Advanced AI and computer vision-based real-time plant disease and pest detection system: a systematic literature review

Abstract

The integration of artificial intelligence (AI) with computer vision through real-time deep-learning models such as YOLOv5, MobileNetV3, and TinySegformer offers revolutionary solutions, achieving 90–96% accuracy at 30–50 frames per second (FPS) with less than 1W power consumption. This systematic literature review (SLR) analyzes 30 studies from 2015–2024, evaluating real-time plant disease and pest detection systems based on performance, computational efficiency, and agricultural applicability. Findings indicate that YOLO-based models provide the best speed-affordability balance, enabling immediate field diagnoses on Raspberry Pi edge devices, while non-realtime systems (e.g., VGG, ResNet) offer higher accuracy at the cost of longer processing times. Barriers such as occlusion, high costs, limited field datasets, and power constraints are addressed through attention mechanisms, low-cost hardware, crowdsourced datasets, and model optimization. Emerging technologies, including federated learning, IoT integration, hybrid CNN-Transformer models, UAV-based systems, and multimodal data fusion, enhance scalability, robustness, and accessibility, reducing pesticide use by 20–25% and recovering 10–15% of lost yields. This SLR outlines research directions for field model optimization, affordable precision agriculture tools, and policy strategies for equitable technology distribution, supporting sustainable agriculture and global food security.

Plant phenotyping relevance

植物病害を画像・AIで検出する手法を対象に、精度、処理速度、計算効率、実用性を比較評価した系統的レビューであり、病害状態の表現型推定手法が中心です。

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.