Research On Metallographic Image Segmentation Based On Computer Vision Technology

Since the 1950 s,titanium alloy material has become the core material of many instruments manufacturing.Metallographic image analysis based on computer vision is one of the most widely used methods to study titanium alloy materials,and metallographic image segmentation is one of the key problems to be solved.There are mainly the following difficulties in the segmentation and analysis process of TA15 metallographic image:(1)there are black spots in the metallographic image and the brightness is not uniform;(2)the colors of some different types of metallographic structures are very similar.To solve these problems,from the perspective of image segmentation,this dissertation uses the traditional image segmentation algorithm and the convolutional neural network image segmentation algorithm to segment the equiaxial phase in TA15 metallographic image,designs and implements the metallographic image analysis system for algorithm research and organization analysis,and improves the recognition accuracy of the organization.The detailed research contents of this dissertation are as follows:Firstly,aiming at the problem of "mis-segmentation" caused by the noise and the similar color,that is,the color of grain boundary α phase or laminar structure is very similar to the color of equiaxial α phase in the light-colored TA15 metallographic image,an metallographic image segmentation algorithm fused with heterogeneous stimulus filtering is proposed: the filter template of heterogeneous stimulus is designed and realized based on the theory of heterogeneous stimulus and mathematical morphology,and the grain boundary α phase and lamellar structure are filtered effectively;the equiaxial αphase and other tissues were marked with foreground and background precisely carried out by combining the distance transformation and the mathematical morphology;the marking watershed algorithm is used to segment the equiaxial α phase,which greatly improves the segmentation accuracy of the equiaxial α phase.Secondly,in view of the dark TA15 metallographic image has obvious characteristics of texture feature,an segmentation algorithm of metallographic image based on TAU-Net:The initial segmentation of equiaxial α phase is carried out based on u-net convolutional neural network;The corresponding segmentation area is clipped on the original image according to the initial segmentation result;The texture analysis and classification of the segmentation area are carried out by using grayscale co-incidence matrix,which improves the segmentation accuracy of equiaxial α phase based on U-Net.Finally,the metallographic image analysis system is designed and implemented,mainly including the following functions: file operation,basic image processing,label preparation,special functions,quantitative analysis.The system adds and deletes the segmentation area of the organization in the way of man-machine interaction,and follows the open design concept,which has good ductility on the basis of satisfying the segmentation of the TA15 metallographic image.