Deep Learning Models for Plant Leaf Disease Detection: A Comprehensive Analysis of CNN, Lightweight, and Explainable Architectures

Abstract

Plant diseases represent a critical global threat to food security, necessitating accurate and timely detection for sustainable agriculture. This systematic synthesis evaluates recent advancements in deep learning (DL) for plant leaf disease detection, analyzing 10 peer-reviewed studies published between 2021 and 2025. Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, this review systematically examines Convolutional Neural Network (CNN)-based architectures, lightweight models (MobileNet, EfficientNet), and the integration of Explainable AI (XAI) to analyze the inherent trade-offs among efficiency, accuracy, and interpretability. Key findings indicate that while modern DL models achieve high classification accuracy ($\geqslant 97 \%$), practical deployment is significantly hindered by limited generalization across diverse field conditions and a critical lack of integrated XAI within lightweight hybrid architectures. This study provides a structured, cross-metric analysis to guide the development of next-generation DL models that are simultaneously efficient, accurate, and interpretable for realworld agricultural applications.

Plant phenotyping relevance

植物葉の病害状態を画像から検出する深層学習手法を体系的に比較・分析したレビューであり、病害表現型の取得・推定方法が中心です。

Code and data availability

公開本文の所在を確認できませんでした。非公開または購読が必要な可能性があります。

No evidence-backed public reproduction asset is currently recorded.