The Research Of Laser Engraving Metal Detonator Character Image Recognition Method And Application

With the rapid development of information technology, the use of science andtechnology to replace manual labor has become a social trend. After recent years oftechnology development, character recognition technology has obtained a very matureresearch achievement. However, with the changes in the environment of application, thereis not a universal character recognition system. So the main purpose of this paper is to comeup with a set of character recognition algorithms to apply in the metal detonator images,to overcome some disadvantages like the low efficiency and accuracy of artificialregistering.Because the metal detonator is small, the surface is smooth and black, the detonatorcoded character is small font and thin strokes, and even showed a uncontinuous distributionof point-like.So detonators images is low-luminosity and low contrast, and there is somenoice on the images due to asymmetric illumination. To solve the irradiation reflectivephenomenon, this paper used Gradation normalization algorithm to normalize theluminance variation of images to a stable range, it effectively enhanced the contrast of theimage, weakened phenomenon of image reflective;To solve the intensity inhomogeneitiesof strokes and not-clear of images,this paper used a binarization algorithm based on thedetection of the wide line,this method combined the geometric characteristics and grayfeature of the character strokes to extract character line-features, Binarization results isvery satisfactory; To solve the problem of strokes fracture appeared in some of the images,this paper use the mathematical morphology algorithm to fill the gap between the characterand the empty in the strokes.Detonators character appears only in Arabic numerals and letters of the alphabet. Asthe detonators character structure is simple,this paper come up with a improved algorithmbased on traditional PDC algorithm. Improved algorithm considered the position of thecharacter strokes in the image as well as stroke shape information. Improved algorithm notonly Reduce the dimension,but also be able to identify characters include breaks, missingblocks, taint. It can tolerate a certain range of tilt, pan, scale differences, it has higherrecognition accuracy,better robustness and more practical than PDC. In the choice of the classifier,this paper choices the BP neural network,because it is fast in classification. Finallywe proved by experiments that the proposed scheme has both high recognition accuracy andfast processing speed, and it has high practical value.