The Research On Recognition Technology For Glass Flaw Based On BP Neural Network

Due to the influence of different processing technique and environmental factors on production progress of glass, there will produce some different types of flaws, which will affect the appearance quality of the glass and reduce the practical value and reprocessing rate. In order to enhance the quality of glass and make the division of production quality more convenient, it is necessary to design and realize the automatic classification method for defective glass images. According to the characteristic of defective glass images, the paper designs a classification algorithm for glass flaws based on the technologies of image processing and pattern recognition. Compared to the traditional manual classification method, it improves the efficiency and accuracy of classification.Firstly, the paper analyses the characteristic of noises in the glass image and then selects weighted median filtering algorithm to remove the interference of noises. Next, according to the characteristic of change of gray value in images, the paper uses the methods of edge detection and region merging to segment the target area image from background. So it is convenient to extract the feature via the above two steps.Secondly, according to the size and shape character of different flaws, the paper uses Hu invariant moments to extract the shape features from the images and proves that the features are invariant to translation and rotation. Considering the integrity of feature description and specialty of distribution of the gray value, the methods of two-dimensional Gabor transform and principal component analysis have been used to extract the texture features of defect. The texture features and the Hu invariant moments are used as the input of classifier.Thirdly, in order to distinguish different types of defects, the paper researches BP neural network. Owning to the problems of the traditional BP algorithm, the paper analyzes and compares several common improved BP algorithms. The most appropriate algorithm which is named conjugate gradient is selected as the classifier.Finally, some experiments are carried out to verify the effect of the total classification algorithm. The experimental results show that the algorithm can achieve a good performance of recognition. The method can lay a solid foundation for application in future practical production.