Exploring artificial intelligence(AI) for machine automation

• Main Author: Meow, Mervyn Zi Yang
• Format: Final Year Project / Dissertation / Thesis
• Published: 2019
• Subjects: T Technology (General); TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://eprints.utar.edu.my/3904/1/fyp_EE_2019_MMZY.pdf; http://eprints.utar.edu.my/3904/

Artificial intelligence (AI) is something intelligent and it could perform things that only human can perform. It might even be more powerful than the human minds if it was well developed. People all around the world are getting more familiar with AI as the technology development are getting more advance. Object detection task is one of the most popular example of artificial intelligence system that used to identify and classify objects. Inside the object detection task, it consists of deep convolutional neural networks as a classifier. This classifier is work together with other object detection technique to detect the region of interest of a particular image. There are many different type of open source frameworks such as Tensorflow, pytorch, Caffe and Keras are available on the internet. Many research had been done using Tensorflow by those huge company such as Nvidia, Uber and Snapchat in defecting object or face. Tensorflow is consider as low-level language which is more flexible in design. It is important to have more flexibility in desiging own functionalities as it allows us to change the architecture of networks based on our requirements. Researcher can understand how the operations are implemented through the network control provided by Tensorflow. It also allows the researcher to keep track of the updated change over certain time period. In this project, we use the Tensorflow Object Detection API which is an open source framework for object detection related task to identify and classify different types of components. Different type of deep learning models is used to make comparison in term of accuracy. In this case, we used Faster R-CNN as our object detection model and Inception-V2 as our feature extraction network. Faster R-CNN to run through the Region Proposal Network in order to obtain the region of interest and then input into the classifier network to obtain the classes for the particular object.