Heliyon (Oct 2021)
Using unmanned aerial systems and deep learning for agriculture mapping in Dubai
Abstract
As part of the sustainable future vision, sustainable agriculture has become an essential pillar of the food security strategies formulated by the Dubai Government. Therefore, the Dubai Emirate began relying on new technology to increase productivity and efficiency. Agriculture applications also depend on accurate land monitoring for timely food security control and support actions. However, traditional monitoring requires field surveys to be performed by experts, which is costly, slow, and rare. Agriculture monitoring systems must be furnished with sustainable land use monitoring solutions, starting with remote sensing using drone surveys for affordable, efficient, and time-sensitive agriculture mapping. Hence, the Dubai Municipality is currently using Unmanned Aerial Vehicles (UAVs) to map the farming areas all over the Emirate, support locating lands conducive to cultivation, and create an accurate agriculture database contributing to the decision-making process in determining areas suitable for crop growth. This study used a novel object detection method coupled with geospatial analysis as an integrated workflow to detect individual crops. The UAV flights were executed using a Trimble UX5 (HP) over twelve communities across the Dubai Emirate for six months. Detection methods were applied to high-resolution drone images, consisting of RGB and near-infrared (NIR) bands. Advanced geoprocessing tools were also used to analyze, evaluate, and enhance the results. The performance of detection of the selected deep learning models are discussed (vegetation cover accuracy = 85.4%, F1-scores for date palms and ghaf trees = 96.03% and 94.54% respectively, with respect to visual interpretation ground truth); moreover, sample images from the datasets are used for demonstrations. The main aim is to offer specialists a solution for measuring and assessing living green vegetation cover derived from the processed images that is integrated. The results provide insight into using UAS and deep learning algorithms as a solution for sustainable agricultural mapping on a large scale.
