Answers are grounded in this paper and cite their sources.
Weed mapping using deep learning remains underexplored despite its potential to support site-specific weed management and spatial analysis of weed distribution. The objective of this study was to develop and evaluate a proximal sensing pipeline for generating individual-plant weed maps using deep learning and high-precision Global Navigation Satellite System (GNSS) positioning. The system integrated tractor-mounted RGB cameras, YOLO-based object detection, and coordinate transformation to georeference nutsedges ( Cyperus spp.) across two field conditions. Mapping was conducted in a 98 × 26 m fallow field and a 24 × 105 m strawberry field. System evaluation included detection performance, positional accuracy, and comparison of map-derived weed densities with manual counts from ten 1 m 2 quadrats in each field. The generated maps identified 11,241 nutsedges in the fallow field (4 plants m -2 ) and 128 in the strawberry field (<1 plant m -2 ). Detection models achieved F1 scores of 0.88 in the fallow field and 0.75 in the strawberry field. Mean positional error was 7 cm in the fallow field and 11 cm in the strawberry field. Map-derived densities ranged from 4 to 55 plants m -2 in the fallow field (RMSE = 7.47) and from 0 to 10 plants m -2 in the strawberry field (RMSE = 2.35). Underestimation was the primary error source mainly associated with missed detections caused by occlusion of plant structures. The resulting maps preserved spatial structure and enabled interpretation of field-scale weed distribution. This approach provides a scalable ground-based method for generating high-resolution weed maps to support monitoring and site-specific management.
No takes yet. Share an insight, caveat, or question.
Rodriguez et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: