Research On The Object Segmentation Technology Based On Human Vision And Its Applications

To realize Industry 4.0,automatic quality assessment of products has become the development tendency.Among all the assessment methods,machine-vision-based image detection technique is considered as the most important and promising one.In fact,most defects,either on the product surface(e.g.steel sheet surface defects in rolling process)can be characterized as the objects with specific features.Therefore,by visually analyzing and quantitatively characterizing the features,each different object might probably be recognized and measured so that the product quality can be evaluated,and then the efficiency of production and the degree of industrial automation could be improved.Although a large number of detection methods and systems have been applied to industrial production processes,these studies often exist as a result of the strong particularity of the industrial production process,and the wide adaptability of the proposed methods or systems for other industrial scenarios.The large amount of calculation results in many problems,such as the poor immediacy and the poor online real-time feedback capability,the poor robustness to the environment and the special working conditions,which greatly restrict the generality,degree of automation,intelligence and batch processing capability of industrial detection.On the other hand,in actual production detection process,it is found that the object edge and low-contrast interference often impede the improvement of detection accuracy.Therefore,it is very important to study the adaptive and automatic object detection technology and apply it to the actual industrial production.The paper studies the object segmentation by putting forward the theoritical technology and applying it to the actual project cases to improve the accuracy,objectivity,efficiency and application scope of the detection process,.The main research contents are as follows:(1)By imitating the classification and processing mechanism of human vision to different scenes,an adaptive object segmentation technique is proposed and verified by four practical application cases in industrial production.The accuracy,detection rate and robustness of the proposed technique are discussed according to the experimental results.The technique not only provides a solution for the wide application,automation and intellectualization of detection technology,but also has certain application value in engineering practice;(2)Aiming at the objects with non interference and high sensitive edge,a segmentation strategy based on superpixel preprocessing and local gray statistical features is proposed.The experimental system is set up with the assessment of the ground surface quality of brittle materials.The rationality and accuracy of the strategy are verified by qualitative and quantitative analysis,and the potential value of the segmentation strategy in grinding model judgment and grinding parameters adjustment is also verified;(3)In order to improve the segmentation accuracy of the interfered and high sensitivity edge targets,a two-level superpixel object segmentation strategy is adopted.The experiment of transparent oil detection in the process of metal cold rolling is conducted,and the experimental results are analyzed to verify the effectiveness of the strategy.It is also proved that the segmentation strategy is practical for monitoring of the running state of the cold rolling system;(4)The segmentation strategy based on the combination of saliency estimation and fuzzy clustering is applied to solve the segmentation problem of objects without interference and low sensitivity.The segmentation strategy is applied to the subsurface damage detection of brittle materials after grinding.The correctness and accuracy of the strategy are verified through experiments and result analysis.Through further analysis of the detection speed and robustness,it is proved that the segmentation strategy could be widely applied in batch processing;(5)Combined with saliency map and grayscale original image,a segmentation strategy against interfered and low sensitive edge objects is constructed.The rationality and efficiency of the strategy are verified through the application of subsurface white layer detection after metal machining.By further analyzing the automation and robustness,it is proved that the segmentation strategy is feasible and advanced.