Scholarlink

Abstract <p>This study demonstrates the efficacy of the transfer learning method for a convolutional neural network in addressing the inverse problem of photoluminescence spectroscopy for multicomponent analysis. A convolutional neural network model, pretrained to solve a 6-parametric inverse problem (quantifying concentrations of Cu<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\({}^{2+}\)</EquationSource> <!--BPhysMGU2570297Chugreeva-m1--> </InlineEquation>, Ni<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\({}^{2+}\)</EquationSource> <!--BPhysMGU2570297Chugreeva-m2--> </InlineEquation>, Co<InlineEquation ID="IEq3"> <EquationSource Format="TEX">\({}^{2+}\)</EquationSource> <!--BPhysMGU2570297Chugreeva-m3--> </InlineEquation>, Al<InlineEquation ID="IEq4"> <EquationSource Format="TEX">\({}^{3+}\)</EquationSource> <!--BPhysMGU2570297Chugreeva-m4--> </InlineEquation>, Cr<InlineEquation ID="IEq5"> <EquationSource Format="TEX">\({}^{3+}\)</EquationSource> <!--BPhysMGU2570297Chugreeva-m5--> </InlineEquation> and NO<InlineEquation ID="IEq6"> <EquationSource Format="TEX">\({}^{-}_{3}\)</EquationSource> <!--BPhysMGU2570297Chugreeva-m6--> </InlineEquation>), was successfully fine-tuned to solve an extended 7-parametric problem incorporating Pb<InlineEquation ID="IEq7"> <EquationSource Format="TEX">\({}^{2+}\)</EquationSource> <!--BPhysMGU2570297Chugreeva-m7--> </InlineEquation>. The transfer learning method significantly enhanced the prediction quality for all target cation concentrations. It was found that fine-tuning the model on a reduced dataset, but containing a new ion, leads to a lower error in its determination compared to using the full dataset. This methodology proves effective not only for improving the resolution of the inverse problem but also for substantially reducing the requisite volume of experimental data, thereby decreasing the associated costs and development time for optical nanosensors.</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

Application of the Transfer Learning Method for Convolutional Neural Network to Improve the Quality of Solving the Inverse Problem of Photoluminescent Nanosensors

  • 期刊论文

  • 发表日期：

    2025年12月1日

查看全文
  • G. N. Chugreeva,
  • K. A. Laptinskiy,
  • T. A. Dolenko

摘要

Abstract

This study demonstrates the efficacy of the transfer learning method for a convolutional neural network in addressing the inverse problem of photoluminescence spectroscopy for multicomponent analysis. A convolutional neural network model, pretrained to solve a 6-parametric inverse problem (quantifying concentrations of Cu \({}^{2+}\) , Ni \({}^{2+}\) , Co \({}^{2+}\) , Al \({}^{3+}\) , Cr \({}^{3+}\) and NO \({}^{-}_{3}\) ), was successfully fine-tuned to solve an extended 7-parametric problem incorporating Pb \({}^{2+}\) . The transfer learning method significantly enhanced the prediction quality for all target cation concentrations. It was found that fine-tuning the model on a reduced dataset, but containing a new ion, leads to a lower error in its determination compared to using the full dataset. This methodology proves effective not only for improving the resolution of the inverse problem but also for substantially reducing the requisite volume of experimental data, thereby decreasing the associated costs and development time for optical nanosensors.