Chung Cheng University Institutional Repository
Not a member yet
889 research outputs found
Sort by
智慧型駕駛狀態監測系統;Intelligent Driving Condition Monitoring System
No full text[[abstract]]本研究針對駕駛人狀態特徵建置一套主動式安全輔助系統－「智慧型駕駛狀態監測系統」，擷取兩項生理資訊建置專屬的生理模組，分別是脈搏頻率和眨眼頻率。取得這兩項生理資訊建立二維向量模組，去除離群值以後，就可以在行車的時候使用偵測模式做駕駛狀態監測。 眨眼頻率是使用Webcam擷取影像，使用P.Viola提出的臉部偵測方法[20]定位出臉部，然後用霍夫轉換的原理找尋虹膜位置，便可以偵測出眨眼的動作，從而算出眨眼頻率。脈搏頻率使用Arduino Uno[1]和搭配的光反射感測器，感測器打出光線到手指上，光反射回感測器上的接收器，血管中的血液會吸收光，影響反射光的強度，偵測出接收器接收的光線強弱變化的規律，就可以判斷脈搏的發生，依照此原理偵測可以抓到脈搏的間隔時間，從而計算出脈搏頻率。 使用上述原理取得生理資訊以後建立二維模組，使用密度方法（Outlier detect based on Density），濾除離群值以後便完成個人專屬的偵測模組。在行車的時候載入針對該駕駛狀態訓練完成的專屬模組，同時做即時的生理資訊偵測，就可以即時偵測目前駕駛人的生理狀態是否正常。 本研究可以應用在計程車總部監控部門，或是客運公司的監測單位，可以即時監測旗下的駕駛人狀態，評估駕駛狀態做及時調度，避免憾事發生，提高行車品質。關鍵詞：眨眼頻率，脈搏頻率，二維模組，離群值，監測，密度
In this study, by investigating the pulse frequency and the blink frequency, two physiological information models were used to establish an active safety aided system - the "intelligent driving condition monitoring system", and the two physiological information were established as two vectors Module, after the removal of the outliers, you can use the detection mode in the driving time to do driving state monitoring . Blink frequency is to use Webcam to capture images, through P.Viola proposed face detection method[20] to locate the face, and then use the Hough conversion principle to find the pupil position, you can detect the blink of an action to calculate the blink frequency. Pulse frequency detection is the use of Arduino Uno[1] and with the light reflection sensor, the sensor sends light to the finger, the light reflected back to the receiver on the sensor, blood vessels will absorb light, resulting in reflected light Strength changes, in accordance with this principle can be detected to grasp the pulse of the interval time to calculate the pulse frequency. After using the above principle to obtain physiological information, use it to establish a two-dimensional module, by using the density method (Outlier detect based on Density) filter out the outliers, you can complete the individual exclusive detection module. In the driving time to load a dedicated module to do real-time physiological information detection, you can immediately detect the current state of the driver's normal or not. This study can be applied to the monitoring department of the taxi headquarters, or the monitoring unit of the passenger company. It can monitor the driver's status in real time, evaluate the driving state and timely dispatch, avoid the occurrence of the trouble and improve the running quality.Keywords：two-dimensional module, Outlier, Blink frequency, Pulse frequency, Density, monitorin
基於社會臨場感理論之近距離感知資訊系統;A Proximity Aware Information System Based on the Social Presence Theory
No full text[[abstract]]在校園內，人們傳遞室內資訊的媒介，包括LED跑馬燈、語音廣播、數位看板等等。雖然這些媒介能利用輪播的方式，傳遞資訊給人們；但是無法讓人們和媒介之間產生如同人與人溝通的社會臨場感體驗。隨著科技日新月異的進步，人們手中的智慧型手機已成為他們獲取訊息最普遍的方式。除此之外，iBeacon無線感知技術的出現，也使得硬體製造商推出了能夠應用於室內環境的Beacon感測裝置。由於Beacon感測裝置的底層傳輸機制是低功耗藍芽無線技術，而這項無線技術也同樣支援於目前所有的智慧型手機。因此，開發專屬App將成為Beacon感測裝置與智慧型手機建立互動的方式。綜合上述所說，本研究藉由物聯網的架構，提出了近距離感知資訊系統。這個資訊系統結合了近距離感知事件管理的網站、數個Beacon感測裝置、近距離感知訊息的iOS App以及雲端資料庫。本研究的目的分成以下兩點:第一點，利用Beacon感測裝置，使得每一個室內地點都被附加上臨場感知的特性。第二點，讓每一位進入室內地點的學生，在使用我們提供的iOS App來接收各種近距離感知訊息的同時，也能有社會臨場感的使用者體驗。
On the campus, there are various communication media which can help people transfer indoor information, such as the LED marquee, the voice broadcast and the digital signage. Although these indoor information are spread from the carousel methods to students on communication media, the experience of a social presence being an interactive sense between people is not created from students and communication media. With science and technology change rapidly, the smartphone which we take to use on the hand has become the most common way to help us get information. In addition, some hardware manufacturers release Beacon devices applied to indoor environments after the iBeacon technology is published by Apple. Owing to that the underlying transport mechanism of the Beacon device is adopted by the wireless technology of BLE, the wireless technology of BLE is also supported on all smartphones including Android and iOS. Therefore, the specific app developed by us can create an interactive method between Beacon devices and the smartphone.As mentioned previously, this study is proposed the proximity aware information system by the architecture of the IoT. This information system integrates the website of proximity aware event management, Beacon devices, the iOS App of proximity aware message and a cloud database. The purpose of this study is subdivided into the following two sections. The first section is that each indoor locations are attached to the proximity aware feature by using Beacon devices. The second section is that students can receive all kinds of proximity aware messages and create the user experience of the social presence after they walk into each indoor locations
使用深度信念網路的標準國語語調分析;Intonation Analysis in Standard Mandarin using Deep Belief Networks
No full text[[abstract]]深度信念網路在機器學習領域是一個常用的方法，像是在圖像辨識、語音辨識等等。我們只需要對原始資料作些許的前處理，他就能夠從原始資料中學習特徵。在全世界的許多語言中，普通話是少數幾種聲調語言之一，不同聲調會讓一個語音有著不同的意思，也因此在語音辨識上普通話較其他語言辨識的困難度較高，例如英文。換句話說，聲調辨識的正確率對語音辨識率有著相當大的影響，在本論文中，使用了深度信念網路作為國語語調分析的工具，對連續語音進行聲調的分析，並與以往實驗的成果進行比較，實驗結果顯示使用深度信念網路作為工具能得到比以往的方法還要好的成果。
Deep belief network is a commonly used method in machine learning domain, such as image recognition, speech recognition, etc. With a little pre-processing, it can learn the features from the raw data excellently.In many languages around the world, Standard Mandarin is one of few tonal languages, different tones will make a speech differently. Therefore speech recognition in Standard Mandarin is highly difficult than other language for instance English. In other words, it means that the correctness of tonal recognition has a great influence on the correctness of speech recognition,In this thesis, Deep belief network is used as the tool for Intonation Analysis of Standard Mandarin, and the analysis of continuous speech is carried out. The results show that using Deep belief network as a tool can obtain more information than previous methods
利用具有群組標記與註釋功能之系統以提升電子書之閱讀性能;To Improve e-Book Reading Performance with a Group Underline and Annotation System
No full text[[abstract]] 網路上的教學平台提供學生一種新的學習環境，也提升老師與學生之間的學習關係。但是有一些線上教學平台未具有協同教學的功能，讓學生之間也能相互學習。 因此，本論文的系統，不僅可以讓老師上傳、編輯與修改教材、瞭解學生的學習情況，還可以讓學生瞭解同學之間的學習狀況，相互觀摩，提升學生的學習成效。
The platform of education supplies new environment of learning for students on the internet. This platform of education also enhances the relationship between teacher and student. However, some platforms of education online don’t have the function of cooperative instruction to enhance the relationship between student and student. Therefore, our system in this thesis can not only upload teaching materials, edit teaching materials, modify teaching materials and comprehend the learning condition of every student, but also offer students the function to understand the self-learning situation between student and student. This function may enhance the learning effects of students
聊天機器人系統設計與實作;The Design and Implementation of a Chatbot
No full text[[abstract]]聊天機器人是現在熱門的應用之一，而聊天是一個有趣的議題，聊天除了可以帶給人們娛樂性之外，也能提供問答的功能 本論文以Dcard論壇的文章做為聊天機器人的訓練資料，從Dcrad資料中產生了對話部分，並以機器學習的工具與統計從中取出重要特徵，進行詞性判斷、情緒分析、尋找重要詞彙，建出配對的模型，當使用者輸入句子時，便能將句子放入模型，找出相對應的回答，並回應給使用者。
Chatbot is one of the popular applications in recent years. Chatting itself is an interesting research topic. Besides being entertained, people are also looking for helpful answers or responses through chatting with the others. In this implementation, training data is crawled from Dcard, which is one of the most popular forums for undergraduate students in Taiwan. Chatting threads of Dcard are processed to form reasonable conversations. We use machine learning and statistical methods to extract effective features from conversations. With these features and predefined dictionary, part-of-speech tagging, emotion, and critical words can then be done efficiently. This study proposed a retrieval-based model trying to identify the chatting thread which shares the most common characteristics with input sentence, and then return the subsequent response of the chatting thread to user
旅遊搜尋系統之設計與實作;The Design and Implementation of Travel Search Engine
No full text[[abstract]]本論文實作一個旅遊搜尋系統，提供一個簡潔的網頁介面，使用者可透過此系統搜尋旅遊文章、景點資訊及收藏文章。除此之外，此系統會將文章分類並篩選出旅遊相關的文章，根據文章的分類以及使用者的瀏覽行為推薦使用者有興趣的文章。 另外，本論文也實作統計景點的熱門度，並以視覺化的圖表呈現熱門度趨勢，幫助使用者快速了解各景點的熱門度。透過此系統的統計及分析，可以幫助使用者在大量的資訊中，快速的篩選自己可能有興趣的文章，並且快速了解景點的熱門趨勢。
In this thesis, a travel search engine is designed with neat web user interface. The users can use our search engine to find wanted travel-related articles, scenic spot information, and collect the interesting articles. Our engine also provides article classification and recommendation function based on user's browsing behavior. Furthermore, our engine analyzed millions of travel related articles to calculate the statistics of scenic spots and show the users a trend of popularity of scenic spots through a visual graph. It can help the users to quickly sift through a vast amount of information to find interested articles and catch the trend of popularity for those interested scenic spots
基於測驗結果並用於認知診斷評量的監督式分群演算法;A Supervised Clustering Algorithm for Cognitive Diagnosis Assessment Base on Test Result
No full text[[abstract]]中文摘要所蒐集的教育現狀資料，是來自就讀於各大學學士生的考核評量測驗成績，各個評量測驗均包括已設定的一些概念。本研究使用各個成就測驗概念的結果，分類學生的能力。因此，我們可以將學生正確分群，依學生所具備的能力，組成具有不同優秀能力的類型。基於馬氏距離的GG和GK的模糊分群演算法，可以改善資料結構為球型的限制，但GG演算法只能用於資料分佈具有多變量正態分佈的數據。GK演算法所受到限制是，它必須知道數據的結構大小及其分佈。眾所周知，GG和GK模糊分群演算法，對於標準情況的正確率並不夠穩定。在本研究中，提出了一種基於FCM的改進的監督模糊分群演算法，我們採用新的閾值和新的收斂算則來改善GG和GK演算法的局限性，刪除了GK算法中協方差矩陣的決定因素的約束，同時應用目標函數中所存在的相關矩陣來代替協方差矩陣。實證資料的實驗結果顯示，當不同群組之間的數據分佈重疊，或各群組的形狀不是球形時，我們提出的新的演算法，可以在較短的時間獲得到較高的正確性，因此新的演算法整體之表現顯著優於其他演算法。關鍵字：模糊分類，GK演算法，GG演算法。
AbstractThe educational status was collected from an achievement test ofuniversity students, which includes some concepts. This study usedtheir results of each concept of achievement test to classify students'abilities. Hence we can group the students into the types they havegood abilities.Fuzzy algorithms of GG and GK based on Mahalanobis distancecan improve those limitations of spherical structural clusters, but GGalgorithm can only be used for the data with multivariate normaldistribution. GK-algorithm is limited by that it must to know thedistribution of data.The well-known ones, GG and GK were not stable enough forstandard situation. In this study, an improved Supervised ClusteringAlgorithm based on FCM has been proposed, we take a new thresholdvalue and a new convergent algorithm to improve those limitations ofGG and GK algorithms, deleted the constraint of the determinants ofcovariance matrices in the GK algorithm, and replace the covariancematrix with the correlation matrix which exists in the objectivefunction. The experimental results of real data sets show that ourproposed new algorithm get the better performance, when datadistribution between different clusters is overlapping or the shape ofclusters is not spherical like.Keywords: fuzzy clustering, GK-algorithm, GG algorithms
基於多視野卷積神經網路之息肉狀脈絡膜血管病變偵測;Detection of Polypoidal Choroidal Vasculopathy Using Multi-FOV Convolutional Neural Network
No full text[[abstract]]根據世界衛生組織統計老年性黃班部變病已是造成全球人類失明或視障的第三大主因。人口高齡化的問題是台灣近年來的現象，老年性黃班部變病也成為國人失明的主要原因之ㄧ。息肉狀脈絡膜血管病變(Polypoidal Choroidal Vasculopathy, 縮寫PCV)是老年性黃班部變病的一種亞型，息肉狀脈絡膜血管病變的特徵在形狀上大小不一且具有多數分支的血管網路，在末端上有高螢光性的肉狀囊腫結構(Polypoidal Lessions)，又稱為息肉(Polyps)。息肉狀脈絡膜血管病變的邊界通常有模糊的問題存在，使得醫生在標示病變區域時，相當花費精力與時間，因此我們提出一個電腦自動化輔助系統，在ICGA(Indocyanine green angiography，眼底循血綠血管攝影)影像上輔助醫生標示PCV變病區域，透過系統的輔助能減少醫生的負擔。我們的系統使用卷積神經網路進行 PCV病變偵測，最大優點是不需要人工設計特徵，透過大量的訓練資料讓電腦自動學習有代表性的特徵，我們希望此系統能協助醫生標示病變區域，透過電腦系統的輔助能讓執行效率提高。我們採用EVEREST的資料庫進行實驗，並且系統在PCV病變偵測表現上比現有方法有更高的準確度。
According to World Health Organization statistics, Age-related Macular Degeneration (AMD) is the third leading cause of blindness and visual impairment around the world. AMD has also became one of the major causes of blindess among elderly people in Taiwan. Polypoidal choroidal vasculopathy(PCV) is a subtype of neovasular(AMD), and is prevalent in (40\% -- 50\%) among Asian patients with exudative maculopathy. Due to the high incidence of PCV among the Asian people, the goal of my thesis is to develop a computer-aided diagnostic system for detecting PCV lesions in indocyanine green angiography (ICGA) images.Our proposed method is based on the convolutional neural network (CNN). In particular, we construct a multi-scale CNN model and use it to determine whether a pixel location belongs to PCV or not. If it is considered as a PCV lesion, our proposed method can further identify whether it is polyps or branching vascular network. We have conducted experiments using 55 ICG sequences from the EVERSET dataset. Compared with the previous methods, our proposed method achieves better segmentation accuracy on the EVEREST dataset
改善GSC演算法：WSN的繞送演算法;Imporved GSC algorithm: An Routing Algorithm for WSN
No full text[[abstract]]近年來網路技術的發展越來越迅速，無線網路的技術也越來越成熟，無線感測網路(Wireless Sensor Network，WSN)的討論度也越來越高，是熱門的討論議題之一，其中包括感測器的布置、封包繞送的路徑、通訊的協議及感測器電量的消耗等等的議題，讓研究家們絞盡腦汁讓感測器的使用更有效率，本篇注重的是如何讓WSN中的感測節點，在繞送封包的時候能夠減少能量的消耗，不浪費多餘的能量在傳送封包上。本論文討論的是Gauss-Seidel correction(GSC)繞送演算法，在無線感測網路的場景中，將場景分割成等間距的網格點，感測節點透過所屬網格點(grid point)的位能值(potential)，來決定該感測節點繞送封包的方向。在繞送封包的同時，也廣播自己的位能值，相鄰網格點接收到位能值，會更新自己的位能值，改善了Distributed Gauss-Seidel Iteration(DGSI)其缺乏同步機制的缺點。由於GSC繞送演算法在繞送封包時，會廣播自己的位能值，為了讓屬於相鄰網格點的感測節點都能接收到位能值，感測節點的溝通範圍必須包含所有屬於相鄰網格點的節點。當我們使用Gauss-Seidel 方法疊代位能值達到收斂時，網格點的位能值不會改變，相鄰網格點可以不用再監聽該網格點的位能值，感測節點的通訊範圍若保持原樣為包含屬於相鄰網格點的感測節點，可能導致額外的能量消耗。本篇論文在GSC疊代位能值達收斂之後，改善傳送時能量消耗的模型，讓溝通範圍縮小為至下一個繞送節點，觀察節點在傳送封包時所花費的能量。結果顯示在收斂之後，藉由縮小通訊範圍，減少傳送時多餘的能量消耗，讓感測節點不會太快能量耗盡，延長網絡的生存壽命。
Advanced development of network technology and wireless network enabled the wireless sensor network become a famous issue. Researchers discuss the issues about sensors development, energy consumption and routing algorithm. Energy consumption is an important factor which affects network lifetime. In this paper, we improve the energy consumption model of Gauss-Seidel correction (GSC) routing algorithm which formulates the routing paths to partial differential equations from a macroscopic model. We adjust the communication range of sensor nodes which forwarding the packet in some circumstances to decrease the energy consumption. Simulation results show that our improved energy consumption model in GSC effectively achieve energy saving and prolong the network lifetime
一種超像素多重對焦影像融合方法;Superpixel-based Multi-focus Image Fusion
No full text[[abstract]]由於光學鏡頭景深的限制，一般數位相機所拍攝出的影像並不能讓所有物體都以清晰的方式成像，造成單張影像只能夠呈現少數資訊。只有位於對焦平面上的物體會清晰的顯示在影像中；反之，在對焦平面外的物體則會以模糊的形態顯示在影像上。為了保留自然場景中清晰的物件及詳細的資訊，可以將多張具有相同場景、不同對焦物件的影像融合為一張全對焦的影像。在本論文中，提出一個以超像素為基礎的多重對焦影像融合方法。首先，將所有多重對焦影像個別作超像素切割，接著計算影像的顯著、深度、及差值影像資訊，並利用這些資訊將超像素分成三類，分別是對焦、離焦、以及不確定是否對焦的區域，對於不確定是否對焦的超像素則使用sum-modified-Laplacian (SML)來做判斷以產生focus maps，接著修正focus maps中對焦及離焦區域的交界處的權重並產生weight maps，最後依據weight maps產生融合影像。依據實驗結果所示，本論文所提出的研究方法優於現有的五種方法。
Due to the finite depth-of-field of optical lenses, it is difficult to capture an image with all clear objects by a common camera that make a signal image only display few information. The finite depth-of-field of optical lenses is the most common limitation which makes a signal image that not all objects are sharp and clear. Only objects within the depth-of-field are captured in focus and sharp, while others are defocus and blurred. To preserve clear objects and detailed information in a natural scene, we can take and integrate a set of images with different focuses to generate the fused image. In this study, a superpixel-based multi-focus image fusion approach is proposed. First, each multi-focus source image is segmented into superpixels. The saliency information, the depth information, and the difference image information are computed to classify the superpixels into three types (focus, defocus, and undefined) of regions. For the undefined superpixels, sum-modified-Laplacian (SML) is computed to determine whether the superpixel is focus or not, and then focus maps are estimated. Subsequently, each boundary of focus and defocus regions in focus maps is refined to generate the weighting maps. According to the weighting maps, the fused image is generated. Based on the experimental results, the performance of the fused images in this study is better than those of five comparison approaches