深度图像引导的岩石颗粒分割方法  

A rock particles segmentation method guided by depth images

作　　者：严良平 潘月梁 姜雄彪 陆秋雨 徐畅 YAN Liangping;PAN Yueliang;JIANG Xiongbiao;LU Qiuyu;XU Chang(Xinjiang Fukang Pumped Storage Co.,Ltd.,Urumqi 830032,China;Zhejiang Ninghai Pumped Storage Co.,Ltd.,Ningbo 315612,China;College of Internet of Things Engineering,Hohai University,Changzhou 213022,China)

机构地区：[1]新疆阜康抽水蓄能有限公司,新疆乌鲁木齐830032 [2]浙江宁海抽水蓄能有限公司,浙江宁波315612 [3]河海大学物联网工程学院,江苏常州213022

出　　处：《应用科技》2022年第2期87-93,共7页Applied Science and Technology

基　　金：江苏省重点研发计划项目(BE2020649);国网科技项目(820038416)。

摘　　要：堆石料的形状、尺寸以及级配等参数直接影响堆石坝的稳定性和抗渗性能。当前主要采用人工的方式筛分堆石料,效率低下;而传统的图像分割算法分割精度低,无法准确测量岩石的参数信息。针对这一问题,本文提出一种深度图像引导的岩石颗粒分割方法。首先对深度图像进行预处理,去除深度图像的噪声;然后提取深度图像与可见光图像的随机特征和显著性特征,并对随机特征进行多次抽样;最后根据多组随机特征和显著性特征得到多个分割预测结果,并选择最优的分割。实验结果表明,本文方法能够实现岩石颗粒的准确分割,并将其应用到岩石颗粒度评估的场景中,计算岩石颗粒参数信息。The shape, size and gradation of rockfill materials are main factors affecting the stability and impermeability of dams. Currently, most rockfill materials are screened manually, which is inefficient. Traditional image segmentation algorithms often lead to the problem of wrong segmentation, which also results in unsatisfactory rockfill segmentation.To solve these problems, this paper proposes a method of rock particles segmentation guided by depth images. Firstly,the noise of depth image is removed by preprocessing the depth image. Then, the stochastic and saliency features of the depth image and the visible image are extracted respectively, and the stochastic features are sampled several times.Finally, according to several groups of stochastic features and saliency features, segmentation prediction results are obtained, and the optimal rockfill segmentation is thus selected. Experimental results show that the proposed method can accurately segment rockfill materials, and calculate rock particle parameters for rockfill evaluation.

关 键 词：图像分割 深度图像 可见光图像 显著性检测 堆石料 随机特征 显著性特征 级配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]