Broadband absorption spectrum processing method based on end-to-end deep learning networks.

Area of Science:

• Spectroscopy
• Artificial Intelligence
• Chemical Engineering

Background:

• Absorption spectrum analysis is crucial for applications like engine temperature measurement and material design.
• Current methods for processing absorption spectra are slow and complex, hindering real-time applications.
• Existing techniques involve multiple steps including filtering, denoising, baseline correction, and database comparison.

Purpose of the Study:

• To develop a rapid and accurate method for processing absorption spectra.
• To overcome the limitations of traditional, time-consuming spectral analysis techniques.
• To enable real-time online processing of absorption spectrum data.

Main Methods:

• A novel neural network architecture combining Long Short-Term Memory (LSTM) and Fully Convolutional Network (FCN) was designed.
• The neural network directly processes unprocessed absorption spectra without requiring pre-processing steps.
• The method was tested for temperature processing of H2O and CO2 absorption spectra.

Main Results:

• The proposed neural network method achieved a 98.4% accuracy rate for H2O and CO2 temperature processing.
• The average processing time per absorption spectrum was 0.00036 seconds, representing a significant speed improvement.
• The method demonstrated high accuracy and unprecedented processing speed for absorption spectra.

Conclusions:

• The developed neural network method offers a highly accurate and extremely fast solution for absorption spectrum analysis.
• This approach has significant potential for real-time online applications in various scientific and engineering fields.
• The combination of LSTM and FCN networks effectively addresses challenges in spectral feature extraction and processing accuracy.