Research On Bar Identification Method Based On Improved Watershed

Bar real-time identification and counting,has been the steel mill is committed to solve the problem.The steel mill initially used the method of manual counting to identify the bar,which can not meet the needs of industrial automation.Then people put forward the electromechanical method,image processing method for bar identification research,for the end surface oxidation dimming,adhesion of the bar,the use of electromechanical method and the existing image processing method are difficult to accurately identify,this topic proposed to improve the watershed method for bar identification count,the method mainly includes bar image preprocessing,bar target recognition and tracking.First of all,the bar image preprocessing,the end surface oxidation dimming bar on the point cloud map there is a multi-peak problem,the use of power spectrum to extract the low frequency part of the bar image to solve the multi-peak problem of the bar image,the extraction of the low-frequency part of the maximum entropy threshold segmentation,to achieve automatic extraction of the bar area.Secondly,aiming at the problem of target recognition of bar,an improved method of marker watershed is proposed to identify the bar,which fuses the marker point with the multi-scale reconstruction,and the fusion results are identified by watershed transformation to obtain the bar contour.The distance transformation of the bar area is carried out,the marker points in the distance graph are filtered,and the local extremum region corresponding to the low frequency part of the bar is extracted,and the marking point supplement is obtained by expanding the local extremum region,and the complete marking point corresponding to the bar target through the marker point screening and the marker point supplement is taken,The foreground marker applies the watershed transform to obtain the background marker.Then combined with multi-scale morphology to reconstruct the bar gradient map.reconstruct the result fusion foreground,background marker,the use of watershed transformation to obtain the bar contour,superimposed on different scales of the results to obtain the complete contour of the bar.Finally,the continuous bar image is tracked,the bar image is projected,the overall displacement of the image is estimated,and the bar count is realized by the position of different frame bars.The experimental results show that the improved watershed method proposed in this paper can accurately identify the count for the bar with dark oxidation and adhesion of the end surface.Aiming at the complex bar production line,the system of this project can meet the real-time and reliability requirements of bar recognition count,and has a good guiding significance for other kinds of circular target recognition count.