Recognition of arable soils from photographs obtained as part of crowdsourcing technologies

Abstract

The study focuses on the possibilities of using photographs obtained using crowdsourcing technologies for the operational inventory of arable soils. The object of the study is the spectral reflectance of the open surface of arable soils of the test plots, measured using a HandHeld-2 spectroradiometer operating in the range of 325–1 075 nm, and their image in photographs taken with conventional cameras. Test sites are located in the Tula, Moscow and Tver regions. The soils of the test plots are sod-podzolic, gray forest, and leached chernozems. Based on the analysis of photographs of the surface and information obtained using a spectroradiometer, a set of spectral parameters in the RGB, YMC and HSI color systems, as well as their ratios (45 parameters), was calculated. These parameters were used to separate the analyzed soil types using classification trees. The accuracy of classification based on the results of validation varies from 63–100%. At the same time, the parameters of the HSI and YMC color systems turned out to be more informative than the parameters of the RGB color system. The established classification rules can later be used to determine the classification position of soils from images collected using crowdsourcing technologies.

Keywords

• soil inventory
• crowdsourcing
• remote data
• decision trees