Advancements in Plant Disease Detection: A Comparative Study of Deep Learning-Based Classification Models

Agriculture is essential to global food security, especially in nations like Bangladesh where it significantly affects people's economies and incomes. However, bacterial, viral, and fungal plant diseases translate into large annual losses in agricultural productivity. For farming to be sustainable, these illnesses must be identified early and treated effectively. In this work, a deep learning-based approach to plant leaf disease diagnosis is presented using a Convolutional Neural Network (CNN) trained using an enhanced version of the Kaggle data. To improve training results and eliminate class imbalances, the dataset used in this work underwent extensive preprocessing. In the proposed model, complex features are extracted for accurate disease diagnosis using multiple convolutional and pooling layers.In order to improve generalisation and decrease overfitting, data augmentation techniques were also used to diversify the training samples. Several deep learning architectures were evaluated, including ResNet50, InceptionV3,DenseNet121, DenseNet169, and others. The combination of DenseNet121 and DenseNet169 produced the greatest results. The test set, which comprised 20% of the dataset and included images of both healthy and sick plants, had an exceptional accuracy of 98% with precision of 97.60%, recall of 98.36%, and F1-score of 97.98%. The most recent advancements and challenges in applying deep learning to plant disease diagnosis are also covered in this paper, including scalability, resource optimisation, and real-time application. Drones and Internet of Things (IoT) technologies for automated reporting and real-time disease monitoring may be included into future versions of this framework, enabling more intelligent and effective farming practices.