基于多聚类中心的加权聚类铸体薄片图像分割方法  

Weighted clustering casting thin slice image segmentation method based on multi-cluster center

作　　者：郑洋洋 张冲[1,2] 伍薇 焦立群 ZHENG Yang-yang;ZHANG Chong;WU Wei;JIAO Li-qun(Key Laboratory of Exploration Technologies for Oil and Gas Resources,Ministry of Education Yangtze University,Wuhan 430100,China;College of Geophysics and Petroleum Resources,Yangtze University,Wuhan 430100,China)

机构地区：[1]油气资源与勘探技术教育部重点实验室,武汉430100 [2]长江大学地球物理与石油资源学院,武汉430100

出　　处：《矿物岩石地球化学通报》2023年第4期882-889,共8页Bulletin of Mineralogy, Petrology and Geochemistry

摘　　要：针对人工利用显微镜对岩石铸体薄片进行鉴定的低效和普通聚类分割算法对图像边缘分割效果差的问题,本文从岩石特征出发,提出了一种基于聚类分割算法的多聚类中心加权分割新方法,并使用该方法训练了一个适用于岩石铸体薄片的分割模型。使用该模型可以快速地完成对大量薄片的鉴定任务。该方法通过构建新的聚类距离提升了对薄片各组分边缘和内部的分割效果,将多聚类中心的方法和加权的聚类方法相结合,进一步增强了对薄片边缘部分的分割。利用来自疏松岩心的铸体薄片图像进行分割实验,并将本文提出的聚类分割新方法与普通的聚类分割算法的结果进行比较,发现本文方法的分割误差比普通分割方法的误差最高降幅达37.2%。与现有深度学习分割算法对数据量和数据类型的高要求相比,多聚类中心的加权聚类分割算法更适合地质领域中的分割任务,具有较高的应用价值。Aiming at the inefficiency of manual identification of rock casting sheets using microscope and the poor effect of common cluster segmentation algorithm on image edge segmentation,this paper proposes a new segmentation method based on cluster segmentation algorithm with multi-cluster center weighing from rock features,and uses this method to train a segmentation model for rock cast sheets which can identify a large number of thin slices quickly.This method improves the segmentation of the edges and interiors of each component of the wafer by constructing new clustering distances,and further enhances the segmentation of the edges of the wafer by combining the multi-cluster center method and the weighted clustering method.The results of the new clustering segmentation method in this paper are compared with those of the common clustering segmentation algorithm using the cast thin slice images from sparse cores for segmentation experiments,and it is found that the segmentation error of this method is reduced by up to 37.2% compared with that of the common segmentation method.Compared with the existing deep learning algorithms with high requirements on data volume and data type,the weighted clustering segmentation algorithm with multi-cluster center is more suitable for segmentation tasks in geological filed and has high application value.

关 键 词：多聚类中心 均值聚类 铸体薄片 图像分割 

分 类 号：P575.9[天文地球—矿物学]