Cold-rolled Strip Plate-profile Recognition Based On Image Processing Method

For producing technology of cold-rolled strip in huge iron and steel enterprises, the quality of plat-profile is a critical control indicator in the manufacturing, which influence directly and determine the final quality of steel sheets. Therefore, the plate-profile recognition technology holds a very important degree in the producing process. Seen from the current practical applications of plate-profile recognition of cold-rolled strips in the iron and steel enterprises, the major method to getting the plat-profile state is still using the inner stress distribution of steel sheets in the process of rolling indirectly. This method has some disadvantages such as the expensive device, complicate operation, destroyed easily and maintained hardly. Furthermore, the reliability of this method cannot be guaranteed completely because of the outside changing condition and the factor of the sensor itself. So the main countries which manufacture steel are studying new plate-profile recognition method and trying to seek a plate-profile recognition system with simple structure, few investment, stable capability and that the detecting accuracy can satisfy the manufacture requirement.Relying on the Chongqing city natural science fund projects, which are"the producing quality monitor system of iron and steel industry based on new intelligent theory (Project No: 7369)"and"the plate-profile control artificial intelligence system of cold-rolled strips"(transverse science and technology project, Contract No: GK05121), this dissertation take plate-profile of cold-rolled strips as study object and aim at establishing a plate-profile recognition system with simple structure, stable capability, and whose detecting accuracy can satisfy the producing requirement. The dissertation mainly studies the plate-profile imaging at the common light sources, data gathering, pre-processing of plate-profile image, segmentation of plate-profile image and image processing based on morphology, and then put forward the plate-profile recognition algorithm based on artificial neural networks, Top-Hat and comparison. All these methods were tested in practical production, and the results verify their efficiency in recognizing plate-profile of common cold-rolled strip.The main contents and conclusions of this dissertation are as follows:●In chapter 2, the method to get effective plate-profile at the common light sources is put forward; the basic requirements and basic parameters of plate-profile recognition system are presented too. The recognizing accuracy is determined by image gathering IV system, the real-time performance is restricted by rolling speed and computer operating speed, and the recognition accuracy is correlative not only with the accuracy and real time performance of the plate-profile recognition system but also with the plate-profile recognition algorithms. The software and hardware design frame of the data gathering and imaging system of plate-profile image is offered.●In chapter 3, image pre-processing method is mainly discussed such as the histogram equalization, direct gray-level transform and histogram normalization. The solution to detail losing and brightness aberrance caused by conquering histogram equalization is developed. In addition plate-profile image noise and its smoothness are discussed, and then the experimental result and analysis of these three image smoothness, i.e. field averaging, reciprocal weighted mean and median filtering, are presented.●In chapter 4, an image binarization algorithm based on class mean is presented which is proved later that is equivalent to Otsu's algorithm. Then, histogram equalization algorithms based on image binarization and image binarization algorithm based on histogram equalization are put forward. By analyzing the processing results of the plate-profile image fringe respectively using three image fringe detecting methods, such as the first order fringe detecting operators( Roberts operator, Sobel operator, Prewitt operator) , the second order fringe detecting operators (Laplacian operator,Kirsch operator) and Canny fringe detecting operators, Canny algorithm is verified most effective in this system.●In chapter 5, the basic criterion and general method of binary image de-noise algorithm and gray-level image enhancement algorithm based on morphology are investigated. By using binary morphology algorithm, the problem about noise after image binarization is resolved effectively, and by using gray-level morphology image enhancement algorithm, the shortcoming about image detail losing caused by histogram equalization is overcame effectively. Above two productions lay the foundation for image analysis and image nature obtainer.●In chapter 6, three practical plate-profile recognition algorithms are developed. Firstly, the plate-profile recognition algorithm based on TopHat enhances plate-profile image by using TopHat. Secondly, to do binarization on the enhanced image, and then get the defective image. Thirdly, eliminate the interferential signal in the binary defective image. Finally, analyzing the image, and then the sort of the plate-profile can be determined. The average accuracy of this method is 50%.The plate-profile recognition algorithm based on artificial neural networks firstly proceeds to histogram equalization on image, which largely improve the contrast of the processed image and the definition of the image fringe. Then do morphology enhancement to improve the image effect. Finally, take the processed image's fringe by using canny operator and classify the characters by taking the average value, square difference and the contrast statistic characters of image as the inputs of the artificial neural networks interpolator. The plate-profile recognition system designed as this method is well-applied during the plate-profile recognition in a cold mill plant and the average accuracy is up to 92%.The plate-profile recognition algorithm based on comparison takes the last image as the"standard plate-profile image"of the current image and gets the defective image by subtracting the current plate-profile image from the standard plate-profile image. Then do binarization on the defective image, eliminate the interferential signal in the binary defective image and analyze it. So the sort of the plate-profile can be determined. The average accuracy of this method is 97%.The plate-profile recognition method based on image processing is a rather new research subject and the relative datum is lack at home and abroad. In order to make the plate-profile recognition system apply to the production of cold-rolled strips more effectively and then improve quality and efficiency of cold-rolled strips production, we must practice constantly to discover problems, analyze the problems and resolve them.