Satellite information fusion with low-altitude unmanned aerial vehicle flights for estimating weed coverage

Abstract

The detection, geolocation, and classification of weeds in agricultural fields is a problem of interest associated with Precision Agriculture (PA). The main contribution of this work is to describe a workflow (feasible to automate) based on open-source software tools and open information to: 1) measure the spatiotemporal evolution of weed patches through satellite images, and 2) register high-resolution images (taken at low altitude) on top of the satellite image to identify the weeds that compose the detected patches. To merge the satellite and low-altitude information, the following problems must be solved: 1) correct distortions in the acquired images; 2) develop an image formation model that allows registering the low-altitude image on top of the satellite image, and 3) analyze green indices to measure patch coverage in both multiespectral satellite images and RGB images obtained from a camera mounted on an unmanned aerial vehicle. Finally, the feasibility of merging information is demonstrated through an analysis of the correlation in the coverage measures obtained from satellite and low-altitude images.