【摘要】 目的:为实现中药性状鉴别智能化升级，建立人工智能图像识别与数据分析方法，实现对不同规格黑豆、混淆品野大豆以及常见伪品黑芸豆的高效、精准区分。方法:收集黑豆及其相关品种的性状样本数字图像制作数据集；利用深度学习方法学习该数据集以获得对黑豆及其相关品种的鉴别能力；通过生成式样本扩充方法与度量学习提升模型性能，从而获得更好的鉴别能力。结果:在黑豆及其相关品种数据集上获得了99.8%的精确率、100.0%的召回率、99.9%的平均精度均值(mean average precision, mAP)、99.8%的mAP@0.5:0.95(阈值从0.5～0.95以0.05为步长计算的多个mAP分数的均值);所训练的模型能够有效区分黑豆及其相关品种，并能够直观展示其检测结果。结论:本研究以黑豆及其相关品种为范例，为中药性状鉴别智能化升级提供参考，以期提升中药鉴别效率与准确性，完善中药质量标准体系，赋能中药产业全链条数字化转型，为中药监管与市场流通提供技术支撑。更多还原

【Abstract】 Objective: To establish an artificial intelligence image recognition and data analysis method to efficiently and accurately distinguish different specifications of black beans, confused wild soybeans, and common counterfeit black kidney beans, thus to achieve intelligent upgrading of traditional Chinese medicine trait identification. Methods: Digital image samples of black bean traits were collected to create a dataset. Deep learning methods were used to learn the dataset to obtain the ability to distinguish various types of black beans and their counterfeits. Sample expansion methods and metric learning methods were used to improve model performance and obtain better discriminative ability. Results: The method showed accuracy of 99.8%, recall of 100.0%, mean average precision(mAP) of 99.9%, and mAP@0.5: 0.95 of 99.8% on the black bean dataset. The trained model could effectively distinguish various types of black beans and their counterfeits, and visually display their detection results. Conclusion: This study takes black beans and related varieties as an example, providing references for the intelligent upgrading of traditional Chinese medicine trait identification, aiming to improve the efficiency and accuracy of the identification, perfect the quality standard system, empower the digital transformation of the entire industry chain, and provide technical support for regulation and market circulation of traditional Chinese medicine.更多还原