可選字及可再訓?眼寫系統;Selectable and retrainable eye writing system

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Electrical Engineering > Electronic Thesis & Dissertation > Item 987654321/86647

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/86647

Title:	可選字及可再訓?眼寫系統;Selectable and retrainable eye writing system
Authors:	林鈺庭;Lin, Yu-Ting
Contributors:	電機工程學系
Keywords:	眼電圖;眼寫;深度學習;卷積網路;數據擴增;EOG;eye-writing;deep learning;CNN;data augmentation
Date:	2021-08-05
Issue Date:	2021-12-07 13:04:35 (UTC+8)
Publisher:	國立中央大學
Abstract:	患有肌萎縮側索硬化症 (Amyotrophic lateral sclerosis, ALS)的病人，由於運動神經元的退化，肌肉會逐漸萎縮，失去運動能力，無法利用口說和肢體對外溝通，較晚開始退化的眼球運動功能成為後期病人希望可以依靠來對外溝通的一個途徑。本研究以阿拉伯數字(0~9)、大寫英文字母(A~Z)以及 4 個特殊符號(空格、輸入、句點、問號)，共 40 個符號，建立一套眼寫系統。使用眼電圖法（Electrooculography, EOG）記錄眼球運動，將 EOG 訊號轉為影像利用卷積神經網路(Convolutional Neural Networks，CNN）進行符號辨識，以眨眼數次當作特殊指令來控制進出書寫辨識次系統、書寫的開始與結束、刪字或選字，以模擬一般的手寫和打字情境。模型訓練所需要的資料包含全部符號各被眼寫多次所收集的原始資料以及擴增資料，以 k 折交叉驗證（k-fold cross-validation）的方式訓練卷積網路。使用者實際使用該系統時，可以辨識或選擇成功的符號軌跡可用來再訓練，使模型更加穩健，越使用越符合使用者的書寫習慣，提高眼寫辨識的準確率(Accuracy)。;In patients with amyotrophic lateral sclerosis (ALS), due to the degeneration of motor neurons, the muscles will gradually atrophy, lose exercise ability, and cannot use oral and physical communication. Eye moving function that degenerates later may become the only way for the later-stage patients to rely on to communicate with others. This study uses Arabic numerals (0~9), uppercase English letters (A~Z) and 4 special symbols (space, input, dot, question mark), a total of 40 symbols, to establish the eye-writing system. This system uses electrooculography (EOG) to record eye movements. It converts the EOG signals into images for symbol recognition with a convolutional neural network (CNN). It uses the number of blinks as special commands to control the entering and exiting of the writing and recognition subsystem, the start and end of writing, character selection and deletion. The system functions to emulate general handwriting and typing situations. The data required for model training include the original data collected by eye-writing all symbols for multiple times and the augmented data. The CNN is trained by k-fold cross-validation. When the system is practically used by the user, any symbol trace that leads to successful character recognition or selection can be used to retrain the neural network model to make the model gradually become more robust. The more the system is used and retrained, the more it will fit the user′s writing habits, and the accuracy of eye-writing recognition will be gradually improved.
Appears in Collections:	[Graduate Institute of Electrical Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	100	View/Open

社群 sharing

Loading...