Research On The Algorithm Used For Grain Boundary Extraction From Metallographic Images And Design Of Intelligent Rating Software

Industrial manufacture is an important part of national economy,and steel is one of the most popular materials in industrial manufacture.Many steels have greatly reduced their operational life span because their internal structure has changed under severe service-environments.In order to ensure the normal operation of equipment and safety of users,testing the performance of steel is extremely important,and metallographic images are often used to analyze it.The vast majority of steel is composed of many small grains through metallographic microscopy.The grain size has a decisive influence on mechanical properties of steel such as tensile strength,toughness and plasticity.Grain boundary extraction from metallographic images of steel is the premise of grain size analysis.Based on digital image processing technology and machine learning algorithms,this paper focuses on the extraction and analysis of grain boundary from metallographic image and develops an intelligent rating software.The main research and achievements are as follows:1.Researching on the acquisition and preprocessing of metallographic images.Metallographic images of steel were obtained by metallographic microscope.Adaptive histogram equalization was applied to metallographic gray and color images of steel to solve the problem of uneven brightness.Bilateral filtering was used to remove the noise in obtained metallographic images without blurring boundaries.2.According to the idea of grain segmentation in metallographic images of steel.Algorithm based on adaptive Mean Shift was proposed to segmented grains in metallographic images.In order to avoid the over-segmentation of traditional algorithms,the marker-based watershed algorithm was used to mark the foreground and background to modify the gradient image,and the contours of regions were obtained.On this basis,the area of each square was mapped to the range of bandwidth parameters,and grain segmentation was finally implemented using adaptive Mean Shift.3.According to the idea of grain boundary extraction in metallographic images of steel.Algorithm used for grain boundary extraction from metallographic images based on structured random forests was proposed.By combining output of multiple decorrelated trees and capturing information from image neighborhood,the number of decision trees which was used to evaluate per pixel was reduced.Large computation and difficult definition of information gain are main problems which was caused by structured labeling.In this paper,information gain of discrete set of labels was computed by mapping to a discrete set of labels,and complete grain boundaries were extracted finally.4.Design of intelligent rating software.Based on MFC framework in Visual Studio and OpenCV vision library,four major functional modules were achieved: image reading and managing,grain boundary extraction,hand-drawn tools,grain size rating.In addition to the algorithms proposed in this paper,a large number of algorithms used in metallographic image processing were also integrated in the software to meet the needs of different steels.