Detection and classification of tomato diseases with artificial intelligence

With the rapidly increasing global population, the demand for food in the world is constantly increasing. Limited agricultural areas and climatic changes make it necessary to increase productivity and prevent losses in agriculture in order to meet this increasing need. Early and accurate detection and classification of plant diseases and pests to prevent losses in agriculture is possible with the use of technology. For this purpose, in the study, tomato leaf diseases were classified using the transfer learning method. Within the scope of the study, images containing 1100 ToBRFV (Tomato Brown Rugose Fruit Virus) of tomato leaves obtained from the Antalya region were increased with data augmentation methods, and a ToBRFV dataset consisting of 3070 images was created. The created ToBRFV dataset was combined with the dataset named "Tomato Disease Multiple Sources", which included images of 11 classes, and a more comprehensive dataset was created. This dataset, created from 35550 images and 12 classes, was trained for 10 epochs using Vgg19, ResNet101 and EfficientNetB3 architectures with the transfer learning method using stratified 5-fold cross validation and the performance results were compared. The average test accuracies of Vgg19, ResNet101 and EfficientNetB3 architectures were 85.02%, 96.96% and 99.27%, respectively. Among 15 different models resulting from 5 cross-validations in each architecture, EfficientNetB3-Model 1 gave the best test accuracy in the study with 99.54%. Moreover, the same model classified images belonging to 3 classes, including ToBRFV, with full accuracy. With the results obtained, it is seen that the EfficientNetB3 architecture can be used reliably and accurately in the classification of different tomato diseases.