Review: Artificial Intelligence and Deep Transfer Learning for Plant Disease Detection and Classification

Abstract

The persistent threat of plant disease epidemics poses significant challenges to global agriculture, making crops susceptible to catastrophic diseases that compromise food security and nutritional well-being. This review critically examines the application of deep transfer learning and convolutional neural networks (CNNs) in classifying plant diseases, such as tomato leaf diseases. By synthesizing recent advancements in the field, the article highlights how pre-trained models, trained on large-scale image datasets, can be adapted to recognize disease-specific patterns in agricultural contexts. The discussion encompasses key methodologies, including the integration of custom architectures and shallow classifiers, as exemplified by works such as Fruit and Vegetable Leaf Disease Recognition based on a Novel Custom Convolutional Neural Network and Shallow Classifier and An Integrated Framework of Two-Stream Deep Learning Models Optimal Information Fusion for Fruits Disease Recognition. A critical analysis of existing approaches is provided, addressing their strengths, limitations, and the role of dataset quality and diversity in model performance, including the use of publicly available datasets of labelled plant disease images, such as PlantVillage. The review underscores the transformative potential of automation and robotics in reducing disease spread while emphasizing unresolved challenges, such as the need for cost-effective, scalable frameworks. By identifying gaps in current research and proposing future directions, this article aims to guide the development of sustainable, AI-driven solutions for agricultural productivity.

Plant phenotyping relevance

植物病害を画像から検出・分類するAI手法を体系的に検討するレビューであり、植物の病徴・病害状態を推定するフェノタイピング手法が中心です。

Code and data availability

This is a review article on deep transfer learning for plant disease detection. It presents no original phenotyping measurements, datasets, images, code, or trained models of its own. Public datasets like PlantVillage are mentioned only as cited prior-work resources, and all references are to third-party studies, none附

No evidence-backed public reproduction asset is currently recorded.