摘要
在金相组织检测环节中,需在不同放大倍数的显微镜下提取晶粒聚集区域并计算参数。使用传统DBSCAN聚类算法进行聚集检测时,因每张图像晶粒聚集的密度不同、显微镜放大倍数不同等问题,需要反复实验以确定DBSCAN算法的两个基本参数。针对上述问题,本文提出一种改进的自适应DBSCAN算法,通过平均晶粒大小,确定领域密度阈值(MinPts),利用自适应的方式调整领域半径(Eps),并采用k-d树数据结构加速聚类过程。实验结果表明,使用本文方法能够自动检测出晶粒聚集区域,具有一定普适性,有望提高检测效率。
In the detection of metallographic structure,it is necessary to extract the grain aggregation area and calculate the parameters under the microscope with different magnification.When the traditional DBSCAN clustering algorithm is used for clustering detection,repeated experiments are needed to determine the two basic parameters of DBSCAN algorithm due to the different density of grain aggregation in each image and the different magnification of microscope.To solve the above problems,this paper proposes an improved adaptive DBSCAN algorithm,which determines the domain density threshold(Minpts)through the average grain size,adjusts the domain radius(Eps)in an adaptive way,and uses the K-D tree data structure to accelerate the clustering process.The experimental results show that the method presented in this paper can automatically detect the grain aggregation area,which has certain universality and is expected to improve the detection efficiency.
作者
周润
滕奇志
ZHOU Run;TENG Qizhi(Institute of Image Information,School of Electronics and Information Engineering,Sichuan University,Chengdu 610065,China)
出处
《智能计算机与应用》
2021年第4期44-48,共5页
Intelligent Computer and Applications
作者简介
周润(1996-),男,硕士研究生,主要研究方向:模式识别与智能系统;通讯作者:滕奇志(1961-),女,博士,教授,博士生导师,主要研究方向:图像处理、图像/视频编码通信,Email:qzteng@scu.edu.cn
