Automated Geological and Structural Mapping of Open Pit Mines using Unmanned Aerial Vehicle (UAV) Systems and Machine Learning

对暴露在坑壁上的地质特征（如岩性单元、蚀变带、断层）进行准确的描述和绘图，对于矿山规划和设计至关重要。因此，地质和结构图及模型需要通过获取和整合新的地质信息，在整个矿山生命周期内频繁更新。这些地质和结构模型的准确性对于短期和长期矿山规划以及矿坑边坡的地质力学稳定性至关重要。传统的地质和结构数据采集和分析通常是通过直接目视检查和测量暴露的岩石表面来手动执行的。这个过程是耗时的、劳动密集型的，并且可能是不安全的，因为采矿人员暴露于危险条件，包括落石和移动设备。该研究项目旨在使用遥感技术和机器学习算法来自动绘制露天矿的地质特征。配备高光谱和RGB相机的无人机系统将用于测量两个露天矿（Kinross Gold的Bald Mountain和麦克尤恩Mining的Gold Bar）的大型坑壁。使用无人机进行数据采集可以从矿山人员无法到达的区域收集矿井壁的高分辨率图像。收集的高光谱和RGB图像将用于训练监督机器学习模型，以检测、分类和绘制坑壁上的目标地质特征。此外，所收集的数据将用于通过建立高光谱数据与岩石力学性质之间的相关性来预测坑壁上暴露的岩石单元的力学性质。该项目的结果可以改善地质和结构坑壁测绘的发展和一致性，这对采矿作业具有重大的经济和安全影响。

An accurate characterization and mapping of geological features (e.g. lithological units, alteration zones, faults) exposed on pit walls, is critical for mine planning and design. As a result, geological and structural maps and models require frequent updates throughout the life of mine by acquiring and integrating new geological information. The accuracy of these geological and structural models is essential for short-term and long-term mine planning as well as for geomechanical stability of pit slopes. Conventional geological and structural data acquisition and analysis is usually performed manually by direct visual inspection and measurement of exposed rock surfaces. This process is time consuming, labour-intensive, and may be unsafe as mine personnel are exposed to hazardous conditions including falling rocks and moving equipment. This research project aims to use remote sensing techniques together with machine learning algorithms to automate mapping of geological features in open pit mines. Unmanned Aerial Vehicle (UAV) systems equipped with hyperspectral and RGB cameras will be used for surveying large pit walls in two open pit mines (Kinross Gold's Bald Mountain and McEwen Mining's Gold Bar). Using UAV for data acquisition allows collecting high resolution images of the pit walls from areas inaccessible to mine personnel. The collected hyperspectral and RGB images will be used to train supervised machine learning models for detecting, classifying and mapping target geological features on the pit walls. In addition, the collected data will be used to predict mechanical properties of rock units exposed on the pit walls by developing correlations between hyperspectral data and rock mechanical properties. The results of this project can improve the development and consistency of geological and structural pit wall mapping which can have significant economic and safety impacts on mining operations.