基于数字图像处理技术的金相分析研究

Research of metallographic analysis based on digital image processing

采用和改进了人际交互式图像分割算法Live-wire,对电气火灾中的典型火烧金相图和一次短路金相图进行处理,结合数字形态学操作,精确提取了两类金相图中的典型胞状晶和柱状晶结构;分析和讨论了多种结构描述子对晶格分类的作用,结果表明,离心率和晶格标记方差两个描述子能有效区分两类晶格结构;采用聚类分析和Fisher判别法,实现了对两类金相图的自动分类。

A human computer interaction image segmentation algorithm named Live-wire was adopted and improved to process typical metallographic image. Digital morphology methods were adopted to eliminate off redundancy information and digital date. Precise outline of crystals were captured. Effects of several segment descriptors on crystal categories classification were analyzed and discussed. It indicated that eccentricity and variance of crystal checkmarks can be used as the final parameters of the automatic classification. Cluster analysis and fisher discrimination methods were used. Burn molten and primary short circuited melted marks were effectively distinguished.