Abstract: To improve the efficiency of feature selection, a feature selection method based on chaos search (CSFS) is proposed. Firstly, a mapping model for feature candidates and chaotic variables is established，which maps the feature candidates to the chaos space and realizes the interconversion between them. Secondly, the feature selection is carried out by means of the evolution of the chaotic variable. Finally, the classifier is used to evaluate the obtained feature vector. The features of weld defects are taken for example to verify the proposed method, which is compared with a gene-algorithm-based feature selection (GAFS) method. The experimental results demonstrate that the computation time of CSFS is only 61.1% of that by GAFS in the case of obtaining the feature vectors with the same recognition performance.

Key words: systematics, pattern recognition, feature selection, chaos search