Geographical origin traceability of chrysanthemum using hyperspectral imaging with class information-guided evolutionary multitasking wavelength selection and spatial feature extraction

• Published in: Journal of food composition and analysis 2025-03, Vol.139, Article 107107
• Main Authors: Wei, Yunpeng, Hu, Huiqiang, Yuan, Minghua, Xu, Huaxing, Mao, Xiaobo
• Format: Article
• Language: English
• Subjects: Chrysanthemum; Geographical origin traceability; Hyperspectral imaging; Particle swarm optimization; Spatial-spectral feature fusion
• ISSN: 0889-1575
• DOI: 10.1016/j.jfca.2024.107107

As a widely favored flower tea, chrysanthemum contains abundant bioactive components for health benefits, and its quality and nutritional value are influenced by the geographical origin directly. Hence, combining hyperspectral imaging (HSI) technique with chemometrics, an accurate approach was proposed to identify different geographical origins of chrysanthemums. First, to exploit the spectral feature sufficiently, this study developed a class information-guided multitasking particle swarm optimization (CIMPSO) framework. The class information criterion was used to evaluate the category uncertainty and assign the appropriate variable selection quantity for each category. Then the multi-class problem was divided into multiple binary classification subtasks to find the characteristic wavelengths for distinguishing each specific category and others simultaneously. Experimental results showed that the proposed CIMPSO achieved a superior accuracy of 94.16–96.02 % on multiple machine learning models, outperforming various wavelength selection methods. Besides, to enrich the feature information, this study also extracted the spatial texture and color features. With more comprehensive feature information, the fused spatial-spectral features reached the optimal accuracy on RF model, and the classification accuracy and F1-score achieved 98.86 % and 98.84 %, respectively. The results reveal that the spatial texture and color features also contain the useful information contributing to chrysanthemum identification. Meanwhile, combining the spectral and spatial information, the fused features can represent the geographic origin attribute more effectively. Overall, this study proposes a novel method for identifying chrysanthemum geographical origins by exploiting the key wavelengths and spatial information sufficiently, which can provide a valuable technical reference for the quality assessment nondestructively in food industry. Furthermore, it will also facilitate the HSI-related development of rapid and online quality detection equipment in agriculture. •Class information criterion was introduced to evaluate the class uncertainty.•Evolutionary multitasking key wavelength selection framework was constructed.•Spatial texture and color information were extracted to construct fused features.•The fused features improved the accuracy for identifying chrysanthemums further.