仓面大粒径堆石的图像处理与粒径识别  

Image Processing and Particle Size Identification of Large Particle Size Rockfill on Placement Surface

作　　者：刘易 付立群 徐小蓉 郭欣晟 金峰[2] LIU Yi;FU Liqun;XU Xiaorong;GUO Xinsheng;JIN Feng(School of Hydraulic&Environmental Engineering,Changchun Institute of Technology,Jilin 130012,China;State Key Laboratory of Hydroscience&Engineering,Tsinghua Univ.,Beijing 100084,China;School of Water Resources&Hydropower Engineering,North China Electric Power Univ.,Beijing 102206,China)

机构地区：[1]长春工程学院水利与环境工程学院,长春130012 [2]清华大学水沙科学与水利水电工程国家重点实验室,北京100084 [3]华北电力大学水利与水电工程学院,北京102206

出　　处：《三峡大学学报（自然科学版）》2022年第6期28-34,共7页Journal of China Three Gorges University：Natural Sciences

基　　金：贵州省2021年科技支撑计划项目(黔科合[2021]一般359);中央高校基本科研业务专项资金项目(2021MS042);国家自然科学基金重点项目(52039005);中国水利水电科学研究院水利部水工程建设与安全重点实验室开放基金(IWHR-ENGI-202106);清华大学水沙科学与水利水电工程国家重点实验室开放基金(sklhse-2022-C-03)。

摘　　要：粒径是表征非均匀堆石骨架特征的重要参数.为解决传统人工巡视目测堆石质量的不足,本文结合图像识别和人机交互技术开发了RockGUI软件,实现了非接触式的堆石粒径半自动化识别.采用阈值分割结果限制分水岭注水点的个数,减少了传统算法中由于种子点过多而造成的过分割现象,并对图形分割结果进行人工修正,如过分割区域合并、欠分割区域再分割等,大大提高了质地不均、重叠严重堆石体的识别精度.本文选取贵州省风光水库作为试验工程,在典型仓取样点拍摄堆石照片,并采用RockGUI进行图像识别.结果发现,图像识别出的堆石粒径略大于人工标注测量出的堆石粒径,相对误差不超过20%;识别的堆石外接圆直径非常接近堆石实际长轴,二者粒径的相对误差小于20%;该仓小于300 mm逊径堆石料的质量占比约2%,堆石体质量良好.软件生成的堆石粒径分布曲线,可指导工程人员对堆石仓面的质量进行量化等级评价.Particle size is an important parameter to characterize the heterogeneous rockfill skeleton.In order to solve the shortage of traditional manual inspection for the rockfill quality,the RockGUI software combining image segmentation and man-machine interaction technology is developed to realize the non-contact semi-automatic identification of rockfill particle size.The result of threshold segmentation is involved to limit the number of watershed water-injection points,which reduces the over-segmentation phenomenon of traditional watershed algorithm.Furthermore,the segmentation results can be modified by the manual identification,such as merging the over-segmented region,re-dividing the under-segmented region,which greatly improves the segmentation accuracy of the non-uniform and overlapping rockfill.The results show that the recognized particle sizes by the image processing are slightly larger than the measured sizes by manual labeling,but the relative error is less than 20%.The diameter of the identified outer circle is very close to the actual long axis of the rock,and the relative error is less than 20%.The mass of small rocks with the diameter of less than 300 mm accounts for about 2%,and the quality of rockfill is good.The grading curve of rockfill particle size obtained automatically can guide engineers to evaluate the quality of rockfill surface quantitatively.

关 键 词：堆石混凝土 图像处理 分水岭算法 人机交互 粒径分布 

分 类 号：TV43[水利工程—水工结构工程]