<![CDATA[Leaf diseases in tomato plants (Solanum lycopersicum), including Early Blight, Late Blight, and Leaf Mold, can cause substantial reductions in crop yield if not detected at an early stage. Conventional manual detection methods are constrained by limitations in speed, consistency, and accuracy, particularly under field conditions. This study proposes a tomato leaf disease classification framework leveraging a transfer learning approach, in which the Inception V3 architecture functions as a feature extractor and the Random Forest algorithm serves as the classifier. The dataset employed comprises four categories of tomato leaf images—Early Blight, Late Blight, Leaf Mold, and Healthy—which were stratified into training (80%) and testing (20%) subsets. All images were resized to 299×299 pixels, normalized, and subjected to optional data augmentation. Feature representations were extracted from the Global Average Pooling layer of Inception V3 pretrained on the ImageNet dataset and subsequently input into a Random Forest classifier with hyperparameters optimized via grid search. Experimental evaluation demonstrated that the proposed model achieved an accuracy of 94.3%, surpassing the performance of a conventional CNN (89.2%) and a Random Forest classifier without transfer learning (76.5%). The confusion matrix analysis revealed the highest classification performance for the Healthy and Late Blight categories, whereas the Leaf Mold category exhibited a higher misclassification rate due to its visual symptom similarity to Early Blight. The findings of this research indicate that a hybrid methodology combining deep learning-based feature extraction and classical machine learning algorithms is highly effective for agricultural image classification in scenarios with limited datasets. Furthermore, the proposed approach holds significant potential for integration into web- or mobile-based decision support systems, enabling rapid and accurate plant disease detection in practical agricultural settings.]]>
Copyrights © 2025
