多模态光谱融合为突破中药复杂体系质量控制的表征瓶颈提供了新的技术路径。单光谱技术因信息载量不足，难以独立完成对中药复杂体系的全面表征。系统综述了多模态光谱融合在中药质量智能检测领域的研究进展，首先将融合策略归纳为初级、中级与高级3类，阐述其原理及适用场景；其次梳理从传统机器学习到深度学习模型的算法演进，并强调可解释性人工智能的关键作用。在此基础上，进一步剖析当前面临的核心挑战，包括数据层面的采集标准化缺失与模态异质性，以及模型层面的异构特征融合困难与跨场景迁移障碍。最后，从数据规范和模型设计2个维度展望未来发展方向。多模态光谱融合的策略与算法，可为构建中药质量智能检测新范式提供核心支撑。

Multimodal spectral fusion provides a new technical pathway to overcome the characterization bottleneck in quality control of complex traditional Chinese medicine systems. Due to insufficient information capacity, single-modal spectral techniques cannot independently achieve comprehensive characterization of such complex systems. This paper systematically reviews the progress of multimodal spectral fusion for intelligent quality detection in traditional Chinese medicines. First, fusion strategies are classified into low-level, mid-level, and high-level categories, with their principles and application scenarios described. Second, the algorithm evolution from traditional machine learning to deep learning models is reviewed, highlighting the critical role of explainable artificial intelligence. On this basis, this paper then analyzes the core challenges at two levels. At the data level, challenges include a lack of standardized collection and modality heterogeneity. At the model level, challenges include difficulties in heterogeneous feature fusion and cross-scenario transfer. The strategies and algorithms of multimodal spectral fusion can provide core support for constructing a new paradigm for intelligent quality detection of traditional Chinese medicines.

R284

国家工信部2024年重点专项（2440STCZB2617）