Identification of the onset of disease within a potato crop using a UAV equipped with un-modified and modified commercial off-the-shelf digital cameras

Abstract

The rapid development of unmanned aerial vehicles (UAV) has resulted in these aircraft being much easier to operate via the use of portable computers or phones, using fully automated flight paths and at a ready to fly price point that’s within the financial reach of most consumers. UAVs are potentially very useful tools for farmers as they allow an overhead view of crops and field boundaries and although they are typically only equipped with commercial off-the-shelf (COTS) digital cameras, recent photogrammetry techniques allow the creation of orthorectified visual data as well as a digital elevation model of the observed scene. This paper investigates the effectiveness of using a UAV with dual COTS cameras, one un-modified and one modified to sense near infra-red (NIR) wavelengths of light, in order to identify the onset of disease within a trial crop of potatoes. The trial was composed of 2 plots of 16 drills containing 12 tubers that had been exposed to the blackleg disease-causing bacterial pathogen Pectobacterium atrosepticum in order to demonstrate best practise tuber storage and haulm destruction methods. 11 sets of aerial data were gathered between 27/5/2016 ~ 29/7/2016 and then compared with ground truth data collected on 14/7/2016. Visual analysis of the data could only detect the onset of disease and not the specific infection and resulted in a users accuracy (UA) of 83% and producers accuracy (PA) of 78% in detecting the onset of disease, with a total accuracy (TA) of 91% and Kappa coefficient (K) of 0.75. The building blocks of an automated classification routine have been constructed using pixel and object based image analysis (OBIA) methods, which have shown promising first results (UA 65%, PA 73%, TA 87%, K 0.61) but requires further refinement to achieve an equivalent level of accuracy as that of the visual analysis.