Dual-stream feature-decoupling network for rapid and robust rice traceability using handheld Raman spectroscopy.

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Background:

• Rice classification and traceability are vital for food safety and quality management.
• Handheld Raman spectroscopy offers rapid, non-destructive on-site analysis.
• Portable Raman data often suffers from fluorescence interference, noise, and poor spectral utilization, limiting accuracy.

Purpose of the Study:

• To develop a robust method for portable Raman-based rice origin identification.
• To overcome limitations of fluorescence interference and spectral under-utilization in portable Raman instruments.

Main Methods:

• Proposed the Raman-Fluorescence Dual-stream Network (RFDNet) framework.
• Utilized wavelet decomposition and autoencoder for signal separation and refinement.
• Employed a dual-branch network (Raman and fluorescence) with attention and morphological convolution for feature extraction.

Main Results:

• RFDNet achieved 91% accuracy in rice origin identification.
• Demonstrated a 2.25% improvement over established baseline models.
• Effectively separated Raman signals from fluorescence and noise.

Conclusions:

• The dual-stream approach offers a robust and generalizable solution for on-site rice traceability.
• RFDNet effectively utilizes complementary information from Raman signals and fluorescence backgrounds.
• Highlights potential for reliable authentication and quality management of agricultural products.