Hybrid Vision Mamba Convnext V2 for Sesame Leaf Disease Classification
K Sivasankari B Pazhaniraj
Abstract
Purpose Sesame (Sesamum indicum L.) is an economically important oilseed crop whose productivity is significantly affected by foliar diseases. This study aims to develop an automated and accurate deep learning framework, VMCAF-Net, for early identification of sesame leaf diseases, supporting crop health monitoring and precision agriculture. Design/Methodology/Approach The proposed VMCAF-Net integrates Vision Mamba and ConvNeXt V2 to capture complementary global contextual and local texture features. An enhanced preprocessing pipeline incorporating CLAHE, gamma correction, image resizing, and data augmentation is employed to improve image quality and model robustness. A Cross-Attention Feature Fusion module combines the extracted representations, while a Label-Smoothed Softmax classifier performs disease classification. Model performance is evaluated using Accuracy, Precision, Recall, F1-score, ROC-AUC, Cohen's Kappa, and Matthews Correlation Coefficient. Findings Experimental evaluation demonstrates that VMCAF-Net outperforms existing deep learning models in sesame leaf disease identification. The integration of Vision Mamba and ConvNeXt V2 effectively captures both local disease characteristics and broader contextual patterns, while cross-attention fusion enhances discriminative feature representation. The results indicate that the proposed framework provides an accurate, robust, and computationally efficient approach for automated disease detection. Practical Implications The proposed framework can support farmers, agricultural researchers, and precision-agriculture practitioners by enabling rapid and automated identification of sesame leaf diseases. Its potential for real-time deployment can facilitate timely disease management, reduce crop losses, and support data-driven agricultural decision-making. Originality/Value The study proposes a novel hybrid architecture that combines Vision Mamba, ConvNeXt V2, cross-attention feature fusion, and enhanced image preprocessing for sesame leaf disease identification. The integration of complementary feature representations with comprehensive performance evaluation provides a potentially effective framework for intelligent and scalable crop disease diagnosis.