摘要
土石料的级配直接影响着土石坝的质量与防渗,为解决传统人工筛分和目测超粒径剔除的低效率与随机性。采用基于MaskRCNN算法的图像识别技术,从标准球体的图像识别入手,研究了不同组别粒径下识别球体数量与真实数量间的关系,提出了将颗粒平面识别向空间体积扩展的椭球计算方法,建立了基于指数函数的图像识别与级配之间的转换方法。采用该方法应用于碎石料和砂石料的粒径与级配识别中,相关系数均提升10%以上,级配曲线所反映的C_(u),C_(c)准确度最高提升35.26%。该研究为土石料级配的图像识别提供了新方法。
The gradation of soil and stone materials directly affects the quality and anti-seepage of earth-rock dams.In order to solve the low efficiency and randomness of the traditional manual screening and the visual removal of oversized particles,the image recognition technology based on the MaskRCNN algorithm is used to start with the image recognition of standard spheres,and the relationship between the number of recognized spheres and the real number under different groups of particle sizes is studied.The ellipsoid calculation method that extends particle plane recognition to space volume is proposed,and the transformation method between the image recognition and the gradation based on the exponential function is established.By applying this method to the identification of particle size and the gradation of crushed stone and gravel,the correlation coefficient increases by more than 10%,and the accuracy of C_(u) and C_(c) reflected by the gradation curve is up to 35.26%.This stud may provide a new method for the image recognition of soil and stone gradation.
作者
刘禹杉
孙淼军
吴帅峰
张丽雅
孙黎明
LIU Yushan;SUN Miaojun;WU Shuaifeng;ZHANG Liya;SUN Liming(State Key Laboratory of Simulation and Regulation of Water Cycle in River Basin,China Institute of Water Resources and Hydropower Research,Beijing 100048,China;Power China Huadong Engineering Corporation,Hangzhou 311100,China)
出处
《岩土工程学报》
EI
CAS
CSCD
北大核心
2023年第S01期59-62,共4页
Chinese Journal of Geotechnical Engineering
基金
浙江省“尖兵”研发攻关计划项目(2022C03009)
中国电建集团重点科技项目(DJ-ZDXM-2019-3)
关键词
土石料
级配
图像识别
颗粒轮廓
空间转换
earth-rock
gradation
image recognition
particle profile
spatial conversion
