Grape Leaf Disease Recognition Model Based on Improved MobileViT
This study put forward an optimized grape leaf disease recognition model,CD-MobileViT.Firstly,the MobileViT was used as the base network,and the coordinate attention(CA)modules were embed-ded after Layer1 and Layer2 to enable the network to more effectively capture key features from different posi-tions.Secondly,a Dropout layer was added after the network's fully connected layer to prevent overfitting.Lastly,the optimizer AdamW(Adam with Weight Decay Regularization)was used to better control model complexity and improve generalization ability.The experimental results showed that compared to the base Mo-bileViT network,the improved network CD-MobileViT improved precision,recall,F1 score and accuracy by 1.77,1.85,1.65 and 1.75 percentage points,respectively.Additionally,it also showed varying degrees of im-provement(0.25～1.47 percentage points)compared to other classic networks(InceptionV1,MobileNetV2,EfficientNetB0,VGG-16).These results indicated that the improved model had good effect on grape leaf dis-ease recognition and could be deployed on mobile devices in the future,which provided a new solution for the accurate recognition of grape leaf diseases.
Recognition of grape leaf diseasesMobileViT networkCoordinate attentionAdamW opti-mizerDroupout layer
万方数据
维普
