In agriculture, plant diseases severely affect productivity and lead to starvation. Hence, early treatment of plant diseases is crucial. However, the existing disease detection approaches are time consuming and require human intervention. To overcome these issues, deep learning models have been extensively used. The primary objective of this study is to develop an automated and efficient hybrid deep learning model by combining a Morphological U-Net, a Convolutional Neural Network (CNN) and a Spectral Morphological Transformer, a Vision Transformer (ViT) to perform accurate plant diseases detection. This hybrid model combines local feature extraction and global contextual learning to improve lesion detection and localization and classification accuracy. To analyse the effectiveness of the hybrid model, the benchmark PlantVillage dataset is used. Among the multiple plant species in this dataset, pepper, tomato, and potato plant leaves and their corresponding classes are selected for training and testing. The proposed model accurately classified multiple class plant diseases with an accuracy of 97.89% and outperformed existing deep learning models. The significance of this study lies in timely intervention for plant diseases, reducing crop loss, and promoting precision agriculture practices. It is apparent that this model is suitable for challenging environmental condition and edge computing deployment.
