Odor approximation is a method to produce an odor by blending odor components. Fewer odor components should produce more odors by blending them. Approximated odor based on sensed odor is called odor reproduction . This method is similar to the vision which can produce many colors by adjusting a combination of 3 primary colors. Since no primary odor has been found, an appropriate set of odor components to reproduce odor is needed.Non-negative Matrix Factorization (NMF) was used to explore odor components in the mass spectrum data space. NMF has a dimensional reduction capability which could be used for odor component analysis . Schematic of NMF is shown in Figure 1. Furthermore, NMF has non-negativity properties which match mass spectrum data. NMF with Kullback-Leibler divergence (NMF-KL) and NMF with Itakura-Saito divergence (NMF-IS) as cost function were compared for their performance in odor approximation. Both methods have different property in treating small peaks . Since we do not know how many primary odors exist, we are pursuing a smaller number of odor components to be used to approximate odor.
展开▼
