结合Mask R-CNN和SAM获取堆石混凝土坝堆石级配曲线  

Obtaining Particle Size Distribution Curves for Rock-Filled Concrete Dams by Combining Mask R-CNN and SAM

作　　者：付立群 金峰[1] 张喜喜 杨家琦 周虎[1] FU Li-qun;JIN Feng;ZHANG Xi-xi;YANG Jia-qi;ZHOU Hu(State Key Laboratory of Hydroscience and Engineering,Tsinghua University,Beijing 100084,China;Sichuan CimInfoTech Co.,Ltd.,Meishan 620599,China)

机构地区：[1]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 [2]四川西沐建信科技有限公司,四川眉山620599

出　　处：《水电能源科学》2024年第11期7-11,130,共6页Water Resources and Power

基　　金：国家自然科学基金重点项目(52039005)。

摘　　要：堆石混凝土坝施工过程质量控制的关键是浇筑密实度,其不仅决定于自密实混凝土工作性能,也与堆石体中的空隙性质密切相关。快速获取堆石粒径级配可以保障浇筑质量的同时,有利于进一步降低成本。现场的堆石数量巨大,依靠手工测量块石粒度不切实际,依靠图像识别技术快速获取堆石粒径是可行的办法。利用MaskR-CNN算法和大模型Segment Anything算法,设计了一套完整的、可操作的获取现场堆石粒径级配曲线的方法。该级配计算方法利用了MaskR-CNN算法和Segment Anything算法各自的优势,提高了堆石料粒径识别结果的准确性,拓展了堆石粒径识别的应用场景。将该方法应用于某堆石混凝土拱坝施工现场,取得了良好效果。The construction quality control of rock-filled concrete dams is heavily dependent on pouring compactness,which is influenced by the workability of self-compacting concrete and the voids between rocks.The particle size distribu-tion(PSD)must be rapidly obtained to ensure compactness and reduce costs.Since manual measurement of particle size is impractical due to the large volume of rocks,this study presents an algorithm combining Mask R-CNN,and the Seg-ment Anything Model to determine the PSD of rocks through image recognition.The advantages of both Mask R-CNN and the Segment Anything Model are utilized,improving the accuracy of particle recognition and expanding the application scenarios for PSD identification.The method has been applied to the construction site of a rock-filled concrete arch dam,yielding positive results.

关 键 词：堆石混凝土坝 SAM 神经网络算法 图像处理技术 级配曲线(PSD) 

分 类 号：TV642.4[水利工程—水利水电工程]