COMPARING PIXEL- AND OBJECT- BASED FOREST CANOPY GAPS CLASSIFICATION USING LOW-COST UNMANNED AERIAL VEHICLE IMAGERY

Forest canopy gaps are an important indicator of ecosystem dynamics. Gap sizes can vary because of several agents, and the spatial distribution is related to abiotic factors. The interest in the study of this forest attribute is old, but the difficulties to detect these areas in situ and with the use of satellite imagery hinder this research approach. Thus, we explore the use of high spatial resolution images obtained with RGB boarded in a multirotor unmanned aerial vehicle (UAV) to evaluate the best method to mapping the forest canopy gaps in Brazil. For this, were utilized the pixel- and object-based approaches, and the algorithms Random Forest (RF) and Support Vector Machine (SVM). The results showed that the ortophotomosaics can overcome the disadvantages of study the forest canopy gaps from conventional methods and reduce the complexity and costs to obtain reliable data of forests remnants. The RF and the pixel-based classification were the best combinations, with an overall accuracy (OA) of 93% in the period of study. However, the SVM presented a satisfactory accuracy to classify the forest canopy gaps, with the precision of user (PU) ranging from 86% to 98% and measure F from 85% to 96%. Therefore, was confirmed the potential of low-cost UAVs boarded with RGB sensors in this research proposal, and the results are promising for future studies.