目的 基于广西海洋中药马刀的潜在药效成分，建立一种快速评价其药材品质与等级的智能分析方法。方法 参考《中国药典》2025年版牡蛎项下方法，测定马刀中碳酸钙、酸不溶性灰分及浸出物的含量，并采用茚三酮法测定总氨基酸含量。构建多任务一维卷积神经网络（1D-CNN）模型，以近红外光谱（near infrared spectroscopy，NIRS）数据为输入，同步预测多成分含量，并引入光谱注意力机制提升模型的可解释性。进一步结合CRITIC客观赋权法与TOPSIS多准则决策算法，建立端到端的AI-NIRS智能评价平台，依据综合得分实现品质等级的自动划分。结果 利用1D-CNN模型在90批次马刀样品的测定结果基础上建立了得分等级预测模型。与传统偏最小二乘（partial least squares，PLS）近红外模型相比，多任务CNN模型的预测性能显著更优。验证结果表明，马刀样品的综合得分呈高斯分布，各质量等级间具有良好区分度，说明本研究建立的等级评价方法具有良好的适用性和可靠性。结论 在现行的监管体系框架下，实现了对马刀药材品质的快速综合评价与等级划分，为贝类海洋中药的数智评价与科学监管提供了新的技术路径。

Objective To establish a rapid and intelligent analytical method for evaluating the quality and grade of the marine traditional Chinese medicinal material Madao, from Guangxi, based on its potential pharmacodynamic components. Methods The contents of calcium carbonate, acid-insoluble ash, and extractives in Madao were determined following the methods specified for oysters in the Chinese Pharmacopeia (2025 edition). The total amino acid content was measured using the ninhydrin method. A multi-task one-dimensional convolutional neural network (1D-CNN) model was constructed, using near-infrared spectroscopy (NIRS) data as input to simultaneously predict the contents of multiple components. A spectral attention mechanism was introduced to enhance model interpretability. Furthermore, the criteria importance through intercriteria correlation (CRITIC) objective weighting method and the technique for order preference by similarity to ideal solution (TOPSIS) multi criteria decision making algorithm were integrated to develop an end-to-end AI-NIRS intelligent evaluation platform. This platform enables automatic classification of quality grade based on comprehensive scores. Results A score-grade prediction model was established based on the determination results of 90 batches of Madao samples using the 1D-CNN. Compared with the traditional partial least squares (PLS) NIRS model, the multi-task CNN model demonstrated significantly superior performance. Validation results showed that the comprehensive scores of the Madao samples followed a Gaussian distribution, with clear differentiation among the various quality grades. This indicates that the established evaluation method is effective and reliable. Conclusion Operating within the framework of the current regulatory system, this method achieved rapid, comprehensive quality assessment and grade differentiation for Madao. It provides a novel technological approach for the digital-intelligence evaluation and scientific supervision of shell-based marine traditional Chinese medicinal materials.

R282.6

“带土移植”人才引育计划项目（桂科AA23026008）；广西海洋中药资源及标准调研（桂药监科(2023)001号）；广西重大专项计划项目（桂科JF2503980037）