Research on Binary Mixed VOCs Gas Identification Method Based on Multi-Task Learning.

Area of Science:

• Environmental Science
• Analytical Chemistry
• Machine Learning

Background:

• Traditional volatile organic compounds (VOCs) detection methods often involve separate component identification and concentration prediction steps.
• This separation leads to suboptimal feature utilization and challenges in learning from limited datasets.

Purpose of the Study:

• To develop an advanced model for simultaneous VOCs component identification and concentration prediction.
• To improve feature extraction and task synergy in VOCs detection using multi-task learning.

Main Methods:

• Implementation of a Residual Fusion Network based on multi-task learning (MTL-RCANet).
• Integration of channel attention mechanisms and cross-fusion modules for enhanced feature extraction.
• Utilization of a dynamic weighted loss function to balance task training progress.

Main Results:

• The MTL-RCANet achieved a high accuracy of 94.86% and an R² score of 0.95.
• Excellent identification performance was observed even when using only 35% of the total input data length.
• Multi-task learning demonstrated superior model efficiency by effectively integrating feature information across tasks.

Conclusions:

• The proposed MTL-RCANet significantly improves VOCs detection performance, especially in small-sample scenarios.
• The dynamic weighted loss function and integrated modules enhance model robustness and feature utilization.
• This approach offers a more efficient and effective alternative to traditional single-task learning models for VOCs analysis.