A multisource fusion approach for roasting evaluation of Gardenia jasminoides

Abstract

To objectively and scientifically monitor the roasting degree of Gardenia jasminoides Ellis, a widely used herb in traditional Chinese medicine, this study employed a computer vision system and an electronic nose to analyse both the visual and olfactory characteristics of the roasted product. The roasting process of G. jasminoides comprises three distinct stages—early, middle, and late—each characterised by progressive changes in colour and aroma arising from complex chemical reactions. Initially, samples were categorised into these three stages based on empirical observations. Principal component analysis was then applied to reduce dimensionality and highlight discriminative features in both image and aroma data. Subsequently, a data fusion strategy was utilised to integrate visual and olfactory information. Machine learning algorithms, including random forest, K-nearest neighbours, and support vector machines (SVMs), were employed to construct classification models. Among these, the data-level fusion strategy combined with the SVM model yielded the highest classification accuracy, achieving 94.44±3.51% on the training set and 97.78% on the test set. These results demonstrate the feasibility and effectiveness of multi-sensor data fusion for accurately determining the roasting degree of G. jasminoides.