一种基于Mask R-CNN和分水岭算法的岩石颗粒图像分割方法  

A Rock Particle Image Segmentation Method Based on Mask R-CNN and Watershed Algorithm

作　　者：司晨冉 王仁超[1] 邸阔 朱品光 SI Chen-ran;WANG Ren-chao;DI Kuo;ZHU Pin-guang(State Key Laboratory of Hydraulic Engineering Simulation and Safety,Tianjin University,Tianjin 300072,China;The Frontier Technology Research Institute Limited Company of Tianjin University,Tianjin 301700,China)

机构地区：[1]天津大学水利工程仿真与安全国家重点实验室,天津300072 [2]天津大学(武清)前沿技术研究院,天津301700

出　　处：《水电能源科学》2020年第11期129-132,128,共5页Water Resources and Power

摘　　要：针对传统的岩石颗粒图像分割方法存在过分割、欠分割、需要人工调整大量参数等问题,提出了一种基于Mask R-CNN和分水岭算法的岩石颗粒图像分割方法,首先利用改进的CNN方法防止大块岩石出现过分割,接着使用提出的算法R对CNN分割出来的掩码图像和原图像进行融合,最后利用改进的分水岭算法对融合后的图像进行分割,在避免过分割的同时防止细骨料区域出现欠分割。将此方法与传统的颗粒图像分割方法应用于某堆石坝工程中,结果表明该方法有效避免了传统方法中存在的问题,实现了岩石颗粒图像的实时、精准分割,应用效果较好。Aiming at the problems of over-segmentation,under-segmentation,and manual adjustment of a large number of parameters in traditional rock particle image segmentation methods,a rock particle image segmentation method is proposed based on Mask R-CNN and watershed algorithm.Firstly,an improved CNN method is used to prevent over-segmented for large block of rock.And then the proposed algorithm R is used to fuse the mask image segmented by CNN with the original image.Finally,the fused image is segmented using an improved watershed algorithm to avoid oversegmentation while preventing fine aggregates under-segmentation of the region.This method and the traditional particle image segmentation method are applied to a rockfill dam project,and the results prove that this method effectively avoids the problems in the traditional method,realizes the real-time and accurate segmentation of rock particle images,and has good application effects.

关 键 词：Mask R-CNN 分水岭算法 岩石颗粒图像 图像分割 过分割 欠分割 

分 类 号：TV523[水利工程—水利水电工程]