Consumer-grade UAV utilized for detecting and analyzing late-season weed spatial distribution patterns in commercial onion fields

Abstract

Studying weed spatial distribution patterns and implementing precise herbicide applications requires accurate weed mapping. In this study, a simple unmanned aerial vehicle (UAV) was utilized to survey 11 dry onion (Allium cepa L.) commercial fields to examine late-season weed classification and investigate weeds spatial pattern. In addition, orthomosaics were resampled to a coarser spatial resolution to simulate and examine the accuracy of weed mapping at different altitudes. Overall, 176 weed maps were generated and evaluated. Pixel and object-based image analyses were assessed, employing two supervised classification algorithms: Maximum Likelihood (ML) and Support Vector Machine (SVM). Classification processes resulted in highly accurate weed maps across all spatial resolutions tested. Weed maps contributed to three insights regarding the late-season weed spatial pattern in onion fields: 1) weed coverage varied significantly between fields, ranging from 1 to 79%; 2) weed coverage was similar within and between crop rows; and 3) weed pattern was patchy in all fields. The last finding, combined with the ability to map weeds using a low cost, off-the-shelf UAV, constitutes an important step in developing precise weed control management in onion fields.