An Intelligent System for Early Prediction of Crop Diseases using Deep Learning

Abstract

Some of the major culprits include plant diseases which immensely lower agricultural production, as well as cause severe economic losses and global food insecurity. Manual diagnosis, the traditional diagnosis of diseases in plants, is far too often incorrect, tardy, and requires acquaintance with experts. The present work presents an understandable deep learning framework that transforms the automated detection and severity estimation of plant diseases using images only. The proposed model applies a combination of Convolutional Neural Networks (CNNs) and Vision Transformers (ViT), capable of obtaining local and global feature descriptions through the corresponding leaf pictures, making it easier to detect even extremely minor disease alterations. Additionally, Grad-CAM++ is used to interpret the decision and display the model’s decision areas visually to the users. The system has been trained on the public Plant Village dataset, and real-field plant images are inserted to strengthen its robustness and generalization. The experimental results proved the proposed method to be superior to classical CNN-based techniques, as they obtained overall accuracy in disease classification above 98% and severity prediction that was trust worthy. The developed framework is completely software-based, scalable, and open to integration with web or cell phone applications. Therefore, it offers an explicable, interpretable, and simultaneously effective solution for precision agriculture.