Akademik Veri Yönetim Sistemi
Applied Fruit Science, cilt.68, sa.1, 2026 (SCI-Expanded, Scopus)
Timely and accurate detection of plant diseases is crucial for enhancing agricultural production efficiency. In this study, a hybrid deep learning and machine learning approach is proposed using image-based data for the classification of diseases observed in cassava and cashew plants. In the first stage, classification is performed directly on the images with the vision transformer (ViT) model. In the second stage, 300-dimensional feature vectors are extracted using ViT and capsule network (CapsNet) models, and they are combined to obtain a 600-dimensional numerical feature vector. This vector is fed as input to various classification models, including CapsNet, TabNet, FT-transformer, support vector classifier (SVC), MLP classifier, gradient boosting, random forest (RF), and naive Bayes (NB). According to the results, 95.31% accuracy was obtained for the cassava dataset and 98.08% for the cashew dataset in the direct classification performed with ViT. However, the success rate increased significantly in the classifications performed with feature extraction. The SVC model showed the best performance with 97.07% accuracy in the cassava dataset and the CapsNet model with 98.65% accuracy in the cashew dataset. The findings show that hybrid feature extraction and appropriate model selection are effective in increasing the accuracy of plant disease classification.