VGG-EffAttnNet: Hybrid Deep Learning Model for Automated Chili Plant Disease Classification Using VGG16 and EfficientNetB0 With Attention Mechanism.

Chili plant diseases significantly impact global agriculture, necessitating accurate and rapid classification for effective management. The study introduces VGG-EffAttnNet, a hybrid deep learning model combining VGG16 and EfficientNetB0 with attention mechanisms and Monte Carlo Dropout (MCD) for robust chili plant disease classification. VGG16 captures spatial and hierarchical features, while EfficientNetB0 ensures efficient, high-accuracy learning. Attention enhances focus on disease-relevant areas, and MCD improves robustness by estimating uncertainty. The study utilizes a chili plant disease dataset sourced from Kaggle, comprising 5000 images across five classes: Healthy, Leaf Curl, Leaf Spot, Whitefly, and Yellowish, after extensive data augmentation techniques, including rotation, flipping, zooming, and brightness adjustment, to improve model generalization. Feature extraction is performed using VGG16 and EfficientNetB0, followed by concatenation and refinement through attention mechanisms, enabling the model to focus on disease-relevant features while suppressing background noise. MCD is integrated to estimate model uncertainty and mitigate overfitting. Experimental results demonstrate the superior performance of the proposed hybrid model. The concatenated VGG16 and EfficientNetB0 model achieved a classification accuracy of 99%, precision, and recall of 99%, surpassing individual model performances (VGG16: 96.8%, EfficientNetB0: 96.5%, and attention-integrated variants reached up to 98%). The F1-score reached 99% across all disease categories, ensuring high precision and recall. Compared to state-of-the-art models like InceptionV3 (98.83%) and MobileNet (97.18%), the proposed hybrid model demonstrates improved classification accuracy and robustness. The study underscores the potential of deep learning-based automated disease classification in precision agriculture, enabling early intervention and reducing reliance on chemical treatments. Future work aims to extend the approach to real-time deployment on mobile and edge devices, integrate explainability techniques for enhanced interpretability, and explore federated learning for decentralized agricultural diagnostics.