Detection of tomato diseases using top-view dataset with deep learning

Tomato (Solanum lycopersicum) is a strategic agricultural product with high economic value and widely consumed worldwide. However, its production faces threats from various plant diseases. Tomato Brown Rugose Fruit Virus (ToBRFV), which has spread rapidly in recent years, poses a serious risk to commercial tomato production. Early detection of plants infected with this virus is critical for preventing the spread, reducing crop losses and supporting sustainable agricultural practices. In this study, a deep learning-based image classification model was developed for the early detection of ToBRFV infection in commercial tomato plants. Within the scope of the project numbered FOA-2023-2587 supported by Eskişehir Osmangazi University Scientific Research Projects Coordination Unit, 40 healthy and 45 ToBRFV inoculated plants were grown in greenhouse conditions; top view RGB video recordings were taken for 30 days. The frames selected from these videos were processed by segmentation method and made ready for classification. The dataset obtained was divided into 60% training, 20% validation and 20% testing. VGG16, ResNet50, InceptionV3, MobileNetV2, Xception and DenseNet169 CNN models were used for ToBRFV virus detection. These models were trained and tested under the same conditions. In the experimental study, the most successful result was obtained from the DenseNet169 model with 98.30% accuracy and high consistency rates. The developed system is adaptable not only for tomato but also for other plant species and diseases and offers a scalable solution that will contribute to smart agricultural technologies.