Integrating UAV-based multispectral imagery and machine learning to classify glyphosate-resistant palmer amaranth and assess herbicide program efficacy in soybean

Palmer amaranth is one of the most troublesome and economically important weeds in soybean production, with widespread occurrence of glyphosate-resistant (GR) populations. This study aimed to map and classify GR Palmer amaranth using UAV-based multispectral imagery and evaluate the effects of herbicide programs [preemergence (PRE) only, postemergence (POST) only, and PRE followed by (fb) POST] on Palmer amaranth coverage (early- and late-season) in soybean field in central Louisiana in 2025. A multispectral drone was used to map GR Palmer amaranth, and images were processed into orthomosaic using Pix4DFields and georeferenced in ArcGIS Pro. Weed classification was performed using pixel-based and object-based approaches implemented in ENVI and Orfeo Toolbox (OTB) within QGIS, employing Random Forest (RF), Multilayer Perceptron (MLP), and Support Vector Machine (SVM) algorithms. Both classification methods exhibited high performance, with RF achieving the highest overall accuracy (99%). UAV-based assessments provided a spatiotemporally efficient method for early-season Palmer amaranth mapping and late-season evaluation of residual weed activity following herbicide applications. UAV-derived Palmer amaranth coverage showed moderate positive correlation with the ground-truth weed density. Herbicide programs significantly reduced UAV-derived Palmer amaranth coverage during early-season (P = 0.0004) and late-season (P = 0.0223) assessments. Among the evaluated programs, PRE fb POST resulted in the greatest reduction in weed coverage (4.3%), compared with PRE only (9.2%) and POST-only (8.8%) programs. In conclusion, integrating UAV-based multispectral imagery with machine learning algorithms, particularly RF, offers a promising, automated approach for classification of GR Palmer amaranth and evaluating herbicide program effectiveness across different crop growth stages.