Waveband Selection Based Feature Extraction Using Genetic Algorithm

In order to explain the geological structure accurately and quickly, we analyze the gas mixture gathered from the well by Infrared Spectroscopy Fourier Transform Spectrometer instead of gas chromatograph. In the process of the spectrum analysis, the reduction of the spectrum data dimention is very neccessary to perform. In this paper, we propose a feature extraction method is based on waveband selections using genetic algorithm, which is named FEWSGA. This approach can directly selecte eigenvalues from the limited waveband spectrum data instead of using mathematical transformation, such as the PCA (principal component analysis) and PLS (partial least squares) algorithm. Experiments results show that our method can reduce the spectrum data dimention from 1866 to 317, and the mean relative error (MRE) of the analysis model decrease from 34.68% to 26.59%. Moreover, the feature extraction from the whole waveband spectrum data using GA only reduce the data dimention from 1866 to 937. The MRE of the analysis model only reduces from 34.68% to 32.97%. Our approach has a better performance.