A Hybrid CNN–Vision Transformer Framework for Optimized Leaf Disease Detection Using Feature Fusion and Transfer Learning
Journal of Robotics and Control (JRC) · 05.01.2026 · 6 (6) · 3002-3012
DOI: 10.18196/jrc.v6i6.28426 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Robotics and Control (JRC), online erschienen am 05.01.2026. Verfasst von Kotranaik Dhakyanaikabdel und 1 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref). Der Volltext ist frei zugänglich (Open Access, CC BY-SA 4.0).
Abstract (Originalsprache)
Plant disease of leaves drastically affects the yields of crops, which is a threat to food security and financial sustainability of agricultural enterprises. Early identification of diseases with accuracy is vital for precision farming, i.e., smart farming. Random Forest (RF), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN) classifiers, being conventional, have already been employed for the sake of disease classification. These Machine Learning (ML) classifiers, nevertheless, are liable to be biased towards handcrafted features, sensitivity to lighting, background, and severity of disease, leading to lower accuracy and generalizability. To address these limitations, this study suggests a new hybrid deep learning architecture called Convolutional Neural Network – Vision Transformer (CNN-ViT), which unifies the spatial feature extraction of CNN with the Vision Transformers' global attention mechanism. As a result, the combined model can automatically capture hierarchical features and long-distance relationships from enhanced and preprocessed images of leaves, facilitating effective multiclass classification. Evaluation of the model is conducted based on a publicly available plant disease dataset, resulting in better performance than conventional ML techniques and individual deep learning models. The hybrid CNN–ViT framework combines the CNN’s ability to extract spatial features with the Vision Transformer’s long-range attention. We evaluated the framework on the Plant Village (banana subset) with a batch size of 32, learning rate 0.001, and 50 epochs. The hybrid framework was 8.4 % more accurate than comparable baseline CNN and 7.1 % and 6.5 % more precice and recall compared with the comparable baseline ViT, respectively. The model, while showing increased robustness, requires increased computational complexity and inference latency and is therefore key to scaleable real-time deployment. This combined methodology presents a scalable and efficient solution for real-time monitoring of leaf diseases and contributes towards the development of AI-powered smart agriculture.
Zitieren
APA
Dhakyanaikabdel, K., & J, A. (2026). A Hybrid CNN–Vision Transformer Framework for Optimized Leaf Disease Detection Using Feature Fusion and Transfer Learning. Journal of Robotics and Control (JRC), 6(6), 3002-3012. https://doi.org/10.18196/jrc.v6i6.28426
Vancouver
Dhakyanaikabdel K, J A. A Hybrid CNN–Vision Transformer Framework for Optimized Leaf Disease Detection Using Feature Fusion and Transfer Learning. Journal of Robotics and Control (JRC). 2026;6(6):3002-3012. doi:10.18196/jrc.v6i6.28426
BibTeX
@article{Dhakyanaikabdel2026,
author = {Dhakyanaikabdel, Kotranaik and J, Anandababu},
title = {A Hybrid CNN–Vision Transformer Framework for Optimized Leaf Disease Detection Using Feature Fusion and Transfer Learning},
journal = {Journal of Robotics and Control (JRC)},
year = {2026},
volume = {6},
number = {6},
pages = {3002--3012},
doi = {10.18196/jrc.v6i6.28426}
}
EUVIMED