Practical implementation of neural networks for the interpretation of infrared spectra

✍ Scribed by Q.C. van Est; P.J. Schoenmakers; J.R.M. Smits; W.P.M. Nijssen

• Publisher: Elsevier Science
• Year: 1993
• Tongue: English
• Weight: 818 KB
• Volume: 4
• Category: Article
• ISSN: 0924-2031
• DOI: 10.1016/0924-2031(93)80001-v

✦ Synopsis

The feasibility of applying neural networks for the interpretation of infrared spectra in practice is demonstrated.

An implementation of a modular system of networks, that can easily be expanded and is user friendly, is described. The present application allows the spectroscopist to consult neural network modules to obtain information on the presence or absence of functional groups. A few examples are described. The performance of the networks appears to parallel that of human experts.

Keywora!r: Infrared spectrometry; Neural networks

Following initial attempts in 1985 to automate the interpretation of infrared spectra, in subsequent years an ambitious expert system was developed, with the aim of determining the complete structure of unknown compounds from their infrared spectra [ll. In practice, this EXPER-TISE system has not approached the desired performance [21. Experiences with this expert system made it questionable whether a system as ambitious as EXPERTISE can be created by mimicking the actions of an expert.

Meanwhile, research was started into the possibility of using artificial neural networks @INS) for the interpretation of IR spectra. The possibilities of this approach were underlined by Munk and co-workers [3,4]. In a recent paper [5], it was concluded that artificial NNs appear to perform reasonably well.