A method derived from Genetic Algorithm, Principal Component Analysis and Artificial Neural Networks to enhance classification capability of Laser-Induced Breakdown Spectroscopy

Selection of characteristic lines is a critical work for both qualitative and quantitative analysis of laser-induced breakdown spectroscopy; it usually needs a lot of time and effort. A novel method combining genetic algorithm, principal component analysis and artificial neural networks (GA-PCA-ANN) is proposed to automatically extract the characteristic spectral segments from the original spectra, with ample feature information and less interference. On the basis of this method, three selection manners: selecting the whole spectral range, optimizing a fixed-length segment and optimizing several non-fixed-length sub-segments were analyzed; and their classification results of steel samples were compared. It is proved that selecting a fixed-length segment with an appropriate segment length achieves better results than selecting the whole spectral range; and selecting several non-fixed-length sub-segments obtains the best result with smallest amount of data. The proposed GA-PCA-ANN method can reduce the workload of analysis, the usage of bandwidth and cost of spectrometers. As a result, it can enhance the classification capability of laser-induced breakdown spectroscopy.