Scholarlink

<p>Discontinuity traces are important parameters for characterizing the structure of rock mass. In order to solve the problems such as sample imbalance, high feature dimensionality, and poor recognition performance in discontinuity trace recognition using machine learning model, a new method based was proposed. Firstly, the slope point cloud model was established using the unmanned aerial vehicle (UAV) multi-angle photography method. Secondly, the point normal vector and roughness are selected as input features. Then, CloudCompare was used to select the training samples manually. Based on the SMOTE technique, the imbalanced ratio was optimized to 1:1 to improve the type performance. Finally, a new CNN classification model is developed to recognize discontinuity traces. The new method was applied to a rock slope along the Nujiang (NJ) River for verification. The research results show that the new method can accurately recognize discontinuity traces from 3D point clouds. In addition, the new method was compared and analyzed with three machine learning models: SVM, DT and ELM. The experimental results show that CNN model achieved an F1 score of 0.92 with strong generalization ability. The new method provides technical support for the application of artificial intelligence in rock masses.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Recognition of rock discontinuity trace based on UAV photogrammetry and a convolutional neural network

  • 期刊论文

  • 发表日期：

    2026年2月24日

查看全文
  • Mingzhe Zhou,
  • Haiying Fu,
  • Yufan Li,
  • Li Kong,
  • Yaoxian Xie,
  • Yanyan Zhao

摘要

Discontinuity traces are important parameters for characterizing the structure of rock mass. In order to solve the problems such as sample imbalance, high feature dimensionality, and poor recognition performance in discontinuity trace recognition using machine learning model, a new method based was proposed. Firstly, the slope point cloud model was established using the unmanned aerial vehicle (UAV) multi-angle photography method. Secondly, the point normal vector and roughness are selected as input features. Then, CloudCompare was used to select the training samples manually. Based on the SMOTE technique, the imbalanced ratio was optimized to 1:1 to improve the type performance. Finally, a new CNN classification model is developed to recognize discontinuity traces. The new method was applied to a rock slope along the Nujiang (NJ) River for verification. The research results show that the new method can accurately recognize discontinuity traces from 3D point clouds. In addition, the new method was compared and analyzed with three machine learning models: SVM, DT and ELM. The experimental results show that CNN model achieved an F1 score of 0.92 with strong generalization ability. The new method provides technical support for the application of artificial intelligence in rock masses.