基于深度学习的滇黄精"辨状论质"研究OA

Objective To explore new identification methods and technical means for evaluating the"quality evaluation through morphological identification"of Dianhuangjing(Polygonatum kingianum).Methods The appearance characteristics of fresh and dried samples were collected,and permanent sections were prepared.The Densenet-121 network model was trained to automati-cally identify Dianhuangjing(Polygonatum kingianum)slices.Microscopic features were observed and collected under a micro-scope,and the corresponding image sets were established to train the YOLOv7 model for automatically recognizing and counting the number of vascular bundles and needle crystals in Dianhuangjing(Polygonatum kingianum)microscopic slices.Results The Densenet-121 network model achieved an average recognition rate of 90.97％for Dianhuangjing(Polygonatum kingianum)slices.Compared to manual counting,the YOLOv7 model achieved an average accuracy of 98.14％for identifying vascular bun-dles in Dianhuangjing(Polygonatum kingianum)slices,and an average accuracy of 53.2％for identifying calcium oxalate needle crystals.Conclusion The Densenet-121 network model can achieve relatively high and stable automatic identification between Dianhuangjing(Polygonatum kingianum)and adulterants.The YOLOv7 model can effectively count the number of microvascular bundles and needle crystals in Dianhuangjing(Polygonatum kingianum)slices.Taking Dianhuangjing(Polygonatum kingianum)as an example,a new research approach and technical method for rapid and accurate AI identification of traditional Chinese me-dicinal materials was constructed.