Automatic classification of mass spectra by means of digital learning nets-existence of characteristic features of chemical class in mass spectra

摘要

The application of digital learning nets to the classification of mass spectra assumes that there is a relationship between the data and the defined classification categories. The validity of this assumption is demonstrated. A well defined subset of data from a chemical class can be compiled, which is sufficient to enable other spectra from that category to be classified. Further evidence is provided by a comparison of the behaviour of the net with experimental spectral data and random patterns and the consideration of the results of a 28 group classification.