Plant Leaf Disease Detection Using Efficient Net V2-S with Transfer Learning

Abstract

Early and accurate detection of plant leaf diseases plays a vital role in improving crop productivity and ensuring sustainable agriculture. This paper presents a deep learning-based framework for multi-class classification of banana leaf diseases using transfer learning. Initially, a baseline model based on ResNet50 is developed to evaluate standard performance. To enhance classification accuracy and computational efficiency, a transfer learning approach employing EfficientNetV2 is proposed. The pretrained EfficientNetV2-S model is fine-tuned by integrating a custom classification head comprising global average pooling, dropout, and fully connected layers.
The proposed model is trained and validated on a dataset containing four classes of banana leaf images, namely Cordana, Healthy, Pestalotiopsis, and Sigatoka. Experimental results demonstrate that the proposed approach achieves an overall accuracy of 95%, along with high precision, recall, and F1-score across all classes. The confusion matrix and training curves further confirm the robustness, stability, and generalization capability of the model. Comparative analysis indicates that the proposed EfficientNetV2-S-based framework outperforms the baseline ResNet50 model while maintaining reduced computational complexity.
To further evaluate practical applicability, the proposed model was tested on real-world banana leaf images captured under natural field conditions. The model achieved a detection accuracy of 76.19%, demonstrating its robustness and ability to generalize effectively beyond controlled datasets.
The results show that the proposed framework provides an efficient and scalable solution for real-world plant disease detection in precision agriculture. Future work will focus on expanding dataset diversity and exploring advanced architectures to further improve classification performance.