中原大學電機工程學系
Loading
專題成果首頁>專題成果
以人工智慧深度學習進行手寫數字辨識
指導老師:洪穎怡   組長:楊明浩   組員:劉山廷、莊承翰
阿拉伯數字是我們在日常生活中頻繁使用到的,透過數字可以組成我們的手機號碼、帳號密碼、表達數量等等。但鑒於每個人書寫阿拉伯數字的習慣不同,導致筆跡也會不同,除了撰寫者本身可能可以輕鬆一眼辨識數字為何,但是讀者卻可能難以辨識出正確的數字,造成傳達重要資訊的過程可能無法完全。本研究目的為透過神經網路的深度學習來輕易辨別各式各樣的數字筆跡,並且透過比較6種不同的神經網路來得知準確率最高的為何。
深度學習中有很多方法都可以完成本實驗,但這項實驗的主要目的就是用6種方法來判定出哪一種方法最精準且最節省時間,是本次實驗的主要目的。此次實驗有6種不同的方法:
(1) 多層前饋式神經網路(Multilayer Feedforward Neural Network, MFNN)
(2) 多層前饋式神經網路深度學習(Multilayer Feedforward Neural Network Deep Learning, MFNN-DL)
(3) 支持向量機(Support Vector Machine, SVM)
(4) 深度信念神經網路(Deep Belief Network, DBN)
(5) 自動編碼機(Autoencoder)
(6) 卷積神經網路(Convolutional Neural Network, CNN)
Arabic numerical digits are frequently used in daily life. We can form our mobile phone numbers, account passwords, and express numbers, etc. However, given the different habits of each person writing Arabic numerals, the handwriting will be different, except that the author himself may be able to easily identify the number at a glance, but the readers may have difficulty recognizing the correct number, and the process of conveying important information may not be complete. The purpose of this study is to easily identify a wide variety of digital handwritings through deep learning of neural networks, and to compare the six different neural networks to determine the highest accuracy.
There are a lot of deep learning neural networks that can be used to conduct the experiment, but the main purpose of this experiment is to use these six methods below to determine which method is the most accurate and most time-saving, which is the main purpose of this experiment. There are six different methods for this experiment:
(1) Multilayer feedforward neural network (MFNN)
(2) Multilayer feedforward neural network deep learning (MFNN-DL)
(3) Support Vector Machine (SVM)
(4) Deep Belief Network (DBN)
(5) Automatic encoder (AE)
(6) Convolutional Neural Network (CNN)
我們決定將壓電陶瓷晶片和鞋子做結合。
   
 
32023 桃園市中壢區中北路200號 No-.200, Zhongbei Rd., Zhongli City, Taoyuan County 320, Taiwan(ROC) TEL:03-2654801 FAX:03-2654899