Gas Sensor Drift Compensation Using Semi-Supervised Ensemble Classifiers with Multi-Level Features and Center Loss.

Area of Science:

• Sensor technology
• Artificial intelligence
• Machine learning

Background:

• Gas sensor drift significantly degrades electronic nose (E-nose) system performance.
• Conventional drift compensation methods struggle with complex data relationships and often require unrealistic labeled data for both drifted and non-drifted states.

Purpose of the Study:

• To develop an effective and reliable algorithm-level solution for gas sensor drift compensation in E-nose systems.
• To improve E-nose performance by accurately compensating for sensor drift using a semisupervised domain adaptive approach.

Main Methods:

• Proposed a semisupervised domain adaptive convolutional neural network (CNN) incorporating ensemble classifiers, multilevel feature extraction, pretraining, and center loss.
• Utilized Maximum Mean Discrepancy (MMD) in Hilbert space to evaluate domain similarity across feature levels and weighted ensemble predictions.
• Employed MMD as a pretraining loss for robust feature learning and center loss for focused intra-class feature representation.

Main Results:

• Achieved average classification accuracies of 76.06% (long-drift) and 82.07% (short-drift) on two datasets.
• Attained an average R-squared score of 0.804 in regression tasks, demonstrating significant improvements over conventional methods.
• Validated the effectiveness and reliability of the proposed method in addressing gas sensor drift compensation.

Conclusions:

• The proposed semisupervised domain adaptive CNN effectively compensates for gas sensor drift, significantly enhancing E-nose system performance.
• The method's ability to leverage multilevel features and domain adaptation techniques offers a robust solution for real-world E-nose applications.
• This work provides a valuable algorithmic advancement for tackling the persistent challenge of sensor drift in gas sensing technology.