Highly-accurate solvent identification using dynamic evaporation reflection spectra from an inverse opal sensor...

Area of Science:

• Analytical Chemistry
• Materials Science
• Artificial Intelligence

Background:

• Accurate solvent identification is crucial, especially for structurally similar compounds.
• Existing spectroscopic methods often require complex sensor materials.
• Dynamic evaporative reflection spectra (DERS) offer potential for solvent analysis.

Purpose of the Study:

• To develop a low-cost, rapid, and accurate method for solvent detection using DERS.
• To apply deep learning for direct analysis of DERS data.
• To evaluate the performance of a deep learning model for solvent classification.

Main Methods:

• Fabrication of a simple inverse opal (IO) sensor using co-assembly and sacrificial template methods.
• Acquisition of 985 DERS datasets from 14 different solvents.
• Classification of DERS using classical machine learning and deep learning algorithms, including ResNet18-CBAM.

Main Results:

• The ResNet18-CBAM deep learning model achieved 97.7 ± 0.9% accuracy in 5-fold cross-validation.
• The model demonstrated 100% accuracy on the independent test set.
• The developed method proved effective in distinguishing between various solvents.

Conclusions:

• A deep learning approach enables automatic, rapid, and accurate solvent identification from DERS.
• The proposed method offers a simple, efficient, and repeatable solution for solvent detection.
• This strategy has significant potential for practical applications in solvent analysis.