OBJECT-BASED APPROACH FOR CROP ROW CHARACTERIZATION IN UAV IMAGES FOR SITE-SPECIFIC WEED MANAGEMENT

A list of color-infrared images captured from the new generation of remote platforms known as unmanned aerial vehicles (UAV), specifically a quadrotor, was tested for site-specific weed management applications. The aim was to identify and classify the crop rows within a maize crop-field, with the ultimate objective of distinguishing small weed seedlings at early stages for in-season site- specific herbicide treatment. An object-based image analysis (OBIA) procedure was developed by combining several scene, contextual, hierarchical and object-based features in a looping structure. The procedure integrates several features from the crop-field patterns: 1) field structure, such as field limits and row length, 2) crop patterns, such as row orientation and inter-row distance, and 3) plant (crop and weeds) characteristics, such as spectral properties (NDVI values) and plant dimensions; as well as 4) hierarchical relationships based on different segmentation scales, and 5) neighboring relationships based on distance, position and angle between objects. The algorithm identified and counted the rows with 100% accuracy in most of the images and the definition of the longitudinal border of the crop rows was successful with 90% of overall accuracy, comparing to on-ground measures of weed emergence.