Hybrid Deep Learning Framework for Multi-Class Plant Disease Classification and Detection

Abstract

Plant disease detection plays a vital role in precision agriculture by enabling early diagnosis and reducing crop losses. However, challenges such as complex backgrounds, class imbalance, and limited training samples often hinder the performance of standalone deep learning models. This paper proposes a hybrid deep learning framework that integrates five pre-trained Convolutional Neural Networks (CNNs) using a weighted ensemble strategy for multi-class plant disease classification. Laplacian sharpening is applied during preprocessing to enhance leaf texture and edge information, improving feature extraction and classification accuracy. The model is evaluated on the PlantVillage dataset and a custom IIITDMJ Maize dataset containing realfield and synthetic images. Experimental results demonstrate that the ensemble model outperforms individual architectures, achieving improved accuracy, robustness, and generalization across multiple disease classes. The proposed approach offers a scalable solution for real-world agricultural monitoring and smart farming applications.