在1959年發表,以不同年齡的區隔,給予一張正常的左手掌、左手腕X光片當作參考片,去比較出骨齡受檢者X光片的差異,並給予分數供診斷。本研究將設計一套可辨析手掌骨格結構之影像處理模組,來去除20筆選定之標準手部X-ray造影影像的軟組織。從結果得知可消除較薄的軟組織,並保留骨質部分,再經後續分析流程,有效辨別出正確之骨骼年齡。
<br>Non-destructive classification of melon sweetness levels using segmented rind properties based on semantic segmentation models
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/124401
title: Non-destructive classification of melon sweetness levels using segmented rind properties based on semantic segmentation models abstract: Melon is one of the most consumed crops worldwide and has high marketability. Consumers prefer sweet melons. However, the nondestructive determination of melon sweetness is challenging because of its thick rind. In this study, we presented a novel approach for predicting melon sweetness levels using features extracted from segmented rind images and machine learning techniques. We extracted various features from melon rinds images, such as the net density, net thickness, and rind color, using a semantic segmentation model. These features were used as factors in grading melon quality. Experiments on various machine learning models showed that the one-dimensional convolutional neural network model achieved the best performance with 85.71% accuracy, 96.00% precision, and 87.27% F-score. Moreover, It indicated that the sweetness classification performance over a binary class (combining sweet and ‘very sweet’ classes into one class) achieved better result than over multiple classes.
<br>Sentiment Analysis by Lexical Analysis Combined with Machine Learning, International Journal of Advances in Soft Computing & Its Applications
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/124400
title: Sentiment Analysis by Lexical Analysis Combined with Machine Learning, International Journal of Advances in Soft Computing & Its Applications abstract: This paper aims to sentiment analysis users evaluate products using lexical analysis method combined with machine learning. This study analyze the emotions of users through analyzing their comments and evaluations for information posted or shared about services and products at the Explorascience QuyNhon. The first we retrieve the data of the user's product review comments and then build a Vietnamese emotional dictionary using vocabulary-based methods by calculating the semantic value of words or phrases in documents, finally, using a machine learning model to analyze and evaluate emotions with two problems: classifying sentences with feelings or without emotions and classifying sentences with positive or negative emotions. With input is a set of raw Vietnamese comments of users our methods will return outputs are Vietnamese comments which have been classified into three categories: without, positive or negative emotions. The input data will be a set of Vietnamese comments that are then evaluated and then put into processing Vietnamese errors with accents, processing emoticons, processing stop words collectively referred to as preprocessing. After the preprocessing has been standardized, the system begins to extract the characteristics of each sentence based on the emotion dictionary and the factors affecting the emotions in the sentence. From the characteristics obtained, subjective classification and emotional classification of comment sets to finally output the set of comments are classified into three categories: without emotions, with positive emotions or with negative emotions by machine learning model.
<br>Video based basketball shooting prediction and pose suggestion system
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/124377
title: Video based basketball shooting prediction and pose suggestion system abstract: Video based motion analysis, which aims to acquire the whole posture data by simple camera and without placing sensors on the body parts, has become the major analysis method in the sport domain. However, most video based motion analysis approaches either work only for some specific domain action recognition, or suffer from low prediction rates for practical applications in the sport domain. This paper presents an effective system to predict basketball shooting and to suggest corrected postures, as based on video based motion analysis with the OpenPose system. Given a basketball shooting video sequences, the proposed system first detects the human joint points acquired from the OpenPose system, and then, the video frames of the shooting period are detected by two important features of the shooting process. Basketball shooting is predicted using the adopted trajectory curves matching method and the K-nearest neighbor classification method. Finally, the wrong shooting posture is corrected and suggested based on the pix2pix conditional GAN (cGAN) model. Experimental results show that our approach can effectively estimate shooting results with high accuracy.
<br>Modified YOLOv4-DenseNet Algorithm for Detection of Ventricular Septal Defects in Ultrasound Images
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/124282
title: Modified YOLOv4-DenseNet Algorithm for Detection of Ventricular Septal Defects in Ultrasound Images abstract: Doctors conventionally analyzed echocardiographic images for diagnosing congenital heart diseases (CHDs). However, this process is laborious and depends on the experience of the doctors. This study investigated the use of deep learning algorithms for the image detection of the ventricular septal defect (VSD), the most common type. Color Doppler echocardiographic images containing three types of VSDs were tested with color doppler ultrasound medical images. To the best of our knowledge, this study is the first one to solve this object detection problem by using a modified YOLOv4–DenseNet framework. Because some techniques of YOLOv4 are not suitable for echocardiographic object detection, we revised the algorithm for this problem. The results revealed that the YOLOv4–DenseNet outperformed YOLOv4, YOLOv3, YOLOv3–SPP, and YOLOv3–DenseNet in terms of metric mAP-50. The F1-score of YOLOv4-DenseNet and YOLOv3-DenseNet were better than those of others. Hence, the contribution of this study establishes the feasibility of using deep learning for echocardiographic image detection of VSD investigation and a better YOLOv4-DenseNet framework could be employed for the VSD detection.
<br>Joint Hamming Coding for High Capacity Lossless Image Encryption and Embedding Scheme
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/124281
title: Joint Hamming Coding for High Capacity Lossless Image Encryption and Embedding Scheme abstract: Encryption is a widely used solution to prevent privacy leakage and illegal spread when sensitive images are uploaded to cloud storage. Hiding technology also allows confidential data to be embedded into encrypted images for secret communication. As image accuracy without distortion is essential within certain fields (such as medicine and the military), sensitive images must be completely decrypted back into the original images. However, an encrypted image is a noise-like pattern that is meaningless to a user; thus, it is difficult for a user to find the accurate image they desire. Take keywords as search indexes and embed them in encrypted images for encrypted image retrieval as an example. This idea has been extended by Chen and Line’s scheme to achieve higher capacity with reversibility. The proposed scheme adjusts the coding results according to smooth and complex images to increase its hiding capacity. In addition, two thresholds are designed to adjust the predicted pixel value to be close to the original one. Experiments show that compared with the other schemes, the proposed method achieves superior results. In addition, a hidden encrypted image can be extracted from the cover image. Afterward, the hidden secrets can be completely extracted, and sensitive images can also be perfectly restored.
<br>Joint Hamming Coding for High Capacity Lossless Image Encryption and Embedding Scheme
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/124244
title: Joint Hamming Coding for High Capacity Lossless Image Encryption and Embedding Scheme abstract: Encryption is a widely used solution to prevent privacy leakage and illegal spread when sensitive images are uploaded to cloud storage. Hiding technology also allows confidential data to be embedded into encrypted images for secret communication. As image accuracy without distortion is essential within certain fields (such as medicine and the military), sensitive images must be completely decrypted back into the original images. However, an encrypted image is a noise-like pattern that is meaningless to a user; thus, it is difficult for a user to find the accurate image they desire. Take keywords as search indexes and embed them in encrypted images for encrypted image retrieval as an example. This idea has been extended by Chen and Line’s scheme to achieve higher capacity with reversibility. The proposed scheme adjusts the coding results according to smooth and complex images to increase its hiding capacity. In addition, two thresholds are designed to adjust the predicted pixel value to be close to the original one. Experiments show that compared with the other schemes, the proposed method achieves superior results. In addition, a hidden encrypted image can be extracted from the cover image. Afterward, the hidden secrets can be completely extracted, and sensitive images can also be perfectly restored.
<br>Design of Multi-Receptive Field Fusion-Based Network for Surface Defect Inspection on Hot-Rolled Steel Strip Using Lightweight Dataset
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/124119
title: Design of Multi-Receptive Field Fusion-Based Network for Surface Defect Inspection on Hot-Rolled Steel Strip Using Lightweight Dataset abstract: With the advancement of industrial intelligence, defect recognition has become an indispensable part of facilitating surface quality in the steel manufacturing process. To assure product quality, most previous studies were typically trained with many defect samples. Nonetheless, a large quantity of defect samples is difficult to obtain, owing to the rare occurrence of defects. In general, deep learning-based methods underperformed as they have inherent limitations due to inadequate information, thereby restraining the application of models. In this study, a two-level Gaussian pyramid is applied to decompose raw data into different resolution levels simultaneously filtering the noises to acquire compact and representative features. Subsequently, a multi-receptive field fusion-based network (MRFFN) is developed to learn the hierarchical features and synthesize the respective prediction scores to form the final recognition result. As a result, the proposed method is capable of exhibiting an outstanding performance of 99.75% when trained using a lightweight dataset. In addition, the experiments conducted using the disturbance defect dataset showed the robustness of the proposed MRFFN against common noises and motion blur.
<br>Risk assessment of metabolic syndrome prevalence involving sedentary occupations and socioeconomic status
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/124118
title: Risk assessment of metabolic syndrome prevalence involving sedentary occupations and socioeconomic status abstract: Objectives
To determine whether occupation type, distinguished by socioeconomic status (SES) and sedentary status, is associated with metabolic syndrome (MetS) risk.
Methods
We analysed two data sets covering 73 506 individuals. MetS was identified according to the criteria of the modified Adult Treatment Panel III. Eight occupational categories were considered: professionals, technical workers, managers, salespeople, service staff, administrative staff, manual labourers and taxi drivers; occupations were grouped into non-sedentary; sedentary, high-SES; and sedentary, non-high-SES occupations. A multiple logistic regression was used to determine significant risk factors for MetS in three age-stratified subgroups. R software for Windows (V.3.5.1) was used for all statistical analyses.
Results
MetS prevalence increased with age. Among participants aged ≤40 years, where MetS prevalence was low at 6.23%, having a non-sedentary occupation reduced MetS risk (OR=0.88, p<0.0295). Among participants aged >60 years, having a sedentary, high-SES occupation significantly increased (OR=1.39, p<0.0247) MetS risk.
Conclusions
The influence of occupation type on MetS risk differs among age groups. Non-sedentary occupations and sedentary, high-SES occupations decrease and increase MetS risk, respectively, among younger and older adults, respectively. Authorities should focus on individuals in sedentary, high-SES occupations.
<br>Tamper detection and recovery for medical images using near-lossless information hiding technique
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/122981
title: Tamper detection and recovery for medical images using near-lossless information hiding technique abstract: Digital medical images are very easy to be modified for illegal purposes. For example, microcalcification in mammography is an important diagnostic clue, and it can be wiped off intentionally for insurance purposes or added intentionally into a normal mammography. In this paper, we proposed two methods to tamper detection and recovery for a medical image. A 1024 × 1024 x-ray mammogram was chosen to test the ability of tamper detection and recovery. At first, a medical image is divided into several blocks. For each block, an adaptive robust digital watermarking method combined with the modulo operation is used to hide both the authentication message and the recovery information. In the first method, each block is embedded with the authentication message and the recovery information of other blocks. Because the recovered block is too small and excessively compressed, the concept of region of interest (ROI) is introduced into the second method. If there are no tampered blocks, the original image can be obtained with only the stego image. When the ROI, such as microcalcification in mammography, is tampered with, an approximate image will be obtained from other blocks. From the experimental results, the proposed near-lossless method is proven to effectively detect a tampered medical image and recover the original ROI image. In this study, an adaptive robust digital watermarking method combined with the operation of modulo 256 was chosen to achieve information hiding and image authentication. With the proposal method, any random changes on the stego image will be detected in high probability.
<br>Image stitching and computer-aided diagnosis for whole breast ultrasound image
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/122980
title: Image stitching and computer-aided diagnosis for whole breast ultrasound image2-D ultrasound strain images for breast cancer diagnosis using nonrigid subregion registration
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/122979
title: 2-D ultrasound strain images for breast cancer diagnosis using nonrigid subregion registration abstract: Tissue elasticity of a lesion is a useful criterion for the diagnosis of breast ultrasound (US). Elastograms are created by comparing ultrasonic radio-frequency waveforms before and after a light-tissue compression. In this study, we evaluate the accuracy of continuous US strain image in the classification of benign from malignant breast tumors. A series of B-mode US images is applied and each case involves 60 continuous images obtained by using the steady artificial pressure of the US probe. In general, after compression by the US probe, a soft benign tumor will become flatter than a stiffened malignant tumor. We proposed a computer-aided diagnostic (CAD) system by utilizing the nonrigid image registration modality on the analysis of tumor deformation. Furthermore, we used some image preprocessing methods, which included the level set segmentation, to improve the performance. One-hundred pathology-proven cases, including 60 benign breast tumors and 40 malignant tumors, were used in the experiments to test the classification accuracy of the proposed method. Four characteristic values—normalized slope of metric value (NSM), normalized area difference (NAD), normalized standard deviation (NSD) and normalized center translation (NCT)—were computed for all cases. By using the support vector machine, the accuracy, sensitivity, specificity and positive and negative predictive values of the classification of continuous US strain images were satisfactory. The Az value of the support vector machine based on the four characteristic values used for the classification of solid breast tumors was 0.9358.
<br>Solid breast masses: classification with computer-aided analysis of continuous US images obtained with probe compression
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/122978
title: Solid breast masses: classification with computer-aided analysis of continuous US images obtained with probe compression abstract: PURPOSE: To prospectively evaluate the accuracy of continuous ultrasonographic (US) images obtained during probe compression and computer-aided analysis for classification of biopsy-proved (reference standard) benign and malignant breast tumors. MATERIALS AND METHODS: This study was approved by the local ethics committee, and informed consent was obtained from all included patients. Serial US images of 100 solid breast masses (60 benign and 40 malignant tumors) were obtained with US probe compression in 86 patients (mean age, 45 years; range, 20-67 years). After segmentation of tumor contours with the level-set method, three features of strain on tissue from probe compression-contour difference, shift distance, area difference-and one feature of shape-solidity-were computed. A maximum margin classifier was used to classify the tumors by using these four features. The Student t test and receiver operating characteristic curve analysis were used for statistical analysis. RESULTS: The mean values of contour difference, shift distance, area difference, and solidity were 3.52% ± 2.12 (standard deviation), 2.62 ± 1.31, 1.08% ± 0.85, and 1.70 ± 1.85 in malignant tumors and 9.72% ± 4.54, 5.04 ± 2.79, 3.17% ± 2.86, and 0.53 ± 0.63 in benign tumors, respectively. Differences with P < .001 were statistically significant for all four features. Area under the receiver operating characteristic curve (AZ) values for contour difference, shift distance, area difference, and solidity were 0.88, 0.85, 0.86, and 0.79, respectively. The AZ value of three features of strain was significantly higher than that of the feature of shape (P < .01). The accuracy, sensitivity, specificity, and positive and negative predictive values of US classifications that were based on values for these four features were 87.0% (87 of 100), 85% (34 of 40), 88% (53 of 60), 83% (34 of 41), and 90% (53 of 59), respectively, with an AZ value of 0.91. CONCLUSION: Continuous US images obtained with probe compression and computer-aided analysis can aid in classification of benign and malignant breast tumors.
<br>Classification of breast ultrasound images using fractal feature
https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/122977
title: Classification of breast ultrasound images using fractal feature abstract: Fractal analyses have been applied successfully for the image compression, texture analysis, and texture image segmentation. The fractal dimension could be used to quantify the texture information. In this study, the differences of gray value of neighboring pixels are used to estimate the fractal dimension of an ultrasound image of breast lesion by using the fractal Brownian motion. Furthermore, a computer-aided diagnosis (CAD) system based on the fractal analysis is proposed to classify the breast lesions into two classes: benign and malignant. To improve the classification performances, the ultrasound images are preprocessed by using morphology operations and histogram equalization. Finally, the k-means classification method is used to classify benign tumors from malignant ones. The US breast image databases include only histologically confirmed cases: 110 malignant and 140 benign tumors, which were recorded. All the digital images were obtained prior to biopsy using by an ATL HDI 3000 system. The receiver operator characteristic (ROC) area index AZ is 0.9218, which represents the diagnostic performance.
<br>