Remotely Piloted Aircraft System for monitoring coffee crops

Abstract

Coffee is a commodity of great importance for the Brazilian economic balance and the state of Minas Gerais stands out with the highest production and exports worldwide. New techniques and technologies are applied in this agricultural sector, seeking practical gains in productivity and profitability of crops, combined with environmental gains. In this sense, Remotely Piloted Aircraft Systems (RPAS) are used as platforms for Remote Sensing (SR) for monitoring crops, combined with machine learning tools, they make it possible to identify problems that can be solved through adequate and efficient management. Given this scenario, this thesis analyzed the potential of using RPAS as aerial imaging technology in coffee plantations through scientific studies. (I) In the first study, a bibliometric survey was proposed, contextualizing the state of the art on the topic of RPAS in coffee farming, based on 20 years of searches in the most relevant databases, highlighting the temporal evolution of publications, performance analysis grouping the main publications, main journals, main researchers, main institutions, main countries and scientific co-citation mapping, keywords, trends and future possibilities on the research topic. (II) The second study aimed to classify and map the area occupied by weeds in coffee growing areas, determine the percentage of area occupied, and indicate treatment control strategies to be adopted in the field. To this end, two machine learning algorithms (Random Forest - RF and Support Vector Machine - SVM) were tested to classify regions of interest due to spectral differences between targets, highlighting the RF with the best classification performance. Furthermore, the savings obtained by treating only areas with the presence of weeds compared to treating the entire study area was approximately 92.68%. (III) The third study related parameters derived from aerial images based on different vegetation indices (VIs) and the canopy height model (CHM) to soil compaction in a coffee plantation area. Data collection was carried out on plant height, soil characterization, resistance to soil penetration, in situ productivity, and VIs calculated by aerial images. The multispectral data correlated with the penetration resistance data, making it possible to determine the NDRE and MTCI VIs with better estimation performance. This way, the possibility of monitoring coffee crop height variations using RPAS to demarcate compacted areas was highlighted. (IV) The fourth study aimed to classify and differentiate, using a machine learning algorithm (Random Forest – RF) coffee plants subjected and not subjected to foliar application of the biostimulant chitosan, indicating a valid approach to model the presence of the biostimulant in coffee plants, confirming that the model can assist in precision agriculture practices efficiently. (V) The fifth study compared coffee plant height data obtained from RGB/SfM point clouds and LiDAR data collected by RPAS, and estimated soil compaction through penetration resistance in a coffee plantation. No statistically significant differences were identified between the sensors used, and accurate estimation indirectly of soil compaction via remote sensing was demonstrated.