Research And Application Of Counting Method For Bundle Bar Based On Machine Vision

Automatic counting of bars is a difficult problem for steel enterprises. At present, for the steel production enterprises, the counting method mainly relies on manual counting method, photoelectric detection method and image processing method. The manual counting method mainly depends on the workers’count holding point count pen. The photoelectric detection is effective for the single rod and bar counting, for a bundle of bars will lose effect. Although the image processing method can improve the counting efficiency, but because of the adhesion problem of the bars often appear in the image, so that the counting accuracy has a great deviation. These counting methods of bars have great limitations to the counting problem. The efficiency and accuracy of manual counting is low. Photoelectric counting is not flexible, it can not adapt to the counting problem of the bundle of bar. The image processing method is still not a good solution to the adhesion problem. Therefore, production enterprises need a new efficient and accurate method for counting bars.In order to realize the high efficiency and high precision counting method, in this paper, the research of bars counting method based on machine vision. This paper mainly studies the image preprocessing, image segmentation, image target recognition, and image processing software. In this paper, we mainly study the image segmentation and image object recognition. For image segmentation, this paper has abandoned the traditional segmentation method like geometric morphology, watershed segmentation and region growing segmentation. This paper creatively changed the traditional method, uses the intelligent algorithm to identify the classification, then segmentation, and finally identify the counting results.In this paper, the classification algorithm of support vector machine is adopted. After a large number of experiments about the adhesion of bars, we find four types of adhesion, which is one single bar, two adhering bars and several adhering bars. In this paper, we select the optimal parameters of the support vector machine to train this kind of bar, and then calculate the best classification results of various types of adhesion. As a result of the support vector machine classification, this paper uses the wavelet analysis method to carry out the wavelet decomposition, the structure of the segmentation line, and ultimately the image target recognition. This identification method has high speed and high accuracy, and has high adaptability.In this paper, the method based on support vector machine is not only suitable for the problem of bars; the problem of other target identification is still the guidance and promotion.