K. Kolev
Abstract. The purpose of this research is the reduction of feature spaces to decrease the features for qualifying the ready-to-cook minced meat products. The elaboration uses hyperspace reduction methods, through the selection of features or through the extraction of features. The capacities have been analyzed of the selection method through branches and borders selection (BBS), the method of sequential forward selection (SFS), the maximum autocorrelation factor method (МAF) and the method of multi-resolution approximation (MRA). In the experiment ready-to-cook minced meat products complying with the “Stara Planina” standard have been used. The experimental results demonstrated that the selection methods for the features are not suitable for determining the quality of ready-to-cook minced meat products in real time. For comparison of efficiency one and the same classification algorithm C4.5 has been used. The results obtained showed that a combination of MRA and МAF is most suitable for reducing the hyperspectral feature spaces of ready-to-cook minced meat products for the realization of a computer platform for objective determination of the quality of ready-to-cook minced meat products.