A LIGHTWEIGHT PARIS L. DECOCTION PIECES RECOGNITION METHOD BASED ON ATTENTION MECHANISM

Aimed at the automatic recognition of Paris mairei H. Lév., Paris polyphylla var. yunnanensis (Franch.) Hand.-Mzt. and Paris polyphylla Sm. Var. Alba H Li et R. J., a lightweight Paris L. decoction pieces classification model based on attention mechanism is proposed. Two multi-scale feature extraction modules were proposed to comprehensively extract multiple scale features. On the basis of ECA-Net and spatial attention mechanism, ECSA-Module (Efficient channel and spatial attention module) was proposed to make full use of feature map channels and spatial information. The backbone network was densely connected, and the random erasing method was used for data enhancement. The experimental results show that the classification accuracy of the model is as high as 96.74%, which is 3.26 percentage points, 2.82 percentage points and 2.22 percentage points higher than that of MobileNet-V2, VGG16 and Xception respectively. The recognition system of Paris L. based on this model has high recognition accuracy and high speed, which has important practical application value.