were in high agreement with the judgment results by the application. In addition, the model built on the application was compared with the conventional Python-based model building method, and no significant differences were found between the two results. These results suggest that the application-based learning and judgment process performs as well as conventional programmingbased methods, supporting its high practicality in the field. Figure 3 Rock classification results using hyperspectral images. (a) Prediction by trained model by the application. (b) Comparison with Python-based method. (c) Ground truth by visual inspection 3.4 Result of RQD measurement The RQD values obtained from the actual measurements were then compared with the RQD values calculated by detecting cracks using the application. The RQD measurements and the crack detection results are shown in Figure 4. The mean discrepancy between the RQD measurements and the application-generated measurements was 11.8, as presented in Table 1. When this RQD measurement error was applied to the Rock Mass Rating (RMR) rock mass rating index, the difference in scores averaged 3 points. Consequently, despite the presence of a certain degree of error, the application-based evaluation method was found to be significantly more time-efficient than the conventional approach. These findings imply that automation through application may enhance the efficiency and expediency of RQD evaluation, thereby underscoring its potential as a practical rock quality evaluation tool in fieldwork.
RkJQdWJsaXNoZXIy MTM0Mzk2