Smart Detection Techniques for Plant Leaf Diseases Using Deep Learning : A Systematic Literature Review

Abstract

Agriculture is an essential part of ensuring global food security, but crop productivity is frequently threatened by plant diseases. Leaf diseases on tomato crops, in particular, can cause substantial losses if not identified early. Traditional disease identification methods rely on visual inspection, which is labor-intensive and prone to error. In recent years, deep learning techniques, especially convolutional neural networks (CNNs), have gained attention for automated plant disease diagnosis using leaf images. This paper presents a comprehensive review of existing deep learning approaches for tomato leaf disease detection, focusing on CNN architectures, attention mechanisms such as squeeze-and-excitation (SE) blocks, data augmentation strategies, and training optimizations. The reviewed studies are categorized according to network architecture, dataset, plant species, and reported performance metrics. Publicly available datasets, such as PlantVillage, are discussed, along with limitations in real-world applicability due to domain shift. Challenges faced by existing methods, including dataset bias, class imbalance, and overfitting, are highlighted. Finally, research gaps are identified, and directions are suggested to enhance the robustness, generalization, and practical applicability of deep learning-based plant disease diagnosis systems for sustainable agriculture.