A Multimodal Tensor Data Fused Astragalus Slices Quality Grades Method

Astragalus, a popular traditional Chinese herbal medicine celebrated for its ability to tonify Qi and nourish blood, plays a crucial role in clinical applications and health maintenance. Current quality assessment methods for astragalus predominantly concentrate on individual evaluation metrics. However, the intrinsic polygenicity, heterogeneity, and complexity of astragalus present challenges in achieving a thorough and precise grading of its quality. To overcome this challenge, we have developed a novel method that is a Multimodal Astragalus tensor data fusion method for Quality Grading (MAQG). Compared to existing deep learning methods, MAQG demonstrates unique advantages in handling the complexity of traditional Chinese medicine tablets. Our approach innovatively incorporates a collaborative guidance mechanism to construct a comprehensive and precise multimodal fusion learning model. This model not only meticulously captures external features, internal constituents, and intricate inter-image correlations but also reveals the strong relationship between the appearance traits, intrinsic key quality indicators, and quality grades of traditional Chinese medicine tablets from a broader perspective. Meanwhile, MAQG effectively addresses the high-dimensional complexity in multimodal data fusion by introducing the Tucker tensor decomposition technique, significantly improving prediction accuracy and computational efficiency while avoiding the simple stacking and interaction of features. Additionally, the collaborative guidance module implements a dual-channel aggregation and collaborative guided learning mechanism, further enhancing classification efficacy. Comprehensive experimental results indicate that MAQG surpasses existing techniques in the quality evaluation of traditional Chinese herbal medicines, achieving an average classification accuracy of 88.14% for astragalus quality grades, significantly improving the recognition accuracy of the intricate multimodal data associated with traditional Chinese herbal medicine slices.