Next-generation Site-specific Weed Management: A Critical Synthesis of Sensing, Artificial Intelligence and Precision Actuation

Abstract

Weed management is entering a transition from uniform field-wide treatment to spatially selective intervention at patch, row and individual-plant scales. This critical narrative review evaluates the technological and agronomic evidence underpinning site-specific weed management, with emphasis on the coupling of sensing, artificial intelligence, positioning, decision logic and precision actuation. The review focuses on peer-reviewed literature from 2000 to 9 June 2026, supplemented by foundational work within that interval that established practical sensor-based systems. Evidence was appraised for methodological quality, field realism, generalisability and the degree to which technical performance translated into weed-control efficacy, crop safety, input reduction and operational viability. The literature shows that sensing and computer vision have advanced rapidly, especially through high-resolution unmanned aerial vehicle imagery and deep-learning-based crop–weed discrimination. Yet benchmark classification accuracy remains an incomplete predictor of field performance because domain shift, occlusion, weed growth stage, illumination, georeferencing error, latency and actuator footprint can propagate through the control chain. Targeted spraying currently has the strongest field evidence for preserving agronomic efficacy while reducing herbicide use when weeds are spatially heterogeneous, although savings are highly context-dependent. Camera-guided hoeing and robotic intra-row cultivation can reduce hand labour and herbicide dependence, but crop safety, soil conditions and work rate remain decisive constraints. Laser and electrical methods demonstrate credible biological mechanisms and increasingly realistic prototypes, but their large-scale energy, throughput, safety and economic performance remains less established. The central conclusion is that next-generation weed management should be judged as a closed-loop agronomic system rather than as an isolated detection algorithm or actuator. Research priorities therefore include multi-environment validation, uncertainty-aware decision thresholds, standardised field metrics, interoperable sensing–actuation architectures, resistance-aware treatment rules and transparent whole-system economic and environmental assessment.