In the wild blueberry (Vaccinium angustifolium Aiton) system, perennial weeds are a significant challenge. Weeds compete for resources and have a patchy, heterogeneous distribution, often growing among the rhizome of the crop. Producers broadcast spray herbicides to control weeds, but currently seek to lower costs, reduce environmental impact, and mitigate herbicide resistance. Site-specific weed management (SSWM) using weed maps offers a targeted solution, but its adoption is limited by the lack of software compatible with commercial spray equipment and by technological constraints, such as weed patch size thresholds. The objective of this work is to suggest new weed-mapping workflows for ground and drone spraying in wild blueberry fields using RGB drone imagery processed in drone-compatible GIS software: PIX4Dfields. Drone aerial imagery from two commercial wild blueberry fields in Maine was captured in May 2025. To create maps for ground spot spraying, PIX4Dfield’s machine learning-driven “Magic Tool” and five vegetation Index Tool workflows were assessed. For drone zone spraying, maps created with minimum weed patch sizes of 9.3 m2 and 405 m2 were evaluated. All maps were validated using image-based and ground-based accuracy assessments. The “Magic Tool” workflow proved most effective for ground spray maps, achieving ≤98% (image-based) and ≤86% (ground-based) overall accuracy. Among the vegetation indices, the Triangular Greenness Index performed best (≤93% and ≤85% image-based and ground-based accuracies, respectively) at a 20% to 25% pixel exclusion level. For drone zone spraying, the 9.3-m2 minimum weed patch size yielded higher accuracy than the 405-m2 size, although performance was site dependent. Classification accuracy decreased at sites with greater weed species diversity, more uniform weed distribution, and when evaluated against more reliable ground-based reference data. Nevertheless, the resulting maps identified weed coverage ranges of 39% to 52% and 58% to 61% across the two fields, giving growers a practical tool to reduce inputs and implement SSWM.
