Research on fast recognition system of chinese materia medica based on VGGNet,EfficientNet and CaiT models
Objective To establish a rapid recognition method of CMM(chinese materia medica)by independently collecting image data set of CMM and comparing three different deep learning models.Methods A total of 402 samples of 248 CMM from 10 manufacturers were collected in the market,and the images were taken by digital camera to establish the CMM image dataset.The related image data are summarized and amplified into a training set,a validation set and a test set.VGG19,EfficientNetV2-B and CaiT were selected as the backbone models respectively.Transfer learning was performed based on ImageNet,20 rounds of training were performed to test their fit degree,and 10-fold cross-validation was performed to select suitable models and establish the recognition system.Results VGG19 and CaiT showed a good fit.Poor convergence occurs in EfficientNetV2-B3.VGG19 and CaiT are selected as the backbone network,and the verification set data is used for 10-fold cross-validation,and the accuracy rate of VGG19 is 81.6%and the accuracy rate of CaiT is 77.06%.Conclusion In this study,deep learning technology can effectively build a recognition system of common CMM with high accuracy under the selection of appropriate backbone model.
Traditional Chinese medicineArtificial intelligenceDeep learningNeural network
万方数据
维普
