岩石种类和结构面几何参数是岩体质量和稳定性评价的两个核心要素，在“智能建造”的大趋势下，如何实现岩体地质参数解译信息化和智能化是岩体工程领域的热点和难点。本项目综合采用现场调查、室内试验、理论分析和程序研发等方法，开展基于无人机和深度学习的岩石智能识别和结构面定量表征研究。通过搭建可扩充的区域性岩石样本数据集和深度迁移学习算法，构建岩石种类智能识别模型，基于无人机采集的岩体图片、岩石识别模型和图像分割算法，实现实际岩体不同区域的岩石种类识别与边界标定；针对传统方法无法处理无人机海量点云数据的问题，通过研究基于GPU的点云法向量并行计算方法，实现约10~1000倍的加速比。同时，探寻以法向量为核心输入参数的结构面识别算法，可以实现海量点云的高效率、轻参数识别。最终形成一套从岩体数据采集到地质信息解译的智能方法，研究成果可为实现岩体质量和安全评价的数字化、智能化提供有益的理论和技术支撑。

Rock types and discontinuity geometry parameters are two key factors for rock mass quality and stability evaluation. Under the general trend of "intelligent construction", how to realize informatization and intelligentization of rock mass geological parameter interpretation is a hot and difficult research topic in rock engineering field. In this project, a comprehensive method combining field investigation, laboratory test and development, theoretical analysis and procedure development will be adopted to study the intelligent identification of rocks and quantitative characterization of discontinuity based on unmanned aerial vehicle (UAV) and deep learning. An intelligent identification model of rock will be trained using a specific designed deep transfer learning algorithm and an expandable data set of regional rock samples. Moreover, the rock type identification and boundary demarcation for rock mass will be realized based on the rock image collected by UAV, rock identification model and image segmentation algorithm. Aiming at the problem that the traditional method cannot deal with the massive point cloud data generated by UAV, the parallel computing method for the calculation of normal vector based on GPU will be studied to achieve an acceleration ratio of about 10~1000 times. At the same time, the algorithm of discontinuity recognition based on normal vector will be explored, which can realize the high efficiency and light parameter recognition of massive point cloud. Eventually, a complete set of methods including rock mass data collection and geological information interpretation will be formed. The research results can provide theoretical and technical support for digitization and intelligentization of rock mass quality and safety evaluation.