Grapevine fruits disease detection using different deep learning models

In India, grapes are one of the most important crops for business. Grapes and their byproducts are one of India’s leading exports. The leaves of grapes are susceptible to a variety of diseases. Large-scale production of grapes can be affected by these diseases. The purpose of this research paper is to classify and detect disease on leaves using deep learning before taking a big picture. It describes the advances in detecting leaf disease and shows how to improve results. A total of seven different types of pre-trained deep Convolutional Neural Networks (CNNs) were used for transfer learning: MobileNet, InceptionResNetV2, DenseNet121, InceptionV3, Xception, VGG16 and ResNet101V2. There are 4062 images of grapevine leaves in total, divided into four classes: Leaf Blight, Black Rot, Black Measles, and Healthy. The major challenge was to detect if the leaf was healthy or had been infected, followed by classifying the type of disease of the leaf. In order to learn and detect the disease, these images were trained to seven different transfer learning models. The image classification accuracy obtained is 99.672% by DenseNet121 and it is the highest accuracy obtained compared to any accuracy reported in the literature. It is followed by 99.500% of Xception and 99.345% of VGG16, 99.345% of InceptionV3. Therefore, the proposed research paper can be useful for detecting Grapevine disease at an early stage and preventing its spread. This process can increase the production rate and profit for farmers.