Verificação de métodos e algoritmos em imagens advindas de ARP para a extração de parâmetros estruturais e morfológicos para a quantificação de árvores e parâmetros de copa, dossel e clareiras

Abstract

Combining the use of geotechnologies with forest monitoring is increasingly sought after in the forestry sector, as the search for tools that can provide high quality results quickly and non-destructively is seen as the most sustainable way to expand the monitoring of large forests in the Amazon. However, the use of remotely piloted aircraft (RPA) imagery, combined with the application of artificial intelligence algorithms, has shown promise in identifying factors that indicate changes in forests. This research aims to test methods and algorithms on ARP images to extract structural and morphological parameters for native forest monitoring, and to evaluate the quality of the results. In order to achieve it, aerial surveys were carried out in two areas of the Experimental Farm of the Federal University of Amazonas (FAEXP) that had undergone silvicultural treatments. From the generation of orthophotos and digital elevation models (DEMs), 2 tests were carried out: (1) evaluation of four classification algorithms (CART, RandomForest, Mahalanobis and GradienteTreeBoost) for automatic identification of gaps and (2) evaluation of the SAMGEO algorithm for automatic identification of palm samples. The analysis of the orthophotos combined with the DEM allowed a better evaluation and obtaining of metrics for the studied area. This analysis made it possible to evaluate metrics such as canopy profile, canopy area and changes in floristic composition. Furthermore, this joint analysis provided better answers to the identification of gaps. The results of the gap classifications showed that the RandomForest algorithm was the most suitable for identifying gaps in forest areas. Regarding the identification of palm trees, the SAMGEO algorithm proved to be efficient in identifying palm trees with an accuracy of 87.96%. However, it has limitations in terms of geometric delineation of the individual palm tree. These results contribute to research knowledge about the identification and development of remote sensing methods to improve the extraction of forest metrics in aerial photogrammetric products.