Methodological approaches to identifying plants on high resolution images of multispectral monitoring using UAVs

Abstract

Plant growers need accessible and effective information about the state of crops to implement crop management. The purpose of the study is to develop a method for identifying plants on high-resolution multispectral images for continuous sowing crops, using the example of winter wheat. The studies are conducted in the Left-Bank Forest-Steppe zone, on industrial crops of winter wheat, Mulan variety. At the time of remote monitoring through UAVs (2019.03.17), the plants were in the tillering stage. Monitoring from an altitude of 100 meters is conducted using the Slantrange 3p spectral system installed on the DJI Matrice 600 UAV. A full-screen copy of the snapshot window is made to extract reference graphic data from the SlantView programme. Statistical processing of graphical data of spectral monitoring results is performed in the MathCad programme. It is noted that reliable determination of the spectral portrait of the soil for its pixel filtration from multispectral images is a difficult task, since its colour substantially depends on the state of moisture and may differ in open and shaded areas. A fundamentally new way to filter out random inclusions is to use a spectral portrait of plants based on the intensity ratios of their components. A promising parameter for assessing the condition of crops is the estimation of their horizontal surface area, which can be determined by pixel-by-pixel image analysis. A filtering option that requires debugging is suggested. In further studies, it is advisable to consider the issue of methodological support for assessing the quality of filtering data from spectral monitoring of plantings