Light fuel classification based on Raman spectroscopy and region-adaptive convolutional neural networks

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Artificial Intelligence in Chemistry

Background:

• Modern warfare demands robust energy security for combat effectiveness and strategic success.
• Field-adaptable, rapid, and accurate detection of light fuels is crucial for supply chain integrity.
• Portable Raman spectroscopy is promising for light fuel analysis but faces challenges with spectral similarity and low signal-to-noise ratios.

Purpose of the Study:

• To develop a novel, highly accurate, and efficient method for classifying light fuels using Raman spectroscopy.
• To overcome limitations of traditional feature extraction in Raman spectroscopy for complex fuel mixtures.
• To provide a practical solution for rapid, on-site identification of fuels in demanding environments.

Main Methods:

• Collected Raman spectra from 16 different grades and sources of light fuels (gasoline, diesel, jet fuel).
• Proposed a novel one-dimensional convolutional neural network (1D-CNN) model for fuel classification.
• Implemented region-specific feature extraction with adaptive weighting to enhance discriminative ability, focusing on spectral peak importance.

Main Results:

• The 1D-CNN model achieved high classification performance: 96.16% average accuracy, 96.23% precision, 96.08% recall, and 96.12% F1 score.
• The model demonstrated superior performance compared to other methods, with low parameter complexity (0.7 M) and a short training time (226.32 s).
• Sensitivity analysis confirmed the optimality of the selected model parameters for robust fuel identification.

Conclusions:

• The proposed region-aware 1D-CNN offers a novel and practical solution for rapid battlefield identification of light fuels.
• The method effectively addresses challenges of spectral similarity and low signal-to-noise ratios in Raman spectroscopy.
• This work provides valuable insights into applying deep learning models for fine-grained spectral analysis and classification.