Advancements in Plant Disease Detection: A Comprehensive Review of Traditional, Modern, and AI-Driven Approaches

Abstract

This paper provides a comprehensive review of advancements in plant disease detection, moving from traditional to modern and AI-driven approaches. It highlights that traditional methods, such as visual inspection, microbiological isolation, culturing, and molecular and serological techniques, are often limited by being time-consuming, subjective, or requiring specialized expertise and lab processing. These limitations can lead to significant crop yield losses, economic setbacks, and threats to food security. The review then discusses modern, non-destructive sensor technologies, which are crucial for detecting diseases in their early stages, often before visible symptoms appear. These technologies include: * Hyperspectral Imaging (HSI): Captures detailed "spectral fingerprints" of plants to detect subtle physiological changes. * Multispectral Imaging (MSI): Uses a limited number of spectral bands, often including near-infrared (NIR), to identify abnormal plant conditions more cost-effectively than HSI. * Thermal Imaging: Detects temperature fluctuations in plants caused by physiological changes during infection. * Chlorophyll Fluorescence Imaging (CFI): A non-invasive technique that detects early stress responses by analyzing chlorophyll emissions. * LiDAR and Drones: Used for aerial analysis of crop health, enabling early diagnosis and monitoring of large agricultural areas. Finally, the paper details how Artificial Intelligence (AI) and Deep Learning (DL) have revolutionized this field through automated, highly accurate diagnostic capabilities. The document covers various deep learning architectures, including Convolutional Neural Networks (CNNs) like AlexNet, VGG, ResNet, and YOLO, which are used for image classification, feature extraction, and real-time disease localization. It also mentions the use of semantic segmentation models like U-Net for pixel-level disease mapping, and the role of transfer learning and explainable AI (XAI) in improving model performance and transparency. The review concludes with an emerging paradigm of federated learning for decentralized, privacy-preserving model training.

Plant phenotyping relevance

植物病害の症状・生理状態を画像およびセンサーで検出する手法を中心に扱う包括的レビューであり、植物フェノタイピング手法のレビューに該当する。

Code and data availability

This is a comprehensive review article with no original plant-phenotyping measurements, datasets, images, or author analysis code. All listed URLs are cited third-party references (prior publications, tutorials, public datasets like PlantVillage/PlantCLEF, and unrelated GitHub projects), not paper-specific assets of an

No evidence-backed public reproduction asset is currently recorded.