Machine learning-based fusion of E-nose and E-eye for discrimination of chrysanthemum (Chrysanthemum morifolium) from different geographical origins

Chrysanthemums are a valuable medicinal plant widely cultivated in China. Chrysanthemums from different geographical origins exhibit variations in appearance and aroma, which influence their commercial value and consumer acceptance. Currently, there is a lack of rapid and effective methods for large-scale, on-site screening of chrysanthemum types. This study proposes the use of E-nose and E-eye technologies to respectively capture the appearance and aroma profiles of chrysanthemums. Linear, nonlinear, and ensemble learning algorithms were applied to sensor data modeling to differentiate chrysanthemum samples from five geographical origins. Although chrysanthemums from different regions showed variations in color variables of E-eye and metal oxide sensor signals of E-nose, these differences were insufficient for clear differentiation. Classification models based solely on individual sensor data failed to achieve accurate identification. However, data fusion of the E-eye and E-nose significantly improved the accuracy of discrimination. The middle-level fusion based on principal component analysis combined with the XGBoost model achieved the highest classification accuracy, achieving 94% accuracy on a testing set of 50 samples. Feature importance analysis further revealed that E-nose signals contributed more significantly to the classification than E-eye. This study provides a simple, fast, and reliable method for tracing the geographical origin of chrysanthemums, and offers a reference for the geographical traceability of other agricultural products. • E-eye and E-nose were jointly used to perceive information of chrysanthemum. • XGBOOT showed modeling advantages of data from different geographical origins. • Mid-level data fusion improved the accuracy of the classification model. • E-nose data outperformed e-eye in terms of contribution for classification.