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

Introduction: Individual sensor systems have limitations in the complex task of classifying shredded tobacco. This study aims to overcome these limitations by developing a novel evolutionary algorithm-based feature fusion framework to enhance sensing accuracy. Methods: We fused data from three sensing modalities: GC-SAW, E-nose, and FTIR. A systematic comparison was conducted to determine the optimal fusion strategy. Seven dimensionality reduction methods were rigorously evaluated, leading to the selection of a genetic algorithm (GA) as the cornerstone for feature selection within our fusion framework. Results: Feature-level fusion was confirmed as the most effective strategy. The GA-based feature selection demonstrated exceptional performance, achieving a mean classification accuracy of 99.89% ± 0.79% across 50 independent test runs. This success stemmed from the algorithm's ability to intelligently distill the high-dimensional fused data into a compact, highly discriminative subset. Discussion/Conclusion: Our framework effectively balances information from the three sensing modalities to maximize their complementary strengths. This work confirms that evolutionary algorithm-based feature fusion is a powerful and robust method for unlocking the full potential of multi-sensor data, thereby significantly advancing the accuracy of complex plant material classification.