With the advancement of tools that use machine learning and deep learning to classify minerals and rocks with acceptable accuracy, these limitations can be addressed or minimized. An automated web application designed to identify and characterize different rock types using artificial intelligence was proposed by Okada (Okada et al., 2020). This application was tested using thin section images and reported an accuracy of 89.2%. Another application, APiS (Okada et al., 2025), is MATLAB-based and allows users to interactively create classification models and apply them to unknown images. However, these applications only support pre-trained models, meaning their performance and accuracy heavily depend on the datasets used during training. Users are not able to flexibly modify these models, which may result in a mismatch between the model’s functionality and the user's specific objectives. Additionally, since MATLAB-based applications require a paid license, this can become a barrier to broader adoption. In this study, we developed a new application called IROMIE (a blend of Japanese words "iro" for color and "mie" for appearance) which features an intuitive graphical user interface (GUI) designed to enhance accessibility and usability in mineral processing. IROMIE retains the analytical capabilities of existing Python-based scripts while significantly improving ease of use, efficiency, and accessibility. A notable feature of IROMIE is that users can build models tailored to their specific objectives using their own datasets. This enables AI-based mineral processing to be applied to a wide variety of rock types without relying on predefined datasets. With simple operations, users can perform rock classification and hyperspectral data analysis, and the entire workflow—from model creation to prediction on unknown images—can be carried out without requiring programming expertise. Moreover, since the application is built on the open-source Python platform, it can be used without the need for license registration. Furthermore, IROMIE integrates functionality for measuring Rock Quality Designation (RQD), a key metric for evaluating rock mass quality. It is evident that RQD performs a pivotal function within the domain of civil engineering, particularly in the contexts of tunnel excavation and foundation design (Zhang, 2016), (Narimani et al., 2025). Recent studies have demonstrated that integrating lithological identification with structural evaluations such as RQD can significantly enhance the reliability and efficiency of geological interpretation (Xu et al., 2023). The IROMIE system, developed in this study, represents a notable advancement in this field. In addition to its classification capabilities, the apparatus incorporates functions for determining the type of rock and the degree of weathering. Collectively, these enhancements demonstrate the feasibility and effectiveness of a more advanced and comprehensive geological evaluation system that leverages deep learning and spectral analysis. In this paper, we first provide a detailed description of the methodology employed in this study. Firstly, a comprehensive account of the project development environment is presented, followed by a flowchart showing how the Convolutional Neural Network (CNN) model was built and how Rock Quality Designation (RQD) was measured. The algorithms utilized in the application are then described in detail, and the differences between using the application and training a CNN on the same dataset using a Python script are compared and verified. Regarding the calculation of RQD, we will also compare the results of the application calculations with the actual measurements and evaluate the impact of the results on the final RMR score.
RkJQdWJsaXNoZXIy MTM0Mzk2