Comparison Of Neural Network Paradigms For Classification Of TM Images

Backpropagation, learning vector quantization, counterpropagation, functional link, probabilistic, and. self -organizing map are among the most popular neural network classification paradigms. Each paradigm finds a unique set of weights that will map the input pixels into their respective output classes. The usefulness of each network depends on its overall accuracy, ease of use, speed and scaling. The fastest and most accurate networks in this study were the functional link and counterpropagation. These networks, however, do not scale well to large input vectors or large training sets. A better choice is the LVQ network which achieves high accuracy, is fast, requires a minimum number of input parameters and presents fewer scaling problems.