Empirical Analysis of Deep Learning Models for Tomato Leaf Disease Detection

Currently, the detection of plant diseases relies heavily on manual observation, which is time-consuming and susceptible to inaccuracies. To address the significant threat posed by plant diseases and harmful insects in the agriculture sector, researchers are actively working on developing an automated system for detecting diseases in plant leaves. This paper specifically focuses on the detection of diseases in tomato leaves using various models, both custom-built and pre-trained, with an emphasis on CNNs. Two datasets, PlantVillage and Tomato Leaf Disease dataset, are employed for evaluation. The results of the analysis reveal that the CNN-KNN Hybrid model performs the best for the PlantVillage dataset. For the Tomato Leaf Disease dataset, VGG16 and VGG19 prove to be the most efficient models. It is noteworthy that there are significant variations in model performance depending on the dataset type. Six models exhibit superior performance on the imbalanced dataset, while the other two show improved results on the balanced dataset. The study offers valuable insights for practitioners aiming to optimize model selection and enhance the accuracy and effectiveness of disease detection systems in agriculture. The integration of Deep Learning and Computer Vision holds great promise for improving crop management practices and mitigating losses for farmers.