ABSTRACT
A temperature-constrained cascade-correlation network (TCCCN) was used to identify ultraviolet (UV) spectra of organic environmental pollutants. Library search for UV spectra is more difficult than that for infrared (IR) spectra, because the UV spectra overlap more severely than IR spectra. Besides, drift and noise in the measurement will have significant effect on UV library spectra search. Therefore, neural networks with fuzzy output should be a better alternative for the library search. The TCCCN is different from the commonly used BP networks in architecture. The processing units in the TCCCN are connected in a cascade mode, and a temperature constraint is introduced. Therefore, the TCCCN can reduce overtraining and fast training speed. TCCCN was used for library search of UV spectra in the present work and the effects of network parameters and noise were investigated. Results showed that better results were obtained with the TCCCN than with conventional correlation method.
