Login

Publications  •  Project Statistics

Glossary  •  Schools  •  Disciplines
People Search: 
   
Title/Abstract Search: 

Dissertation Information for Min He Ji

NAME:
- Min He Ji
- (Alias) Minhe Ji

DEGREE:
- Ph.D.

DISCIPLINE:
- [No Discipline Recorded]

SCHOOL:
- University of South Carolina (USA) (1996)

ADVISORS:
- None

COMMITTEE MEMBERS:
- None

MPACT Status: Incomplete - Not_Inspected

Title: Towards a complete fuzzification of supervised classification for remote sensing digital image processing

Abstract: In characterizing biophysical phenomena through remotely sensed digital image data, a fuzzy partition matrix has proven more representative of the real situation than the conventional thematic model. The geographical concepts of concentration and diffusion, important to many areas of geographical research and applications, can be readily supported with such a data model. In a statistical supervised classification, however, a fuzzy partition matrix based on spectral signatures derived through a conventional two-value logic is a faulty representation of the real situation. In order to improve the accuracy of a fuzzy partition matrix as well as a conventional thematic map, an adequate fuzzy approach must be incorporated into supervised training, which is the determinant of the final accuracy level for a classification. This dissertation was attempted to achieve a complete fuzzy supervised classification by fizzifying both training stage and error analysis stage of the conventional maximum likelihood procedure. It is reasonable to assume that a certain degree of uncertainty exists in both human knowledge about the training site and spectral coherence of the training pixels delineated by the analyst with respect to the class in definition. Fuzzy training copes with the data uncertainty in the early stage of image classification in order to derive statistical parameters that are closer to real ones. It is also a design to ensure a properly transmission of uncertainty information from the raw image data to the final product. A self-organizing iterative fuzzy parameter estimator (ISOFPE) was developed by modifying Bezdek's Fuzzy c-Means engine. Fuzzy error analysis was performed through the construction of a fuzzy concentration table and the computation of a non-fuzziness index for each class. Both synthetic data and Landsat Thematic Mapper imagery were used to test the fuzzy procedure. In this study, classification results based on the fuzzified spectral signatures were favorable over those with the conventional training method. Major findings include a statistically significant improvement in the overall classification accuracy and the accuracy of the classes with a high degree of confusion, and a fuzzy membership distribution that is more realistic and consistent to human knowledge. Future directions include class-based fuzzification, fuzzy Kappa coefficients, and the incorporation of contextual information in fuzzy segmentation.

MPACT Scores for Min He Ji

A = 0
C = 1
A+C = 1
T = 0
G = 0
W = 0
TD = 0
TA = 0
calculated 2008-01-31 06:21:34

Advisors and Advisees Graph