Research And Implementation Of Web Inspection System Based On Machine Vision

Dirty spots, bright spots, holes, cracks and other flaws in the appearance of papers are collcetively called the paper defects. With the improvement of the consumers' requirements for the paper quality, especially for the specialty paper, recognizing the paper defects with artificial eye has been unable to meet the requirtments. In recent years, web on-line inspection for defect has become one of the hot issues in paper industry. The research and application at baord in the web inspection area has become mature, but has just started at home. The paper is partly surported by the Shaanxi Provincial Innovation Engineering Project (project number:2012KTCQ01-19). An in-deepth and meticulously reserching work has been done about the paper defects on-line detection and implementations in the laboratory. A web inspection system based on high speed linear array CCD camera and machine vision principle is proposed.The main work can be summarized as the following four aspects.?. The overall scheme design of the web inspection system based on machine vision. The typical components of the machine vision system is emphasisly analyzed. The hardware platform of the web inspection system is designed according to the selection rules of each module. The linear array CCD camera and Nikon Nikkor industrial grade lens is adopted in the paper image acquisition system. The lighting system using LED light source, has a front-back combined lighting meathod. The acquisition card is connected to the PC by PCI-E interface. By using the client software, the camera parameters are setted and the camera configuration file of.cff format is generated. Then, the overall design scheme of the system software is given, and the principle of the web inspection system is described.?.Study on the methods of paper defects region extracting and image features computing. The digital image processing technology and BMP device independent bitmap data structure is introduced. On the background of researching on the method of single paper defect extraction, a scheme of multi paper defects extraction is designed. Aiming at the problem of edge tracking algorithm, which can only search for signal target boundary but can not used to search for multi targets boundaries, a improved edge tracking algorithm is designed. And all of the paper defects in the paper images are extracted completely. Then, through researching on the image texture feature description and analysis method, the texture feature of paper defect images is analyzed. The calculation of texture characteristic parameters based on gary level co-occurrence matrix is also realized by programming with C++language.?. Research and design of paper defects classifier based on BP neural network. The basic principles of BP neural network and the method of determining the network structure have been learned. Combined with the characteristics of four kinds of paper defect like spots, holes, folds and cracks, the design scheme of the web inspection classifier baesd on the BP network is put forward. And firstly, the texture characteristic value of the 40 images of each paper defect kind is calculated. And these values as the training sample are used to train the BP neural network. Secondly, the texture characteristic value of another 10 images of each paper defect kind is calculated. And these values as the testing sample are used to test the BP neural network. Finally, the identification results of the classifier are analyzed, which shows that the classifier can perfectly distinguish the difficence among the four kinds of common paper defects.?. Design and implemetation of the paper defect detection system. On the basis of the basic methods and characteristics of object oriented programming, the paper video capture module, image feature extraction module, paper defect classifier module, GUI human-computer interaction interface are realized by using MFC class and Sapera++class. On view of the interference of paper defects imaging caused by imbalanced local illumination, the flat field correction module is designed to improve the imaging quality. The ability of the web inspection system to recognize four kinds of common paper defects is verified, and the anti-interference ability of the system under the interference of external lighting scource is analyzed and verified.Around the web inspection experimental platform design, video acquisition, image pre-processing, paper defects identification, flat field correction and other aspects, a in-deepth reserching work has been done and a web on-line inspection system based on BP neural network is designed. Combined with the geometric features of paper image, the texture features are imported into the paper defects classifier to recognize the paper defects. And the common paper defects such as holes, spots, folds and cracks, can be effectively identifid. The expereriment results show that the scheme can meet the design requirements.