Chung Cheng University Institutional Repository
Not a member yet
889 research outputs found
Sort by
家事事件審理程序法理之研究－以子女扶養費之聲請為中心－;
No full text[[abstract]]本論文分成三大部分，第一部分說明家事事件法之特色，及各個家事事件類型之性質，並說明程序法理中訴訟法理及非訟法理。 第二部分，因未成年子女對父母請求扶養費之實體法依據尚屬不明，整理學說實務見解，再針對本文討論代墊型、將來型及協議型，三種類型學說實務上所認為實體上之請求權依據。 第三部分則是具體討論，扶養事件中：一、夫妻以兩願離婚自行訂立扶養契約之情況，嗣後兩方發生紛爭，法院應如何運用程序法理？二、過往代墊扶養費之請求，性質上為訴訟事件，此種事件經由非訟程序解決是否妥適？如需以非訟程序中定爭止紛，則如何妥善運用非訟法理及訴訟法理，為兩造解決紛爭？三、將來扶養費之請求，考量未成年子女之最佳利益，法院依非訟法理職權介入，法院職權範圍為何？並應如何職權調查證據以保障兩造未成年子女之利益？ 最後結論認為須視家事事件類型調和是用非訟法理或訴訟法理。再者，而未成年子女對父母請求扶養費實體法之請求權依據，並無完全構成要件法條，應有明確規定，以利實務操作及當事人主張。另代墊扶養費事件及協議型扶養費事件，相較於未來型扶養費事件，較具有訟爭性，雖當事人間仍為家庭成員，性質上直接改為一般民事訴訟案件，恐為突兀，則應分類為丙類家事訴訟事件，以明此二種家事事件類型之審理程序
論信託受託人之權利義務與責任;A Study on the Rights, Duties and Responsibilities of Trustees
No full text[[abstract]]英美「信託制度」相繼為大陸法系國家所繼受，優點之一在於將所有權與經營權分離，以期發揮專業管理績效。而信託當事人中，受託人管理信託財產有極大的自主權，但為保護信託財產之安全及防止其權利濫用，嚴格規定受託人各種義務；並課予受託人高度的注意義務及忠實義務，對於健全信託制度及保障受益人之權益實具有重大意義。受託人依信託契約或信託本旨管理處分信託財產，因此本文從受託人之定義與基本概念出發，分別介紹受託人之各項權限、權利、義務與責任，第三章開始介紹信託法中規定受託人的各項權利及學說補充的其他權利，第四章介紹受託人之義務如善管義務、分別管理義務、自己管理義務、忠實義務等等，並由學說或外國法理見解補充來填補法條疏漏的規定，第五章則探討受託人之責任以及違反時之效果及救濟方法，並藉由外國立法來分析我國信託法規定是否合理，有無修正的必要，最後結論提出傳統民事信託在我國施行的困境及改善空間。
The Anglo-American “Trust system” was inherited by civil law country. The most important advantage of the trust law is to separate the ownerships and managements in order to achieve a professional management performance; the trustee has great autonomy in the management of the trust property. However, in order to protect the security of the trust property and to prevent the abuse of its rights, the trustee's obligations are strictly stipulated, and the trustee's high degree of duty of care and loyalty are required. It is a great significance for the trust system and the protection of the beneficiaries' rights.The trustee shall manage the trust property according to the trust contract or the trust purpose. Therefore, this paper introduces the rights, duties and responsibilities of the trustee separately from the definition and the basic concept of the trustee, and the third chapter begins to introduce the trust Chapter 4 introduces the obligations of the trustee, such as duty of care, duty of loyalty, duty of segregation, etc., and by the doctrine or foreign law of the supplement to fill the law. Chapter 5 is to explore the responsibility of the trustee and the effect of the violation and remedies, and by foreign legislation to analyze the provisions of trust law in Taiwan is reasonable or not
EtherCAT在不同Kernel上的效能分析;Performance Analysis of EtherCAT for Different Kernels
No full text[[abstract]]在CNC工具機的應用中，馬達運轉的即時性和同步性是在機器運作時非常重要的一環，EtherCAT在這兩者的性能上是比起其他傳統的工業網路，如:CANopen、Sercos等來得出色，也可以和傳統的工業網路相容，結合成CoE(CANopen Over EtherCAT)、SoE(Sercos over EtherCAT)等，是現在最主流的工業網路。由於現在EtherCAT的主站軟體大多是由驅動器廠商提供，在Windows系統上才能使用，不像在Linux系統下可測試在不同版本下即時性的差異，因此本論文使用開源軟體EtherCAT IGH以及SOEM來連接馬達驅動器，並且嘗試在不同的Kernel版本下，傳輸驅動器指令的速度是否會有不同的變化，並且找尋比較適合的Kernel版本，未來如果要開發EtherCAT的相關程式時，就可以選擇最適合的Kernel版本去做開發，以達到最佳的即時性。
In the application of CNC machine, the immediacy and synchronization of the motor is very important, EtherCAT excels in both performance over other traditional industrial networks such as CANopen, Sercos, and it can compatible with tranditional industrial networks, combine into CoE(CANopen Over EtherCAT)、SoE(Sercos over EtherCAT) and so on, is most mainstream industrial network.Because EtherCAT’s master software is now mostly provided by the manufacturer, can only use in Windows system, can't test the immediacy different between kernels like Linux system. This thesis uses the open source software EtherCAT IGH and SOEM to connect motor driver. And try different versions of Kernels, the transmission speed of the driver instructions will have different changes, find the suitable Kernel version. In the future, if we want to develop EtherCAT programs, we can choose the most suitable Kernel version to get the best immediacy
工業自動化生產之碰撞偵測技術研發;Collision Detection of Industrial Automation
No full text[[abstract]]自古以來，工業技術的進步一直是人類追求的目標，也是改善人類生活品質的關鍵之一。1760年工業革命開始，在這段時間裡，人類生產逐漸轉向新的製造過程，出現了以機器取代人力、獸力的趨勢，以大規模的工廠生產取代個體工場手工生產的一場生產與科技革命。由於機器的發明及運用成為了這個時代的標誌，因此歷史學家稱這個時代為機器時代（the Age of Machines）[1]。隨著科技的日新月異，現今人們追求工業技術進步的渴望並沒有改變。工廠內的生產線逐漸自動化，朝著機械完全取代人力的目標前進。因此，機械手臂在產線中的運動規劃以及安全指數尤為重要，本論文即針對機械手臂的碰撞偵測技術進行了改善，並且實作出模擬的產線。過去的碰撞偵測軟體或演算法多半利用Graphics Processing Unit(GPU)進行計算，物體的模擬是使用多邊形網格（Polygon mesh）進行繪製[2]，此方法將需要很高的成本，並不符合工業需求。本論文改善此一缺點，利用Oct-Tree Algorithms分解物體造型[3]，再使用Bounding Box技術將其形成包覆盒[4]，藉此簡化計算量，在可接受的誤差範圍內，大幅降低計算時間。最後我們也在Rhino這個軟體上實作演算法，建構了一套工業上非常便利的產線模擬系統。
Since ancient times, advances in industrial technology have been the goal pursued by mankind and one of the keys to improving the quality of human life. The Industrial Revolution began in 1760. During this time, human production gradually shifted to a new manufacturing process. There was a trend of replacing manpower and animal power with machines, and a production and technological revolution that replaced large-scale factory production with manual production in individual workshops. Since the invention and application of the machine became a symbol of this era, historians call this age the Age of Machines [1]. With the rapid development of science and technology, the desire of people to pursue industrial technological progress has not changed. The production line in the factory is gradually automated, moving towards the goal of fully replacing manpower by machinery. Therefore, the motion planning and safety index of the robot arm in the production line are particularly important. This paper has improved the collision detection technology for the robot arm and actually simulated the production line. In the past, collision detection software or algorithms were mostly calculated using the Graphics Processing Unit (GPU). Object simulation was performed using Polygon mesh [2]. This method will require high costs and does not meet the requirements. Industrial demand. This dissertation improves this shortcoming by using Oct-Tree Algorithms to decompose the object shape [3] and then use the Bounding Box technique to form a cover box [4]. This simplifies the calculation and reduces the error within an acceptable error range. calculating time. Finally, we also implemented the algorithm on the Rhino software to construct an industrially very convenient line simulation system
基於動態運動偵測的帶位機器人;Ushering Robot Based on Dynamic Motion Detection
No full text[[abstract]]近年來台灣服務業佔國內生產總值(GDP)比重高達63.15%，為總就業人數比重中的59.17%。由此可知，服務業已為我國經濟活動之主體。服務業顧名思義，以提供勞務換取報酬，其工作內容大多為高重複性、高勞動力且長時間性的工作。但長時間且高勞動力的工作往往容易因為注意力下降而產生人因事故，甚至導致顧客受傷。隨著科技的進步，硬體愈做愈小且效能卻愈來愈好。因此我們可以運用高性能的硬體來製造與真人體型相當的機器人，用於取代人類以進行高重複性、高勞動力且長時間性的工作，不僅能提升工作效率及準確度，還可以大幅降低人因事故發生的機率及傷害。一般而言當顧客欲於店內用餐時，店家與顧客第一個遇到並需要解決的問題即為桌位的選擇。以店家的角度，希望將桌子完全填滿以達到桌子有效利用以及利益最大化；以顧客的角度。有人希望享受自己獨處的時光，有人希望跟別人併桌結交新朋友。本論文以餐廳用服務型機器人的帶位為研究主體，針對如何判斷桌位是否空閒與如何選擇合適桌位以及如何規劃路徑三大重點做討論，並由此實作出帶位機器人。本論文藉由結合選取合適位子演算法、感興趣區間偵測以及路徑規劃演算法實作帶位機器人系統。其中選取合適位子演算法能夠即時地依照用餐人數判斷對店家以及顧客適合的桌子；感興趣區間偵測可即時監測某一區域並為系統提供當前該區域的狀況；路徑規劃則依當前規劃路線和當前周遭障礙物決定即時路徑規劃是否需重新計算。
In recent years, Taiwan's service industry accounted for 63.15% of the gross domestic product (GDP), accounting for 59.17% of the total employment. It can be seen that the service industry has become the main body of China's economic activities. As the name suggests, the service industry provides labor services in exchange for remuneration. Most of its work is highly repetitive, high-workforce and long-term work due to accidents. However, long-term and high-labor work tends to generate people because of decreased attention, and even lead to customer injury.With the advancement of technology, the hardware is getting smaller and better and the performance is getting better and better. Therefore, we can use high-performance hardware to make robots that are comparable to real human bodies. It can replace humans for high reproducibility, high labor and long-term work, which not only improves work efficiency and accuracy, but also greatly reduces The probability and harm of human accidents.Generally speaking, when a customer wants to dine in the store, the first problem that the store and the customer encounters and needs to solve is the choice of the table. From the store's point of view, I hope to completely fill the table to achieve effective use of the table and maximize the benefits; from the customer's point of view, someone wants to enjoy their own time alone, and some people hope to make new friends with others. This thesis is a service-oriented robot for restaurants. The position of the research is the main body of the research. It discusses the three key points of how to judge whether the table is free and how to choose the appropriate table and how to plan the path, and thus makes the robot with the position.In this paper, by combining the appropriate seat algorithm, the interest interval detection and the path planning algorithm are implemented as the robot system. The appropriate seat algorithm can be used to judge the table suitable for the store and the customer according to the number of people. Interval detection can instantly monitor an area and provide the current status of the area for the system; path planning determines whether the immediate path planning needs to be recalculated based on the current planned route and the current surrounding obstacles
支援瞬時錯誤檢測與糾錯之行程層級重複技術;A Process-level Redundancy Technique for Transient Fault Detection and Correction
No full text[[abstract]]現代電子產品已經融入了我們的日常生活，不過隨著這些產品的複雜和精緻程度日漸提高，卻也造成它們發生瞬時錯誤的機會隨之提升。為了要保證這些產品日復一日的正常運作，確立軟體可靠度相關的技術是必須的。 軟體可靠度對於現代電腦運算系統的運作是十分重要的。這篇論文提出了一套基於軟體用於提升錯誤容忍力的程式碼編寫模板，用來減少系統受到瞬時錯誤的影響。這套程式碼編寫模板是基於UNIX POSIX的系統調用函示完成，提供了可移植性。此外，我們藉由LLVM編譯器開發了一項程式碼轉換工具，用來幫助程式撰寫者自動地將程式碼套用我們提出的程式碼模板。實驗的受測程式經由我們的程式碼轉換工具轉變為符合我們的程式碼模板的程式，實驗的結果顯示這些經過轉換的程式具備了監控運作狀態、偵測、修復瞬時錯誤的能力。
Modern electronic devices have entered our daily life, but their increasing sophistication increases the likelihood of transient faults occurring. To guarantee their consistent normal operation day after day, techniques to ensure software reliability are necessary. Software reliability is critical to the operation of modern computer systems. This thesis proposes a software-based, process-level programming model for tolerating transient faults, which are errors that temporarily occur in systems. The programming model is portable as its design is based on UNIX POSIX system calls. Besides, we develop a code translator in LLVM compiler for programmer to generate a fault-tolerant code automatically. The tested benchmark programs are modified by the translator to conform to the proposed programming model. Preliminary experiments using the tested benchmark programs show that the proposed approach can monitor working status, detect errors, and recover from transient faults
應用多向神經網路於多圖片多標籤分類;Multi-Stream Networks for Multi-Sampled Multi-Label Image Classification
No full text[[abstract]]在電腦視覺領域，圖片分類問題一直都是個熱門的議題。在本論文中我們提出一個深度神經網路，以實現電影海報和服裝的圖片分類。在此深度神經網路中，我們同時考慮視覺外觀及物件資訊以及考慮相同實體常以多張圖片呈現，將電影海報及服裝圖片依類別分類。由於一張圖可能屬於多個類別，我們將這個分類問題定義成一個多標籤分類問題。為了進行這項研究，我們蒐集了一個大型的電影海報資料集和一個服裝的資料集。基於這兩個資料集，我們訓練了一個卷積神經網路取得圖片的視覺資訊，並且使用現時最好的物件偵測方法以取得圖片的物件資訊。另外，我們整合相同實體的多張圖片呈現，提出多向深度神經網路進行整合。在最後的實驗中，我們達到比以往工作更好的分類效果。我們證明同時考慮相同實體中的多張圖片呈現，可以達到比只使用單張圖片更好的效果。
In the computer vision research field, image classification has always been a hot topic. In order to achieve multi-label image classification for movie posters and clothing images, we propose a deep neural network in this thesis. In this network, we jointly consider visual appearance, object information, and multiple images of the same entity, and classify movie posters and clothing images. Since an image may belong to multiple categories, we define this classification problem as a multi-label classification problem. We collect a large-scale movie poster dataset and an in-shop clothes dataset, associated with various metadata. Based on these two datasets, we fine-tune a pretrained convolutional neural network to extract visual representation, and adopt a state-of-the-art framework to detect objects in images. In addition, we integrate multiple images of the same entity and propose a multi-stream deep neural network. In the evaluation, we show that the proposed method yields encouraging performance, which is much better than previous works. We also prove that jointly considering multiple samples of the same entity yields performance better than considering only one sample
基於判決文之量刑系統設計與實作;The Design and Implementation of a Sentencing Prediction
No full text[[abstract]]在司法界公平與正義一直以來都是一個重點，法官在審判過程中進行查詢該案件適用法條及過去的判例來決定量刑，本論文主要目的為協助查詢法條及量刑預測能夠對待被告公平的量刑。本論文以植根法律網上的判決文作為量刑系統的訓練資料，從判決文資料中，使用Pointwise Mutual Information進行產生詞頻庫，作為jieba產生的字詞之合併，使用TF-IDF、word2vec及人工的方式抽取法條的關鍵字。接著統計各法條的量刑並透過線性回歸的方式產生自動量刑的模組，當使用者輸入案件後，進行法條關鍵字比對，能得到該案件的適用法條，透過量刑預測模組，找出該案件的量刑，並提供給使用者參考。
Fairness and justice are always a key point in the judicial circle. During the trial, judges determine the sentencing by using the law articles and past jurisprudences. The main purpose of this thesis is to treat the defendant with a fair sentence with the assist in searching the law, sentencing prediction and other projects. In this thesis, the judgment text from “rootlaw.com” is used as the training data of the sentencing system. The judgement text is segmented by jieba, and the library as the combination of the words which generated by jieba is created by using Pointwise Mutual Information and used to retrieve the key words of the law by TF-IDF, word2vec and manual methods. In addition, the sentencing prediction module is generated by counting the sentencing of each law and the linear regression. As the user enter the case, matching the keywords can get the used law in the case and find out the sentencing of this case to the user by the sentencing prediction module
運用機器學習改善乳房超音波彈性影像判讀準確性-以BI-RADS 3、4為例;Using Machine Learning to Improve the Accuracy of Breast Ultrasound Elastography Interpretation - Example of BI-RADS 3 & 4
No full text[[abstract]]超音波彈性影像（Ultrasound Elastography）是乳房疾病臨床診斷實驗時常見的非侵入性腫瘤篩檢儀器之一，它採用色彩來表示組織受到施壓力的變形程度。在臨床診斷時，醫師可從超音波影像確認腫瘤的位置與其形狀、大小，再從彈性影像去觀察腫瘤與其周圍組織的形變程度。藉著從兩者影像所觀察到的特徵，可進一步判斷出腫瘤的良惡性與其等級。依據美國放射學會發布的乳房影像報告與資料系統，可以知道BI-RADS 3、4 級之間為腫瘤良惡性的分界線。本研究從超音波彈性影像運用一系列影像處理技術偵測出腫瘤與其周圍組織為感興趣區域，並依據感興趣區域的組織特性，量測出包括: 型態、紋理以及色彩等特徵；最後，再運用器兩種機器學習模型：支援向量機（Support Vector Machine, SVM）與卷積神經網路（Convolutional Neuron Network, CNN）配合不同特徵值判讀BI-RADS 分級。這兩者分類器所需要的輸入資料類型不同，其中SVM 是以圖像所描述的特徵值作為分類器的輸入；但CNN 則是直接以影像亮度當成輸入資料。針對實驗樣本進行實驗結果顯示，在結合彈性影像與傳統超音波影像的情況下，SVM 判讀效能分別為：準確率95.79%、靈敏度94.59%、特異度96.10%；CNN 則分別為：準確率97.19%、靈敏度97.30%、特異度97.16%。除了比較乳房彈性超音波在兩種分類器類型的判讀結果外，本研究也進一步針對傳統超音波影像、彈性影像和兩者結合的分類結果比較。以不同分類器的實驗效能比較而言，CNN 會比SVM 有更好的分類效能與其適應性。尤其此方法不需要任何傳統統計計算，因其可直接導入影像像素亮度做為分類器的導入輸入資料。此研究結果對於在建立電腦輔助診斷（Computer Aided Diagnosis, CAD）系統的研發人員來說，此方式可以減少工程師在程式開發的繁雜度，且此CAD 系統未來在臨床診斷中能協助醫師進行判讀，讓患者可以避免不必要的手術與治療。
Breast ultrasound elastography is one of non-invasive imaging modalities for diagnosis of breast tumor. It can describe the deformation degree of tissue by strain with colors. In clinical diagnosis, physicians validate the tumor location, evaluate its shape and size from ultrasound, and measure the strain degree of the tumor and its surrounding tissues from ultrasound elastography. The severity of the tumor cant be further assessed as malignant or benign by using these two modalities.According to the Breast-Imaging Report and Analyzing Data System (BI-RADS) by American Community Radiology (ACR), the categories such as the BI-RADS 3 and the BI-RADS 4 are borderline for a tumor identified as malignant or benign. In this study, a series of image processing methods are applied to detect the suspicious tumor as a region of interest (ROI). Then, the features including morphology feature, texture feature, and color feature are measured quantitatively and extracted from the ROI in terms of the tissue characteristics. Finally, the machine learning methods such as Support Vector Machine (SVM) and Convolution Neural Network using different features are performed to identify the actual category of BI-RADS. Quantitative features such as shape feature and morphology feature of the ROI was adopted by SVM, but the intensity values of each pixel of the ROI is chosen as features by CNN. Experimental tests on the collected sample when combining with B-Mode ultrasound and ultrasound elastography on the ROI, the accuracy, sensitivity, and specificity classified by SVM are 95.79%, 94.59% and 96.10%, respectively, but the accuracy, sensitivity, and specificity classified by CNN are 97.19%, 97.30%, and 97.16%, respectively.In comparison with experimental performance with different classifiers, the CNN outperform SVM in classification performance and adaption. Above all, the CNN do not require any traditional statistics, because it can directly induce the intensity values of image pixels as its input. Therefore, the proposed CAD system can reduce the complexity in developing the system for engineers. Moreover, it can assist physicians identify the breast disease in clinical diagnosis so that patients can avoid unnecessary surgeries and treatments in the feature
基於深度神經網路重建數據於土石流感測;Deep Neural Network-based Data Reconstruction for Landslide Detection
No full text[[abstract]]土石流會對生命產生巨大的威脅以及造成財產上的損失，在土石流預警系統中，可以透過感測器蒐集環境資訊來檢測土石流發生的可能性，然而無線傳感系統所收集的資料可能會因為外力干擾或者其他環境因素導致感測器故障或者資料的遺失，而影響到土石流預測的精準度。為了解決資料遺失的問題，本篇論文以降雨強度、土壤濕度、坡度、坡向為例提出資料重建方法，基於異質資料以及時間與空間上關係重建遺失資料。在空間方面的考量除了感測節點間的距離之外，還考慮區域的地形(例如坡度與坡向)以及降雨的區域。在時間上的考量則是以過去一段時間的資料趨勢作為重建考量，並且結合異質資料間相關性(如雨量與土壤濕度)，卷積長短期記憶(ConvLSTM)深度神經網路被訓練來預測丟失數據，我們利用所預測之數據遺失時間點之數據補償遺失資料。實驗結果呈現與其他重建方法相比，本文所提出的方法的重建結果有更好的表現，在各種資料遺失情況即使有90\%資料遺失，重建後的資料與原本資料所計算之RMSE 仍能小於 0.3。
Landslides could cause huge threats to lives and cause property damages. In the landslide prediction system, environmental information can be collected through sensors to detect the possibility of landslide occurrences. However, the data collected by wireless sensor network systems (WSNs) may be lost due to sensor failures, external interferences or other environmental factors, which may affect the accuracy of landslide predictions. In order to solve the problem of missing data, this Thesis proposes a data reconstruction method based on rainfall intensity, soil moisture, slope, and slope direction, and reconstructs missing data based on heterogeneous data and temporal and spatial relationships. In terms of spatial, in addition to considering the distance between the sensing nodes, the terrain of the area (such as slope and slope direction) and the area of rainfall are also considered. In terms of temporal, it is based on the data trend of the past period of time as a data reconstruction consideration, and combined with the correlation between heterogeneous data (such as rainfall and soil moisture). A Convolutional Long Short-Term Memory(ConvLSTM) deep neural network is trained to predict the missing time slot data. We use the predicted data to compensate missing data. As a result, compare with other reconstruct methods our proposed method can get better performance. In the case of various types of data missing, even if 90\% of data is lose, the RMSE calculated by the reconstructed data and the original data can still be less than 0.3