Research On Character Recognition Method Of Metal Detonator Image

Detonator is frequently used in engineering blasting as a blasting equipment, which is very harmful and dangerous.Therefore,all the enterprises of produce or consume detonator must manage it strictly.At present, most enterprises distribute detonator by hand,which not only spend a lot of manpower and time and is easy to produce error, reduce the production efficiency severely.Therefore the demand for automatic recognition of character on detonator has become more and more urgent.In this paper,the problems in image preprocessing,character segmentation,feature extraction and classifier design were studied.Because detonator shell is black and the surface is smooth which infected with dirt easily, and most detonator use laser engrave technology, so detonator character image have problems of low contrast, uneven illumination and background texture interference. Aiming at these problems, this paper first adopts the method of the variance normalization to normalize the image grayscale, then adopts the binarization algorithm based on stroke width and superpixel to binarize the image. Finally aiming at the problem of noise in the binary image, this paper adopts the method of connected region to remove noise.In terms of character segmentation,this paper used the method based on the connected domain combined with vertical projection, which considered the problem of stroke adhesion, stroke fraction and stroke overlap in the binary image.Then normalize the size of the individual character to perform the subsequent character feature extraction.In terms of feature extraction and classifier design, this paper adopted the stroke shape feature and BP neural network classifier. The stroke shape feature can overcome a certain degree of tilt, shift, stroke fraction and defect and so on. BP neural network has strong ability of classification and fault tolerance performance, the speed is fast at the same time. A large number of sample tests show that the recognition accuracy and speed of this system meet the performance requirements for automatic recognition of detonator character.