本文为美国新罕布什尔大学（作者：Andrea Linda Murphy）的硕士论文，共58页。

MATLAB是由MathWorks开发的多范式专用编程语言和数值计算环境。在MATLAB集成开发环境（IDE）中，您可以执行计算机辅助设计（CAD）、不同的矩阵操作、函数和数据的绘制、实现算法、用户界面的创建，并能够与其他语言编写的程序进行交互。自1984年推出以来，MATLAB软件在数据科学领域并没有特别的关联。2013年，他们推出了新的数据科学集中工具箱，包括深度学习、图像处理、计算机视觉，一年后又推出了统计与机器学习，改善了这种状况。

本文的主要目的是研究和探索数据科学领域。更具体地说，是关于一个对象识别应用程序的开发，该应用程序可以完全使用MATLAB IDE构建，并对聋人社区产生积极的社会影响。在这样做的同时，回答了以下这个问题：MATLAB是否可以用于开发这种类型的应用程序？为了回答该问题，同时解决我的主要研究目标，利用MATLAB 2019和附加数据科学工具包构建了两个不同的对象识别协议。我将协议命名为ASLtranslate（I）和（II）。这使我能够在学习使用多种方法解决同一问题差异、优点和缺点的同时，对所有MATLAB数据科学工具箱进行实验。两个版本的设计方法非常相似。

ASLtranslate将美国手语（ASL）手势的二维图像作为输入，对图像进行分类，然后输出相应的字母字符。ASLtranslate（I）是利用机器学习实现图像分类的一种方法。ASLtranslate（II）是通过使用一种称为迁移学习的深度学习方法来实现的，该方法通过微调预先训练的卷积神经网络（CNN）AlexNet来对新的图像集合进行分类。

MATLAB is a multi-paradigm proprietary programming language and numerical computing environment developed by MathWorks. Within MATLAB Integrated Development Environment (IDE) you can perform Computer-aided design (CAD), different matrix manipulations, plotting of functions and data, implementation algorithms, creation of user interfaces, and has the ability to interface with programs written in other languages1 . Since, its launch in 1984 MATLAB software has not particularly been associated within the field of data science. In 2013, that changed with the launch of their new data science concentrated toolboxes that included Deep Learning, Image Processing, Computer Vision, and then a year later Statistics and Machine Learning. The main objective of my thesis was to research and explore the field of data science. More specifically pertaining to the development of an object recognition application that could be built entirely using MATLAB IDE and have a positive social impact on the deaf community. And in doing so, answering the question, could MATLAB be utilized for development of this type of application? To simultaneously answer this question while addressing my main objectives, I constructed two different object recognition protocols utilizing MATLAB_R2019 with the add-on data science tool packages. I named the protocols ASLtranslate (I) and (II). This allowed me to experiment with all of MATLAB data science toolboxes while learning the differences, benefits, and disadvantages of using multiple approaches to the same problem. The methods and approaches for the design of both versions was very similar. ASLtranslate takes in 2D image of American Sign Language (ASL) hand gestures as an input, classifies the image and then outputs its corresponding alphabet character. ASLtranslate (I) was an implementation of image category classification using machine learning methods. ASLtranslate (II) was implemented by using a deep learning method called transfer learning, done by fine-tuning a pre-trained convolutional neural network (CNN), AlexNet, to perform classification on a new collection of images.

1.       引言

2. 研究目标

3. 目前的系统

4. 研究历史

5. ASLtranslate设计

6. 研究结果

7. 评估与讨论

8. ASLtranslate的未来发展

附录A ASLtranslate源代码

附录B ASLtranslate II源代码

更多精彩文章请关注公众号：