基于SAM&ImageJ图像处理的堆石混凝土坝层面露石率研究  

Research on exposed rockfill proportion of RFC surface based on SAM and ImageJ image processing

作　　者：安宇 徐小蓉 尹志刚[1] 金峰[3] 张喜喜 AN Yu;XU Xiaorong;YIN Zhigang;JIN Feng;ZHANG Xixi(School of Water Conservancy&Environmental Engineering,Changchun Institute of Technology,Changchun 130012,China;School of Water Resources&Hydropower Engineering,North China Electric Power University,Beijing 102206,China;State Key Laboratory of Hydroscience and Engineering,Tsinghua University,Beijing 100084,China;Sichuan Cim Info-Tech Co.Ltd.,Meishan 620599,China)

机构地区：[1]长春工程学院水利与环境工程学院,吉林长春130012 [2]华北电力大学水利与水电工程学院,北京102206 [3]清华大学水圈科学与水利工程全国重点实验室,北京100084 [4]四川西沐建信科技有限公司,四川眉山620599

出　　处：《水资源与水工程学报》2024年第1期154-161,共8页Journal of Water Resources and Water Engineering

基　　金：国家自然科学基金重点项目(52039005);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项目(sklhse-2022-C-03)。

摘　　要：堆石混凝土坝层面的外露块石为上下层提供了重要的啮合作用,其投影面积比例是科学评价层间抗剪性能的重要指标。采用国际最新Meta AI模型segment anything model(SAM)对层面外露堆石进行自动图像分割,并基于ImageJ软件对SAM识别后的图片进行再加工与图像计算,利用平滑、差分算法、中值滤波等方法精准标定外露堆石,二值化后计算得到层面露石率。结果表明:SAM图像预分割可识别约90%的外露堆石,经过ImageJ二次图像处理后可有效提高小粒径堆石的识别精度,对比手动标注结果误差在±3%以内。以贵州省两座水库的工程应用为例,对浇筑仓面进行分区预处理,结果发现靠近上游、中部、下游不同区域的露石率差别较大,计算得到的层面露石率以10%~30%居多,其中堆石入仓运输通道区域的露石率较低。研究内容与结论可为堆石混凝土结构层间界面抗剪力学性能和大坝蓄水安全稳定的研究提供参考与借鉴。The exposed rockfill on the lift surface of rock-filled concrete(RFC)dam increase shear resistance at the interface between upper and lower layers,which is crucial to the stability of the dam,and the projected area proportion of the exposed rockfill is an important index for the scientific evaluation of the interlayer shear performance.In this study,the latest international Meta AI model,known as segment anything model(SAM),was utilized for automatic image segmentation of RFC exposed rockfill.The SAM-identified images were further reprocessed and analyzed by ImageJ,which involved techniques such as smoothing,differential algorithm,and median filtering for the accurate location of the exposed rockfill.The binarized images were then used to calculate the exposed rockfill proportion.The results show that SAM image pre-segmentation can identify about 90%of the exposed rockfill,and the secondary image processing by ImageJ can effectively improve the identification accuracy of small rocks,within an error of±3%compared to manual annotation results.Then,this methodology is applied to two reservoir projects in Guizhou Province,each lift surface was pre-processed into different zones.We found that the exposed rockfill proportion near the upper,middle and lower reaches are quite different,mostly falls in the range of 10%-30%,among which the exposed rockfill proportion in the transport area is quite low.The research results and findings can provide some reference for the study of interfacial shear performance,as well as the safety and stability of dam reservoirs.

关 键 词：堆石混凝土坝 segment anything model(SAM) 图像处理技术 露石率 层间抗剪性能 

分 类 号：TV642[水利工程—水利水电工程]