Research On The BP Neural Network Based Algorithm For The Glass Defect Classification

With the advancement of industrial automatization of china, the uses of machine vision and digital image processing in the industrial detection have been the important context of the industry development. The appearance of top grade and multipurpose glass product makes the enterprise's higher requirement of glass quality. The traditional human method could not meet the enterprise's high requirement and criterion in detection precision. The classification of glass defect, which is based on Back Propagation neural network, can quickly classify the defects. By this way, we can reduce the cost, improve the glass quality and production efficiency.Firstly, through the analysis of the characteristic of the glass image, a set of pretreatment flow has been designed to strengthen the image. There are several processes including linear transformation and image smoothing. Secondly, the defect's edges are detected by several edge detection operators. By analyzing the processed images, the gas-laplacian operator can effectively detect the defect's edges in the image. By comparing the two methods of image segmentation, an image segmentation method called optimal threshold iteration is used to segment the defect from the image. The result indicates that the loss of the defect's characteristic is low.After analyzing causation and trait of the usual glass defects, eight characteristic parameters, which are based on the rules of characteristic vectors, are elected as a vector to describe a defect. Then, a normal back propagation neural network algorithm is researched. For the prevalent problems in its use, several measurements are used to improve these problems. At last, the improved back propagation neural network algorithm is applied to identify the glass defects.In the end, all works of the thesis are summarized and all the problems that need to be solved in the future are also pointed out.