| In the course of the quality detection of the glass products , with the enlarging of the production scale , the enhancing of production speed and more and more severe request of quality , the artificial detection on the production line hasn't been competent. Using computer vision , image processing and the technique of artificial intelligent to carry out the detection of the product quality is imperative under the situation instead of artificial detection . The mould number identification system is a subsystem of computer vision detection system on-line of the glass products . The subsystem was developed by the cooperation of the Electronic Technique Research Institute of Guang'xi Normal University and GuiLin Glass Factory .The detection of glassware's mould number have become an important part of the glassware's quality detection . At present, we haven't developed a relevant mould number identification system in our country . The developed mould number identification system overseas identifies the mould number by reading the protuberant point code or the 7-segment type code , but it was uncomfortable for the practical use of domestic products . Under this situation , the mould number identification system developed by writer broke through the way by reading code and direct reads common character. it has common applicability and especially suits the urgent demand of domestic glass products which has not become the standard code of mould number .The mould number identification system's design is as following:(1) Detect the location of mould number , select suitable camera , light source and image gather card , obtain digitize mould number image in the special space .(2) Obtain a single character image which is waited recognition by image preprocessing . The image preprocessing include five steps: extracting sensitive part , automatic binarization , eliminating confusion of voices , automatic segmentation and normalization processing . Among them , the normalization processing include single character's crevice stuffing . edge smothing and frame locating . The character 's frame is located by the character image transfering, the character's barycenter determining and the frame's edge locating .(3) Structure BP neural network , extract the eigenvector of single character and input BP network to recognise . The connection weight between nerve cells is obtained from the BP neural network which has been trained .(4) Order the characters which have been recognised by BP neural network for getting united and standard mould number, check mould number by corresponding standard parameter . Final!}' output the mould number and the check results .Comparatively the design of the same kind of identification system , the design has several notable characters: the most novel aspect of the design is that it adopted the character recognition algorithm based on artificial neural networks , it enhanced the recognition ability and the recognition rate , it made the system imelligentize and it has more agility and more strong adaptable ability ; In addition . it is different from the common course of image preprocessing on the binary algorithm . theimage is binarized using a dynamic threshold's scheme , it has very strong self-adaptability ; Meanwhile , in the course of recognition , the mould number images were analysed by combining the knowledge of special yield and the character segmentation/recognition algorithm based on the recognition feedback was adopted , it adapted the practical problems and it has strong pertinence .In this design, 3-ply structure BP neural network was adopted . During the training of BP network . contraposing local minimal value problem and low convergence speed problem , the writer adopted improved BP algorithm and achieved very good result.Base on the experimental result, the writer analysed it from different mould number images and the corresponding image preprocessing algorithms , and summarized that locating the character right in the center of frame area is the key...
|
PDF Full Text Request