Evolutionary algorithm-optimized feature fusion for accurate classification of shredded tobacco using multi-sensor...

Area of Science:

• Analytical Chemistry
• Computational Biology
• Sensor Technology

Background:

• Individual sensor systems exhibit limitations in accurately classifying complex materials like shredded tobacco.
• Multi-sensor data fusion offers potential to overcome these limitations by integrating diverse data streams.

Purpose of the Study:

• To develop a novel evolutionary algorithm-based feature fusion framework to improve sensing accuracy for shredded tobacco classification.
• To overcome the inherent limitations of individual sensor systems in complex classification tasks.

Main Methods:

• Data from three sensing modalities (GC-SAW, E-nose, FTIR) were fused.
• Feature-level fusion was identified as the optimal strategy.
• A genetic algorithm (GA) was employed for feature selection within the fusion framework after evaluating seven dimensionality reduction methods.

Main Results:

• The GA-based feature selection achieved a mean classification accuracy of 99.89% ± 0.79% over 50 test runs.
• Feature-level fusion proved to be the most effective strategy.
• The framework successfully distilled high-dimensional fused data into a discriminative subset.

Conclusions:

• The developed framework effectively balances complementary strengths from multiple sensing modalities.
• Evolutionary algorithm-based feature fusion is a powerful method for maximizing multi-sensor data potential.
• This approach significantly advances the accuracy of complex plant material classification.