| 摘要 |
FIELD: physics. ^ SUBSTANCE: method uses digital video cameras and comprises imaging by means of spectral illumination, analysis and classification of rail surface images. Rail surface scanned, colour image obtained is converted into zero-contrast picture, normalised and binarised. Interferences are eliminated that distort rail fragments images. Informative feature is estimated to automatically classify surface defects by means of artificial neuron network. In case it fails to define defect class to required accuracy, rule-situational dynamic expert system is automatically initiated to define defect class from morphological and genetic features and defect parameters. ^ EFFECT: higher efficiency of defect identification. ^ 4 dwg |
