Multitemporal weed mapping using UAV imagery for early site-specific control: the case of wheat as a narrow row crop.

Abstract

Obtaining weed patch maps for herbaceous crops in early season for site-specific weed control using remote sensing techniques has been a major challenge due to their spectral and appearance similarities. This is particularly problematic in the case of narrow row crops as wheat, where the weed discrimination has to be undertaken in a short time window for a timely postemergence control, and using images with better resolution than the usually provided by remote platforms such as satellite and conventional aircrafts. Nowadays, the utilization of ultra-high resolution images captured by Unmanned Aerial Vehicles (UAV) has opened the door to the generation of weeds and treatment maps. This article describes the complete workflow developed to achieve the weed patch mapping in a wheat field, as paradigm of narrow row crops. An UAV flying at different altitudes and dates was used in order to determine the best spatial resolution and wheat-weeds growth stage to successfully reach our objective. Main steps of the workflow are as follows: 1) configuration of the UAV flights to acquire a set of overlapped imagery; 2) mosaicking of these images to create a georeferenced ortho-image of the whole crop field; 3) automatic object-based image analysis (OBIA) procedure developed for generating weed patch maps. The UAV was equipped with a commercial camera which provided images in the visible range of the spectra. The vehicle was programmed to overflow automatically a wheat field naturally infested by a grass weed, and to trigger the camera at the moment required to supply images with a previously fixed overlapping between them. Then, overlapped images were mosaicked to create an accurate georeferenced orthoimage of the entire crop field. At last, the mosaicked image was analyzed using a robust and completely automatic OBIA procedure developed by our research group. The OBIA analysis algorithm combines object-based features such as spectral, position, orientation and hierarchical relationships, and consists of three consecutive phases: 1) discrimination of crop rows by application of a dynamic and auto-adaptive classification approach, 2) discrimination of weeds on the basis of their relative positions with reference to the crop rows, and 3) generation of a weed patch map in a grid structure for a further use in early sitespecific weed control. The effect of ortho-image spatial resolution (ranging from 1 cm to 3.5 cm) and wheat-weeds growth stage on the workflow performance using OBIA was studied for the different flight altitudes and dates.