engleski

Heuristic Algorithms for Extracting Relevant Features in Signal Analysis

Extraction of relevant features is essential stage in a pattern recognition and classification system. Goal of the feature extraction algorithm is to find feature subset where relevant information for recognition is contained in minimal number of features. Proposed algorithms are based on the assumption that features with better individual discrimination ability will also be better in combination with other features. Features are first extracted from the initial set, then sorted according to their individual fitness. Sorted set is used to form the search tree. Two heuristic algorithms are proposed: the first one performs the depth first search, bounded with required increase of fitness function and the second one is based on genetic algorithm. Their performances are compared with complete search and sequential search (FSS, BSS) algorithms.

feature extraction; pattern recognition; signal analysis

nije evidentirano