Advancing site-specific disease and pest management in precision agriculture: From reasoning-driven foundation models to adaptive, feedback-based learning

TL;DR

This review surveys how foundation models (FMs) are reshaping site-specific disease management in crops, with an emphasis on large language models (LLMs) and vision-language models (VLMs). It discusses how FMs enable multimodal perception, text-based reasoning, and interactive guidance, enabling new capabilities for targeted spraying and extension services. The paper highlights trends in the adoption of FMs, the rising prominence of VLMs over LLMs, and the nascent integration of reinforcement learning (RL), adaptive learning (AL), and digital twin (DT) frameworks to create closed-loop, feedback-driven SSDM. It also outlines key challenges—data quality, sim-to-real transfer, and safety—and opportunities for practical deployment through domain-adapted models, adapters, few-shot learning, and multi-farm DT networks. Collectively, the work maps a trajectory toward autonomous, data-driven crop protection that can adapt to diverse crops, diseases, and field conditions while reducing chemical inputs.

Abstract

Site-specific disease management (SSDM) in crops has advanced rapidly through machine and deep learning (ML and DL) for real-time computer vision. Research evolved from handcrafted feature extraction to large-scale automated feature learning. With foundation models (FMs), crop disease datasets are now processed in fundamentally new ways. Unlike traditional neural networks, FMs integrate visual and textual data, interpret symptoms in text, reason about symptom-management relationships, and support interactive QA for growers and educators. Adaptive and imitation learning in robotics further enables field-based disease management. This review screened approx. 40 articles on FM applications for SSDM, focusing on large-language models (LLMs) and vision-language models (VLMs), and discussing their role in adaptive learning (AL), reinforcement learning (RL), and digital twin frameworks for targeted spraying. Key findings: (a) FMs are gaining traction with surging literature in 2023-24; (b) VLMs outpace LLMs, with a 5-10x increase in publications; (c) RL and AL are still nascent for smart spraying; (d) digital twins with RL can simulate targeted spraying virtually; (e) addressing the sim-to-real gap is critical for real-world deployment; (f) human-robot collaboration remains limited, especially in human-in-the-loop approaches where robots detect early symptoms and humans validate uncertain cases; (g) multi-modal FMs with real-time feedback will drive next-gen SSDM. For updates, resources, and contributions, visit, https://github.com/nitin-dominic/AgriPathogenDatabase, to submit papers, code, or datasets.

Figures (8)

• Figure 1: An overview of the review performed in this study with a perspective to focus on the individual steps taken to address site-specific disease management.
• Figure 2: This review presents a systematic literature analysis focused on foundation models and their extended applications in site-specific crop disease management. The box on the left illustrates the employed methodology, while the box on the right highlights key questions derived from the retrieved research articles.
• Figure 3: Publication trends from 2019 to 2024 for large language models (LLM) and visual language models (VLM) across ScienceDirect and Scopus databases.
• Figure 4: Different categories of sensing systems and platforms used in disease image acquisition. The inverted triangle includes: RGB, multispectral (MS), and hyperspectral (HS) sensors where its application in sensing disease information increases from RGB to HS.
• Figure 5: Overview of a foundational model-based framework for crop disease diagnostics and management. Multimodal data sources including images, textual descriptions, structured dataset, and agri-sensors data. On the right, diverse downstream applications such as disease classification, image generation, smart spraying, question answering, and robotic navigation are enabled by the model's reasoning and generalization capabilities.