抄録
Deep learning (DL) and reinforcement learning (RL) are rapidly transforming intelligent irrigation by enabling accurate environmental perception, predictive modeling, and adaptive irrigation decision-making. Despite substantial progress, a comprehensive understanding of their methodological characteristics, application scope, and remaining challenges is still lacking, limiting their broader deployment in precision water management. This review systematically synthesizes recent advances in DL- and RL-based intelligent irrigation, covering crop water requirement prediction, soil moisture and evapotranspiration estimation, irrigation scheduling, and autonomous control. DL has significantly improved feature extraction and predictive performance by integrating multisource data from meteorological observations, soil monitoring, remote sensing, and Internet of Things (IoT) sensing networks, whereas RL has demonstrated unique advantages in sequential decision-making, long-term policy optimization, and adaptive control under dynamic environmental conditions. Their integration with crop growth models, remote sensing, and intelligent sensing technologies has further expanded the capability of intelligent irrigation systems. However, several critical challenges remain, including limited data availability and quality, inadequate model generalization and cross-regional transferability, difficulties in reward design under environmental uncertainty, insufficient model interpretability, and barriers to large-scale field deployment. This review provides a comparative assessment of DL and RL from the perspectives of methodological frameworks, application scenarios, and performance characteristics, and identifies key research priorities, including interpretable and trustworthy artificial intelligence, mechanism–data fusion, multi-agent collaborative decision-making, and water–energy–food nexus optimization. By integrating these perspectives into a unified analytical framework, this review establishes a roadmap for the next generation of intelligent irrigation systems and highlights future directions toward scalable, robust, and sustainable precision water management.
