A new kernel discriminant analysis framework for electronic nose recognition

Area of Science:

• Chemical Sensing
• Machine Learning
• Data Science

Background:

• Metal oxide semiconductor gas sensor arrays are crucial for electronic nose (e-Nose) technology.
• Effective dimension reduction and pattern recognition are vital for accurate e-Nose gas detection.
• Existing methods face challenges in handling complex gas mixtures and achieving high recognition accuracy.

Purpose of the Study:

• To propose a novel Non-negative Discriminant Analysis (NDA) framework for dimension reduction and enhanced e-Nose recognition.
• To develop an effective Kernel PCA plus NDA (KNDA) method for rapid detection of gas mixture components.
• To evaluate the performance of the KNDA method against state-of-the-art e-Nose classification techniques.

Main Methods:

• Development of a new NDA framework utilizing between-class and within-class Laplacian scatter matrices.
• Integration of Kernel PCA with NDA (KNDA) to leverage high-dimensional kernel mapping space and Principal Component Analysis (PCA) for dimension reduction.
• Implementation of KNDA for both training and recognition processes in e-Nose data analysis.

Main Results:

• The proposed KNDA method demonstrated superior performance on e-Nose datasets for six different gas components.
• Achieved an average recognition rate of 94.14% and a total recognition rate of 95.06%.
• Outperformed existing state-of-the-art e-Nose classification methods in experimental comparisons.

Conclusions:

• The KNDA method offers a promising approach for feature extraction and multi-class recognition in e-Nose applications.
• The developed NDA framework provides an effective strategy for dimension reduction in sensor data.
• KNDA shows significant potential for improving the accuracy and efficiency of gas mixture detection using e-Nose technology.