A statistical fuzzy grade-of-membership approach to unsupervised data clustering with application to remote sensing
- This graduate work has been published as open access. Learn more... - this link will open in a new window
A fuzzy Grade-of-Membership (GoM) paradigm in the context of unsupervised data clustering is presented. GoM partitioning characteristics are examined through theoretical and empirical means, revealing a robust and general framework. GoM is compared and contrasted with three conventional algorithms, namely, vector quantization (VQ), fuzzy c-means (FCM), and deterministic annealing (DA). The comparison is facilitated by a restatement of equations in a form that reveals fundamental mathematical relationships and suggests an approach to efficient practical implementation. Here, GoM distinguishing characteristics are highlighted, strengths and limitations are identified, and applicability conditions are established. Developed theory is confirmed and GoM partitioning characteristics are further illuminated by experimental comparison of GoM, VQ, FCM, and DA partitioning for several simulated annealing and reduced complexity stochastic relaxation is also developed and presented here. GoM partitioning is newly applied to a remote sensing problem with cloud climatology data. In particular, monthly cloud product data from NASA's International Satellite Cloud Climatology Project (ISCCP), is examined for climatic classification.
0799: Remote sensing