Research On Automatic Segmentation Algorithm And Recognition Calibration System For Stacked Bar End Image

With the rapid development of industry,China's demand for various types of bars has been significantly improved.At this time,the significance of the monitoring and statistics of the number of bars has become more and more important.At present,most of the bar counts rely on manual counting.The labor intensity of employees is large.The labor efficiency is low due to manual counting,and the accuracy is poor.Some high-precision automatic counting methods are urgently needed.Based on machine vision,the detection of bars by image analysis and pattern recognition technology is an effective way to achieve automatic bar counting.The general bar image has more noise,and the end faces of the bar are uneven.The mutual occlusion is likely to cause serious adhesion of the binary image after segmentation.In addition,for some steels,the end faces will be oxidized,and the optical properties will change,resulting in a phenomenon of uneven color.The above reasons cause the bar to automatically count accurately.In view of this,this paper starts from the end face extraction in the image,the center calibration counts two aspects,and realizes the automatic recognition and counting function of the bar end face image.The research content of this paper mainly includes the following parts:Firstly,a cloud model based method for front background separation of bar end face images is proposed.In the image of the end face of the stacked bar,the color of the end face region is close,and these pixel distributions are concentrated.In the color image segmentation algorithm based on cloud model,the uncertainty relationship between pixels is preserved,and the segmentation process is easy to control,which is suitable for the current end face extraction scene.Therefore,based on the cloud image segmentation method based on cloud model,this paper An improved algorithm is proposed in the application scenario,and the cloud background model is used to obtain the front background separation algorithm of the bar end face image.This method combines a new cross-quantization method with the existing non-uniform quantization method,designs a new merging and categorization criterion based on its statistical histogram,and improves the existing concept extraction rules.The proper segmentation of the image.Compared with the improved algorithm,the experimental results show that the proposed method is effective for color image segmentation,and is more suitable for the current bar end segmentation scenario.The average error rate of the test chart is 5.2%.Secondly,a method for extracting foreground image of bar image based on support vector regression is proposed.The method screens the segmentation result of the foreground background,extracts the parameters of the connected domain with larger area,and trains the regression model,and selects the connected connected domain according to the regression result,which is used as the end face area of the stacked bar to be counted,and removes other Image area(background part).The results show that the method can effectively remove the background and retain the image of the end face region in most cases.Thirdly,a round particle detection algorithm based on double edge template matching is proposed.Because the hard judgment method of standard circular detection is not ideal for the actual bar image,it is inspired by Sobel edge detection method and fuzzy theory.This paper designs a new set of templates for template matching to enhance Practicality and robustness to solve problems in irregular circular detection.A series of preprocessing(gradation,filtering,adaptive threshold binarization)is performed on the image of the end region to obtain a binary image to be counted.For the round-like stacked particles of similar size,the estimated radius of the round-like particles is obtained by the particle size measurement method,and no prior knowledge of the bar size is required.A set of edge templates is constructed by using this radius,and the binary image to be detected is detected by the Sobel operator in two directions.Finally,the edge template is used to match the edges,and the central marker is obtained by design constraints.result.Experiments show that this method can effectively detect such particles and can solve the effects of light stacking,hole defects and the like.The detection accuracy of the moderately stacked round particles was 94.9%.The research results of this paper help urban builders to manage various types of bars,introduce automatic processing,solve various problems of manual counting,and improve work efficiency.In addition,the stacked particle detection method proposed in this paper can be applied in other fields(such as cell segmentation)and can provide certain reference value.