One of the most challenging problems are tomato diseases which have been the cause of economic problems that have an effect on the production of agricultural products and food security worldwide. To stop the enormous loss of production and maintain the environmental sustainability, it is very important to locate the diseases at an early stage and be accurate. Six deep transfer learning architectures—VGG16, ResNet50, InceptionV3, DenseNet121, EfficientNetB0, and MobileNetV3- Large—have been studied very closely in this paper to classify tomato leaf diseases using the recently created Tomato-Village dataset. The dataset is very suitable for deployment at the field level as it contains the real photos of both healthy and diseased tomato leaves which were taken in a variety of lighting condition and background. To extend the models, they were updated with a substantial data augmentation and transfer learning to overcome the limitation of the dataset. Experiment results indicate that MobileNetV3-Large achieved the highest classification accuracy of 91.32 percent, thus it outperformed other heavier models with exceptional processing efficiency. In precision agriculture, a lightweight CNN architecture such as MobileNetV3-Large could be a practical and efficient way to automatically diagnose tomato disease in real time.
Bakshi et al. (Tue,) studied this question.