东北大学学报(自然科学版) ›› 2024, Vol. 45 ›› Issue (1): 101-110.DOI: 10.12068/j.issn.1005-3026.2024.01.013

• 资源与土木工程 • 上一篇    下一篇

基于AttentionR2U-net的岩石(体)关键节理智能识别与参数提取

孙浩1,2, 代宗晟1,2, 金爱兵1,2, 陈岩3   

  1. 1.北京科技大学 金属矿山高效开采与安全教育部重点实验室,北京 100083
    2.北京科技大学 土木与资源工程学院,北京 100083
    3.北京科技大学 自动化学院,北京 100083
  • 收稿日期:2022-08-16 出版日期:2024-01-15 发布日期:2024-04-02
  • 作者简介:孙 浩(1992-),男,安徽阜阳人,北京科技大学副教授,博士
    金爱兵(1974-),男,江苏兴化人,北京科技大学教授,博士生导师.
  • 基金资助:
    国家自然科学基金资助项目(52174106);中央高校基本科研业务费专项资金资助项目(FRF-IDRY-20-021)

Intelligent Identification and Parameter Extraction of Key Joints in Rock (Mass) Based on AttentionR2U-net

Hao SUN1,2, Zong-sheng DAI1,2, Ai-bing JIN1,2, Yan CHEN3   

  1. 1.Key Laboratory of Ministry of Education for Efficient Mining and Safety of Metal Mines,University of Science and Technology Beijing,Beijing 100083,China
    2.School of Civil and Resource Engineering,University of Science and Technology Beijing,Beijing 100083,China
    3.School of Automation and Electrical Engineering,University of Science and Technology Beijing,Beijing 100083,China. Corresponding author: JIN Ai-bing,E-mail: jinaibing@ustb. edu. cn
  • Received:2022-08-16 Online:2024-01-15 Published:2024-04-02

摘要:

针对岩石(体)表面复杂节理网中关键节理的智能识别与参数提取问题,提出一种基于AttentionR2U-net网络与节理几何特征模型耦合识别的方法.在R2U-net网络的基础上引入注意门(attention gate)改进网络,通过定性与定量的方法对边坡节理图像和混凝土、龟裂土、常见脆性岩石裂隙图像的识别结果分别作准确性及泛化能力检验;利用AttentionR2U-net网络耦合节理几何特征的方法识别关键节理,提取原始节理和关键节理的几何参数并对其迹长、面积及倾角作差异性分析.研究结果表明:针对岩石(体)节理识别,本文算法的Dice相似系数从U-net网络的0.965提升至0.990,且明显优于传统算法,故本文算法在岩石(体)节理识别上具有更强的可靠性与优越性;针对混凝土、龟裂土和大理岩、花岗岩、砂岩等脆性岩石裂隙的识别,本文算法的Dice相似系数均在0.953以上,故本文算法具有较强的泛化能力.与原始节理网络相比,关键节理网络优势迹长由0.732 m显著增大至1.835 m,节理倾角分布形式和优势倾角组均不变,优势迹长和倾角的节理占比均显著增大.

关键词: 岩石(体), 关键节理, AttentionR2U-net网络, 智能识别, 参数提取

Abstract:

Aiming at the problem of intelligent identification and parameter extraction of key joints in complex joint networks on surfaces of rock (mass), a method based on AttentionR2U-net network and joint geometric feature model is proposed. The attention gate was introduced on the basis of R2U-net network to improve the network, and then the accuracy test and generalization ability study of the recognition results of slope joint images and fracture images in concrete, cracked soil and common brittle rocks were carried out by qualitative and quantitative methods. Finally, the AttentionR2U-net network was used to identify the key joints by coupling the geometric features of the joints. The geometric parameters of the original and key joints are extracted, and then the trace length, area and inclination angle are analyzed. The results show that for joint identification of rock (mass), the Dice similarity coefficient of the proposed algorithm is increased from 0.965 to 0.990 in U-net network, outperforming the traditional algorithm significantly. Therefore, the proposed algorithm has stronger reliability and superiority in joint identification of rock (mass). For crack identification in concrete, cracked soil and brittle rocks such as marble, granite and sandstone, the similarity coefficientDice of the proposed algorithm is above 0.953, so the proposed algorithm has strong generalization ability. Compared with the original joint network, the dominant trace length of the key joint network increased significantly from 0.732 m to 1.835 m, the distribution pattern of joint dip angle and the dominant dip angle group remained unchanged, and the proportion of joints with dominant trace length and dip angle increased significantly.

Key words: rock (mass), key joint, AttentionR2U-net network, intelligent recognition, parameter extraction

中图分类号: