Quantification of flowering in coffee growing with low-cost RGB sensor UAV-mounted

Abstract

The prediction of harvesting in terms of quantity, quality, and peak dates in a coffee crop is of utmost importance both for coffee growers and for the entities in charge of marketing at an international level. Such forecasts make the projection of crop maintenance, fertilization, and bean production costs possible, which has an impact on the social and economic benefits for the families that depend on this crop. In Colombia most of these are small producers. This research proposes a production projection methodology based on the traditional counting of flowers on several trees determined from the correlation with the area of flowers classified in Red-Green-Blue (RGB) sensor images captured with a low-cost sensor transported by a low-cost Unmanned aerial vehicle (UAV) system. The results have led to a model that allows correlating the area of flowers classified in the image with the number of flowers in the crop from which the production season can be programmed. The results obtained show a correlation of Sperman ρ=0.761(p−value<0.001). This allowed the development of an appropriate regression model. The object-based classification model based on the Watershed algorithm was applied for the segmentation of the floral area in the crop with an index of Kappa=0.984, which represents an excellent performance of the classifier. All developments were carried out using low-cost tools and free software, which allows for reducing costs for producers in terms of time, labor, and analysis.