Identification of multi-concentration aromatic fragrances with electronic nose technology using a support vector...

Area of Science:

• Chemometrics
• Artificial Intelligence
• Sensor Technology

Background:

• Electronic nose systems face challenges in distinguishing odors due to varying concentrations.
• Artificial intelligence offers potential solutions for enhancing electronic nose performance.

Purpose of the Study:

• To investigate the effectiveness of support vector machine (SVM) technology for classifying different aroma samples at multiple concentrations.
• To address the concentration effect that hinders accurate odor identification in electronic noses.

Main Methods:

• Utilized an 11-sensor electronic nose to collect responses from four fragrance samples (roman chamomile, jasmine, lavender, orange) at three concentrations.
• Applied data preprocessing techniques including baseline correction, smoothing, and sensor selection, reducing data to 8 sensors.
• Employed normalized maximum signal intensity as a feature for SVM analysis, alongside Principal Component Analysis (PCA).

Main Results:

• Using raw signal intensities resulted in poor classification accuracy (<50%) due to the concentration effect.
• Normalizing signal intensity significantly improved the model's accuracy by mitigating the concentration effect.
• Feature reduction and normalization enhanced sensor cross-interference reduction and improved odor resolution.

Conclusions:

• Normalized features effectively eliminate the concentration effect in electronic nose data.
• Support vector machine (SVM) with normalized data and reduced sensor sets achieved high accuracy in classifying complex aroma mixtures.
• Feature engineering and selection are crucial for enhancing the performance and reliability of electronic nose systems.