Deep Transfer Learning with Hybrid CNN Fusion for Smart Tomato Leaf Disease Diagnosis

Tomato plants are highly susceptible to a wide range of bacterial, viral, and fungal diseases that significantly affect crop yield and agricultural productivity, making rapid and accurate identification essential for effective crop management and early intervention. This paper presents Fusion-TLNet, a hybrid deep learning framework that integrates multiple transfer learning models using a feature-level fusion strategy for robust tomato leaf disease classification. The proposed architecture combines three pretrained convolutional neural networks: EfficientNetB0, MobileNetV3-Small, and DenseNet121 to capture complementary hierarchical representations of texture, shape, and color features from tomato leaf images. Experimental results on the PlantVillage dataset demonstrate that Fusion-TLNet achieves an accuracy of 99.34%, precision of 99.28%, recall of 99.22%, and F1-score of 99.25%, outperforming individual backbone models by up to 1.23% in accuracy. In addition, the model maintains a low prediction latency of 9.7 ms per image, making it suitable for real-time deployment in resource-constrained agricultural environments. The proposed model provides an interpretable, efficient, and scalable solution for intelligent plant disease diagnosis, supporting the advancement of data-driven precision agriculture.