Detection of Tomato Diseases Using Deep Learning Methods

In recent years, significant advancements in artificial intelligence, particularly in the field of deep learning, have increasingly been integrated into agricultural applications, including critical pro-cesses such as disease detection. Tomato, being one of the most widely consumed agricultural products globally and highly susceptible to a variety of fungal, bacterial, and viral pathogens, remains a prominent focus in disease detection research. In the present study, a five-class classi-fication was conducted to distinguish between healthy tomatoes and those affected by the most common diseases: tomato late blight, early blight, gray mold, and bacterial cancer. The dataset used in this study is entirely original and consists of images captured in the production envi-ronment, covering three different categories: leaves, red tomatoes, and green tomatoes. Through twofold data augmentation, the dataset was expanded to include a total of 6,414 samples. The data was split into training and testing sets with a ratio of 80% to 20%, respectively, and 21 deep learning models were evaluated. Among these, the five best-performing models (NasNet-Large, ResNet-50, DenseNet-201, EfficientNet-b0, and Places365-GoogLeNet) were selected, and 1,000 features were extracted from each. From these extracted features, the top 100 most discriminative features were selected using three different feature selection algorithms: Minimum Redundancy Maximum Relevance (MRMR), Chi-Square (Chi²), and ReliefF. These features were subsequently used for reclassification with machine learning algorithms, employing five-fold cross-validation while maintaining the same train-test split. The highest test accuracy of 92.0% was achieved using the EfficientNet-b0 model combined with Chi² feature selection and the Fine K-Nearest Neighbors (KNN) classifier. Overall, EfficientNet-b0 emerged as the most effective deep learning model, while the ReliefF feature selection method and the Subspace KNN classifier yielded higher accu-racies compared to other methods. In contrast, the combination of NasNet-Large and the Wide Neural Network classifier demonstrated the weakest performance with the lowest test accuracy. These findings indicate that features extracted via deep learning, when combined with appropriate feature selection methods and traditional classifiers, can be highly effective in the detection of tomato diseases.