Imagens multiespectrais a partir de um VANT (Veículo aéreo não tripulado) na estimativa da produtividade do trigo

Abstract

The general objective of the research was to understand the application of the usage of UAVs in wheat crop management. The first paper aimed to carry out a bibliometric literature search on the usage of UAVs in wheat cultivation. A search was executed within the Web of Science database of all scientific papers published until the year 2021. Subsequently, bibliometric literature analysis techniques were applied via the VOSviewer software, through which co-authorships between countries and institutions were evaluated, as well as the co-occurrence of words between the sources. The journals that publish the most on this topic were verified. The results indicated a growing trend of publications on the application of UAVs in the last 7 years, and that China, the United States of America and the United Kingdom are the largest researchers on this theme. China stands out with approximately 40% of the publications. This analysis reveals the main current issues and the most influential institutions around the world that have carried out relevant investigations and presented them in scientific publications. This study emphasizes the journals that include the most publications on the topic and the collaborative patterns related to the usage of UAVs in wheat crop. The platforms most used for this purpose are of the multirotor type, embedded with multispectral cameras. About 27.8% of the publications are related to the thematic axis linked to the monitoring of wheat productivity/phenotyping. The theme is in evidence, but it is needed further studies that focus on the application of drones in regions with high wheat production, such as the countries of South America. The second paper evaluated the performance of machine learning algorithms and multispectral aerial images, in the indirect estimation of wheat grain yield. Two sample areas with different growing seasons were considered, where several flights throughout the phenological cycle of the crop were performed. At the end of the crop cycle, grain yield (t ha-1 ) was determined. The supervised machine learning algorithms tested were: Linear regression (RL); Random forest (RF); Support vector machine (SVM); and Artificial neural network (ANN), combined with visible spectrum (RGB), multispectral and banded vegetation indexes. The RL algorithm, combined with the RGB indexes, displayed better performance in the initial stages of culture (tillering) and final (maturation) with R² = 0.61 and R² = 0.58. The SVM algorithm presented higher values in the rubberization – coming into ear phase, with the interaction of the red edge, red and green bands (R² = 0.78 and RMSE = 0.479 t ha-1 ). Therefore, it is feasible to use these variables and algorithms to determine wheat yield. Altogether, it was found that this is a viable, relatively new and promising application of the technology.