(A) versatile adaptive neuro-fuzzy inference system(VANFIS) for flexible object classifier = 변형 가능한 물체 분류기를 위한 다용도 적응형 뉴로-퍼지 시스템

Cited 0 time in webofscience Cited 0 time in scopus
  • Hit : 230
  • Download : 0
A mixed mode neuro-fuzzy classifier named Versatile Adaptive Neuro-Fuzzy Inference System(VANFIS) is presented with versatile structure for different classification problems. The versatile structure which is designed for flexible classification problems is adjustable depending on the number and type of input vectors. Using proposed learning algorithm that evaluates degree of saturation state, the learning sequence is modified according to its speed and accuracy. Additionally, versatile configuration of its front-end architecture sustains more than 90% inference accuracy for all 25 objects in image database. Exploiting mixed mode systems and its bit width optimization, this classifier is implemented with 1.2mW and $0.765mm^2$ power and area consumption, respectively. Also it achieves 11.35MCUPS(Connection Updates Per Second)/mm2 as learning ability and 1MFLIPS(Fuzzy Logic Inference Per Second) as inference ability. Most of all, this chip can operate on-line learning in real-time . And it is verified with extremely occluded object and novel view of object based on 200 images. It scores 94% object classification accuracy which is improved about 14% from previous approach.
Advisors
Yoo, Hoi-Junresearcher유회준researcher
Description
한국과학기술원 : 전기 및 전자공학과,
Publisher
한국과학기술원
Issue Date
2010
Identifier
419161/325007  / 020083301
Language
eng
Description

학위논문(석사) - 한국과학기술원 : 전기 및 전자공학과, 2010.2, [ vii, 58 p. ]

Keywords

object classification; versatile; neuro-fuzzy; flexible classifier; on-line learning; 실시간 학습; 물체 분류; 변형 가능한; 뉴로-퍼지; 적응형 분류기

URI
http://hdl.handle.net/10203/36594
Link
http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=419161&flag=dissertation
Appears in Collection
EE-Theses_Master(석사논문)
Files in This Item
There are no files associated with this item.

qr_code

  • mendeley

    citeulike


rss_1.0 rss_2.0 atom_1.0