Sensoriameno remoto com sensores de aeronaves remotamente pilotadas para aplicações de agricultura de precisão e gestão ambiental

Abstract

The objective of this research was to analyze whether the measurement of invasive species height (CHIS) through images captured by remotely piloted aircraft (drone) could classify with good assertiveness the areas with presence of invasive plants differenting from other areas with vegetation remnant of cerrado. The text was divided in two chapters, the first chapter highlights a first test of CHIS+GPS/GLONASS model, comparing to two other common techniques used in the classification of invasive species: the unsupervised classification k-means and vegetation index NGRDI. MDS and MDT elevation models were produced from the drone images collected in field and posteriorly then processed in PhotoScan software. The CHIS+GPS/GLONASS production was through the subtracting the MDS and MDT models. The comparison between the models occurred in two sample áreas. Some nonparametric statistical tests were used, such as: precision test, general error rate, specificity, sensitivity, Spearman correlation and Cohen's Kappa. In the second chapter, to measure the precision of CHIS+GPS/GLONASS model, the CHIS+RTK model was generated as the observed variable. The comparison between the models took place in the two sampled areas used in the first chapter, by which they were visually compared from graphs and statistical tests. The statistical tests used were: Spearman correlation coefficients (SCC), mean square root canopy height error (RMSEz), mean absolute canopy height error (MAEz) and Wilcoxon test. The results of the first chapter demonstrate that CHIS+GPS/GLONASS model has the best results in identifying invasive species when compared to the k-means and NGRDI models. Precision tests for the CHIS+GPS/GLONASS model in sample area 1 and 2 were 0.973 and 0.9, respectively; k-means 0.209 and 0.6; NGRDI 0.795 and 0.518. The results of the second chapter demonstrate that CHIS+GPS/GLONASS model presents faults in the identification of invasive species when compared to the CHIS+RTK model, being less accurate in the classification of invasive species selection heights. Spearman's correlation test for sample area 1 was 0.56 and 0.55 for sample area 2. RMSEz for sample area 1 was 0.17 cm and 0.12 cm for MAEz. A sample area 2 or RMSEz was 0.24 cm and 0.19 cm for the MAEz. The Wilcoxon test was significant for areas such as sample areas. The CHIS model proved to be a promising technique for identification of invasive species. It was exhibited with other accessory models, and its accuracy was considered good, since the largest difference in height errors was 0.24 cm and could be It is used in research that does not require great precision, as examples of research that has as object research, arboreal vegetation, where small differences in accuracy are not large.