Application of Madaline and Kohonen to Resolution of Overlapping Spectrum and Determination of Tungsten and Molybdenum
Jinghui Li · Jisuanji yu yingyong huaxue · 2002
Kohonen Neural Network has been applied to wavelength selection. Those wavelengths that represent feature of spectrum most can be selected in full spectrum. Absorbance of selected wavelength is taken as input of Madaline Neural Network, and G(λ) =ΣAiexp[-4(In2)(λ-t)2/2] is set as target function of output to resolve overlapping spectrum. Tungsten and Molybdenum have been determined simultaneously. This method which diagnostic spectrum representing the full spectrum has been applied to resolution of overlapping spectrum can cut down a great deal of data, and results are satisfied.