Classification And Retrieval System Of Presensitized Plate Of The Surface Flaw

With the increasing demand of PS plate, It is hard to guarantee the global stability and the overall quality of the PS plate. In order to obtain excellent printing effect, the higher requirements are put forward by the market, and with this parameter evaluate the quality of PS plate as a whole. Artificial visual inspection has been the most PS plate detection method by domestic manufacturer, and it has large limitation, which is incomparable with high speed, mass, automation of production pattern, and runs counter to enterprise’s development goals. In order to improve the testing speed of the PS plate of production, accuracy of detection of surface flaws and defects detection, and real-time control of production line, there is great significance in developing the PS plate flaw classification and retrieval system.According to the research of technical indicators, in this article, the PS plate of the surface defect detection system are discussed in hardware design and software design. Hardware design includes the selection of the CCD light, the selection and calculation of the imaging CCD at the same time for multiple CCD installation method. At Software design, according to the digital image processing technology and BP neural network pattern recognition technology, the PS plate of the common defects image processes, analysis, classification and recognition are did.First, due to various reasons such as imaging environment is not stable, leading to the drawbacks of image quality is poor, and defect target is difficult to distinguish. At first, in order to get a clear defect image, this paper adopts the method of local enhancement-block variance operation method to enhance image, and then adopts the method of median filter to filtering out the noise in the image.After image preprocessing, the method of iterative threshold segmentation is adopted to get the image of binarization segmentation. Because after binarization processing of images there is some small noise dot, this article adopts the method of morphological operation to filter the small noise dot. After Image segmentation is completed, shape features, geometric features, gray feature extraction which are distinguish in the shape of the defect characteristics are adopted, including the area, perimeter, rectangular, aspect ratio, the circular degrees, gray variance.Finally, the BP neural network is designed, and using the above characteristics as input, several common defect types as output, the neural network is trained. Using the trained BP neural network, some random samples are selected to do classification for PS plate samples, the experimental results show that the recognition accuracy of the results meets the requirements of the technical indicators.