Weed {Mapping} in {Early-Season} {Maize} {Fields} {Using} {Object-Based} {Analysis} of {Unmanned} {Aerial} {Vehicle} {(UAV)} Images

Abstract

The use of remote imagery captured by unmanned aerialvehicles (UAV) has tremendous potential for designing detailedsite-specific weed control treatments in early post-emergence,which have not possible previously with conventional airborne orsatellite images. A robust and entirely automatic object-basedimage analysis (OBIA) procedure was developed on a series of UAVimages using a six-band multispectral camera (visible andnear-infrared range) with the ultimate objective of generating aweed map in an experimental maize field in Spain. The OBIAprocedure combines several contextual, hierarchical andobject-based features and consists of three consecutive phases: 1)classification of crop rows by application of a dynamic andauto-adaptive classification approach, 2) discrimination of cropsand weeds on the basis of their relative positions with referenceto the crop rows, and 3) generation of a weed infestation map in agrid structure. The estimation of weed coverage from the imageanalysis yielded satisfactory results. The relationship ofestimated versus observed weed densities had a coefficient ofdetermination of r(2)=0.89 and a root mean square error of 0.02. Amap of three categories of weed coverage was produced with 86% ofoverall accuracy. In the experimental field, the area free ofweeds was 23%, and the area with low weed coverage (<5% weeds) was47%, which indicated a high potential for reducing herbicideapplication or other weed operations. The OBIA procedure computesmultiple data and statistics derived from the classificationoutputs, which permits calculation of herbicide requirements andestimation of the overall cost of weed management operations inadvance.