Recognition of Defects in Steel Surface Image Based on Neural Networks and Morphology

A hybrid image recognition approach based on ´Neural Networks and Morphology is presented to detect various defects in steel surface image, because signal-to-noise ratio of steel surface image is very low, and defection targets are small and their shape is irregular. Firstly, we present the comparison of different segmentation results of steel surface image under BP neural and RGB threshold value method, and draw a conclusion that BP neural network is better than the RGB threshold value method. Secondly, noise is filtered by morphological processing. In this way, different defection was auto-segmented and located. Experiment results show this method can detect different defection and is strongly robust.