Using Support Vector Machine Optimized with ACO for Analysis of Multi-component of Spectral Data

This paper present a improved method based on the principle of soft sensor for analyzing overlapped spectra in the case of small samples. The method combines wavelet packet transform and support vector machine for improving the performance of noise reduction filtering, feature extraction as well as improving the prediction accuracy of the soft sensor model. Wavelet transform decomposes the original signal into wavelets of multiple frequency bands to filter out clutter other than the signal band. The feature vectors of the spectral signals are extracted and applied as inputs to the SVM. Support vector machine is applied for least squares regression of input and output data to solve the nonlinear problem of multi-component systems. Ant colony algorithm is applied for optimizing of training parameters. Proper parameters can improve the accuracy and generalization ability of the method. The multi-component overlapped spectra is analyzed by using the method, three kinds of ions of Cu(II), Co(II), Pb(II) the average relative errors are <6%. The result shows the system performed very well. This method offers an promising method for analysis of multi-component overlapped spectra.