Development of a high-accuracy autonomous sensing system for a field scouting robot

We developed a sensing system that can automatically acquire spectral reflection data using a hyperspectral camera with the aim of developing a field scouting robot. In order to deal with light-source fluctuations that occur during the daytime, we devised a sensing system equipped with artificial light that could be used at night. The fractional vegetation cover (FVC) and the soil plant analysis development (SPAD) value were estimated by calculating the normalized difference vegetation index (NDVI) from the spectral reflection data, and the accuracy was verified by comparing the results with the results obtained during the daytime. When segmenting the vegetation using NDVI and calculating the FVC, the accuracy was 1.4% higher at night than in the day. There was a correlation between the FVC and the measured SPAD value with a coefficient of determination of 0.51 at night. For the regression of the NDVI and the SPAD value of the manually extracted vegetation, the coefficient of determination was 0.85 at night and 0.77 in the day. As a result of sensing specific leaves every hour during the day, the difference in the SPAD value estimated from the NDVI was up to 7.9 SPAD units due to light fluctuation. In the regression of the NDVI and the SPAD value of the vegetation automatically segmented by NDVI, the coefficient of determination did not change from 0.85 during the night, but decreased to 0.71 during the day because of the poor accuracy of the segmentation. The combination of NDVI wavelengths for identifying vegetation and the combination for estimating the SPAD value were almost the same during the night, but not during the day. Since the light environment is constant during nighttime sensing, vegetation information such as the FVC and the SPAD value can be estimated accurately. The method developed in this study, which does not require correction using a standard whiteboard, is suitable for autonomous sensing systems in field scouting robots because this method can minimize the need for human intervention during crop sensing.