Title page for etd-0326118-050537


[Back to Results | New Search]

URN etd-0326118-050537
Author Wei-hao Fu
Author's Email Address No Public.
Statistics This thesis had been viewed 5333 times. Download 1 times.
Department Computer Science and Engineering
Year 2017
Semester 2
Degree Master
Type of Document
Language English
Title Enhancement of Flower Classification with the Profile Feature
Date of Defense 2018-04-19
Page Count 50
Keyword
  • inception-v3 model
  • Adaboost
  • flower classification
  • SVM
  • texture feature
  • color feature
  • segmentation
  • feature extraction
  • Abstract In this thesis, we propose an elegant method, based on machine learning, for the flower classification. There are three steps in our method. The process begins with segmenting the flower images and removing their backgrounds. We utilize the GrabCut approach to the do segmentation, because it provides a good performance; then, we extract the features from the foreground, including color features and texture features; finally, we train the SVM (support vector machine) models and Adaboost models with several feature combinations. The experimental material comes from the Oxford-102 category flower dataset. Our proposed feature, the profile feature, improves about 2% accuracy in the SVM model and about 3% in our ensemble model. However, if we combine the inception-v3 model into our model, the profile feature only improves about 1% accuracy. Our best result is 83.57% in accuracy which is obtained by aggregating several classification models and it outperforms all methods without deep-learning.
    Advisory Committee
  • Chiou-yi Hor - chair
  • Hsing-yen An - co-chair
  • Yung-hsing Peng - co-chair
  • Chang-biau Yang - advisor
  • Files
  • etd-0326118-050537.pdf
  • Indicate in-campus at 0 year and off-campus access at 1 year.
    Date of Submission 2018-04-29

    [Back to Results | New Search]


    Browse | Search All Available ETDs

    If you have more questions or technical problems, please contact eThesys