Pattern recognition for holonic manufacturing systems

The paper addresses to the pattern recognition problem within the framework of holonic manufacturing systems. A novel methodology with enhanced adaptation capabilities devoted to features extraction is suggested. It involves a flexible genetic selection of relevant features, in accordance with the specific properties of the patterns that have to be recognized. The competing features are determined by means of principal component analysis, bi-dimensional Fourier transformation and grey-levels analysis. The problem is formulated as a multi-objective optimisation, addressing to both classification accuracy and parsimony. The experimental results reveal the improvement of overall performances of the pattern recognition subsystem.