Transfer Learning in Grape Disease Leaf Detection Based on Convolutional Neural Network

Abstract

In order to achieve rapid and accurate recognition of grape leaf disease images, this paper introduces a convolutional neural network model based on transfer learning for classifying diseased grape leaves. A new fully connected layer module was designed on the basis of the EfficientNetB0 model, and the convolutional layer of the EfficientNetB0 model, pre-trained on the ImageNet dataset, was transferred into this model. The training image data of thousands of images were obtained from Kaggle, including grape leaves with black rot disease, Esca disease virus, leaf blight disease, and healthy grape leaves. In order to expand the dataset and prevent overfitting, we carried out a series of preprocessing steps on the original dataset and divided the training and test sets in a 4:1 ratio. The test accuracy of our model reached 99.14% and the average F1-score reached 98.79%. This paper also compared the classification results of different model structures such as VGG-16 and RESNet50. Their test accuracy values are 96.29% and 97.06% respectively. To quantitatively evaluate the performance of the model, the accuracy, precision, recall and F1-socre of the model are calculated.