Ceramic Surface Image Feature Extracting and Classifying Algorithms Based on Artificial Neural Networks

The brittleness of ceramic materials causes surface defect of ceramic products in the phase of manufacturing and processing. The defect on surface and subsurface will obviously decrease the fatigue life of ceramics. To improve the quality of end products, we have studied the mechanism of formation and expansion of surface defects in this paper. An improved image processing and recognition algorithm based on ANN is proposed to perform shape features extraction for the separated surface defects. By the results in experiments it is verified that: The previous 7 regularized defect boundary Fourier descriptors can better difference the defects with 10 features according to the features like wear defect area and discontinuous divided objects. We also establish and train two ANN classifiers separately corresponding to the features, whose accuracy is 97.5% in histogram feature classifier and 90.3% in shape feature classifier.