Research on the identification of Tibetan herbal medicine based on STM-A Net

In order to implement the accurate, nondestructive, and fast classification of Tibetan medicinal materials, taking the current images of various Tibetan medicinal materials as the research object, an STM-Attention module of plug-and-play is proposed based on deep learning. In this paper, the STM-Attention modules are stacked according to the ResNet method, forming a new classification backbone network structure-STM-A Net. At the same time, the network has the advantages of small parameters and computation, and can better extract features under the condition of ensuring speed. Experimental results show that under the same conditions, the network achieves a balance between parameter quantity and accuracy, and the accuracy rate reaches 97.18% on the test set of Tibetan medicinal materials.