The Application of Hybrid Evolving Connectionist Systems to Image Classification
Nikola K. Kasabov*, Steven A. Israel** and Brendon J. Woodford*
*Department of Information Science
**Department of Surveying University of Otago PO Box 56 Dunedin New Zealand
This paper presents a methodology for image classification of both spatial and spectral data with the use of hybrid evolving fuzzy neural networks (EFuNNS). EFuNNs are five layer sparsely connected networks. EFuNNs contain dynamic structures that evolve by growing and pruning of neurons and connections. EFuNNS merge three supervised classification methods: connectionism, fuzzy logic, and case-based reasoning. By merging these strategies, this new structure is capable of learning and generalising from a small sample set of large attribute vectors as well as from large sample sets and small feature vectors. Two case studies data are used to demonstrate the effectiveness of the methodology. First, an environmental remote sensing application, and second, large scale images of fruit for automated grading. The proposed methodology provides fast and accurate adaptive learning for image classification. It is also applicable for on-line, real-time learning and classification.
This article is published under a Creative Commons Attribution-NoDerivatives 4.0 International License.