UTAR Institutional Repository

Design and Development of a Practical Macroscopic Wood Identification System Using Deep Learning

Tang, Xin Jie (2019) Design and Development of a Practical Macroscopic Wood Identification System Using Deep Learning. Master dissertation/thesis, UTAR.

[img]
Preview
PDF
Download (8Mb) | Preview

    Abstract

    Wood serves as raw material for countless industries due to its unique material characteristics. As such, different wood types are graded and valued accordingly based on their commercial value as raw material. Hence, wood identification is needed to ensure the correct wood type for usage. Macroscopic level wood identification that has been practiced by wood anatomists for decades can identify wood up to genus level for any commercial timber group. However, this knowledge is difficult to transfer to the industry non-experts. In this research, a rapid and robust macroscopic wood identification system is proposed using deep learning method with off-the-shelf smart-phone and retrofitted macro-lens as image acquisition device. Trained deep learning model is deployed as a cloud service accessible via Internet. This research collects and verifies data by wood anatomists on 100 Malaysian Tropical Timber types using the image acquisition device. A new Convolution Neural Network BlazeNet designed by the author, achieved better accuracy when benchmarked against SqueezeNet in this research. A cloud based wood identification system was deployed accompanied by an iOS application, Mywood-ID.

    Item Type: Final Year Project / Dissertation / Thesis (Master dissertation/thesis)
    Subjects: T Technology > TA Engineering (General). Civil engineering (General)
    Divisions: Institute of Postgraduate Studies & Research > Lee Kong Chian Faculty of Engineering and Science (LKCFES) - Sg. Long Campus > Master of Engineering Science
    Depositing User: Sg Long Library
    Date Deposited: 17 Dec 2019 17:13
    Last Modified: 17 Dec 2019 17:13
    URI: http://eprints.utar.edu.my/id/eprint/3630

    Actions (login required)

    View Item