Bean yield estimation using unmanned aerial vehicle imagery

Abstract

The common bean is a crop of substantial socioeconomic importance that is cultivated throughout the Brazilian territory. Despite that, studies conducted so far have shown limitations in the methodologies used for yield estimation. In this sense, emerging technologies such as unmanned aerial vehicles (UAVs) can help both in crop monitoring and in assessing crop yield. Therefore, this study aimed: (1) to estimate the bean yield using spectral variables derived from UAV imagery and (2) to define the best vegetative stage for yield estimation. For this, data from a field experiment were used. The beans were planted in a conventional system in an area of 600 m2 (20 x 30 m). During the crop cycle, six flights were carried out using a UAV equipped with a five-band multispectral camera (Red, Green, Blue, Red Edge, and Near-infrared). After that, 10 spectral variables composed of the bands and five vegetation indices (VIs) were obtained. At the end of the season, the area was harvested, and the yield (kg ha-1) was determined. Then, the data was submitted to correlation (r), and regression analysis. Overall, all developed models showed moderate performance, but in accordance with the literature, with R2 and RMSE values ranging from 0.52 to 0.57 and from 252.79 to 208.84 kg ha-1, respectively. Regarding the best vegetative stage for yield estimation, the selected models used data from the second flight (52 days after planting) at the beginning of pod formation and filling (between stages R7 and R8).