| Abstrakt: |
With the popularization of unmanned aerial vehicles, low-cost, high-precision photogrammetry has become a useful tool for mapping and obtaining information in areas that are difficult to access. It has been applied in geomechanics because it makes it possible to obtain information from rock masses for which there is no alternative way to obtain information by traditional mapping methods, increases the frequency of analysis and reduces the exposure of professionals to unsafe conditions. In this study, we used the technique called structure from motion to acquire 2D images and create a three-dimensional point cloud. Then, using artificial neural network techniques, we implemented a semiautomatic classification of the rock mass using the Geological Strength Index (GSI). The routine was validated using five datasets with different geological characteristics, choosing the neural network with the best performance, presenting results with confidence intervals above 90%, 100% hit rates, and a low mean squared error. The procedure allows a standardized interpretation, intended to reduce the bias generated by the interpretation through a rapid and repeatable analysis, in addition to the creation of a systematic record report. [ABSTRACT FROM AUTHOR] |
