62 was obtained with the XPL-10X model, showing lower total loss and precision of up to 35% for large objects, whereas PPL-10X yielded the poorest results. Visual evaluation reveals segmentation and overlap issues; however, good performance is observed in gabbros. Although these results were assessed based on model metrics, Vasilev (2019) notes that a common criticism of neural network systems is the interpretation of their outputs. Such models are often regarded as black boxes with logic that is difficult to interpret, leading to reliance on algorithms that are not fully understood and making improvements challenging. Enhancing the explainability of AI models is therefore an important step toward enabling users to understand and trust model outputs. 5.3. The scikit-image image processing library The comparison between measurements obtained from XPL images and their duplicated labels in PPL images shows a high level of accuracy (Figure 7), with better performance in linear measurements than in areal ones. However, the issue is not limited to the precision of the tool. Minerals occupy a threedimensional space, and their representation in two dimensions can introduce significant bias. The main limitation is truncation caused by cropping of the original image, which may lead to minerals being classified as fine-grained when they are in fact medium- or coarse-grained. This effect impacts both crystal size distributions and accurate mineral identification. 5.4. Unified petrographic analysis model The extraction of information from thin sections with high accuracy and precision cannot currently be achieved using a single model. Advanced analyses that involve interpretation and decision-making require multiple interconnected models with capabilities comparable to human vision (Tyagi, 2018). Liu et al. (2022) propose a composite architecture for petrographic analysis of sandstones (Figure 8), which is potentially applicable to igneous and metamorphic rocks and integrates artificial intelligence models with other algorithms. This approach handles high structural complexity and large data volumes, including 210000 images and more than 700000 labels across 12 categories. However, implementing such systems is complex, and challenges remain, including maintenance and update difficulties, costs, and the evaluation of model stability, reliability, and performance. Figure 8 – Petrographic analysis model for sandstones modified from Liu et al. (2022). (*) Indicates that Mask R-CNN outputs were used for Sorting and Particle contacts analysis. 6. WEB APPLICATION The application named AI Mineral Tech is publicly available under two development approaches (Figure 9): a Google Colab notebook, which serves as a functional prototype for code testing and model tuning, and a web application (Hugging Face-Streamlit), which enables deployment of the model and
RkJQdWJsaXNoZXIy MTM0Mzk2