Scholarlink

<p>X-Min Learn is an open-source software for mineral recognition in natural and artificial stone materials from X-ray map data, automated through machine learning (ML) and image analysis. The most innovative software section is the Developer’s Toolkit (DT), which is useful for building up ML models in a user-friendly, no-code environment. In this first version of the software, the DT has been equipped with a fully trainable SoftMax Regressor algorithm, paired with a Stochastic Gradient Descent optimizer. Two additional operational sections are included: 1) Mineral Classifier tool, for achieving supervised or unsupervised mineral classifications based on three classifier types: (a) pre-trained, for user-developed custom models; (b) ROI-based, for supervised algorithms that rely on user-drawn regions of interest (ROIs); and (c) unsupervised, for clustering algorithms; 2) Phase Refiner tool, for removing noisy pixels from classification results through (a) a basic mode, that applies a maximum frequency filter to smooth the entire image; and (b) an advanced mode, which allows class-by-class refinements using morphological image processing. Classified mineral maps, if validated, can be employed in the DT for updating existent ML models or developing new ones tailored for the characterization of samples acquired with specific instrument (e.g., SEM, EPMA) at precise operational conditions (e.g., current, dwell time, pixel size). X-Min Learn’s interactive widgets also allow processing standard image files to populate pixel-wise mineral ground truth datasets. This makes it a versatile and autonomous tool for developing custom ML models from mineral X-ray maps data.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

X-Min Learn: machine learning-powered mineral identification from X-ray maps data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月22日

查看全文
  • Alberto D’Agostino,
  • Gaetano Ortolano,
  • Roberto Visalli,
  • Michele Zucali,
  • Rosolino Cirrincione

摘要

X-Min Learn is an open-source software for mineral recognition in natural and artificial stone materials from X-ray map data, automated through machine learning (ML) and image analysis. The most innovative software section is the Developer’s Toolkit (DT), which is useful for building up ML models in a user-friendly, no-code environment. In this first version of the software, the DT has been equipped with a fully trainable SoftMax Regressor algorithm, paired with a Stochastic Gradient Descent optimizer. Two additional operational sections are included: 1) Mineral Classifier tool, for achieving supervised or unsupervised mineral classifications based on three classifier types: (a) pre-trained, for user-developed custom models; (b) ROI-based, for supervised algorithms that rely on user-drawn regions of interest (ROIs); and (c) unsupervised, for clustering algorithms; 2) Phase Refiner tool, for removing noisy pixels from classification results through (a) a basic mode, that applies a maximum frequency filter to smooth the entire image; and (b) an advanced mode, which allows class-by-class refinements using morphological image processing. Classified mineral maps, if validated, can be employed in the DT for updating existent ML models or developing new ones tailored for the characterization of samples acquired with specific instrument (e.g., SEM, EPMA) at precise operational conditions (e.g., current, dwell time, pixel size). X-Min Learn’s interactive widgets also allow processing standard image files to populate pixel-wise mineral ground truth datasets. This makes it a versatile and autonomous tool for developing custom ML models from mineral X-ray maps data.