Remotely Piloted Aircraft Systems (RPAS) and machine learning: A review in the context of forest science

The combination of machine learning and products obtained by Remote Piloted Aircraft Systems (RPAS) has stood out in the forest sector in the last decade due to the wide range of applications and contributions to the classification and modelling of spatial attributes. The literature presents an overview of these applications, their limitations, and the perspectives related to the theme. Many applications are promising and have different approaches in practices such as identifying rare or invasive trees species, quantifying biomass, early detection of diseases, monitoring deforestation and forest fires, among others. In this context of large data sets generated by RPAS surveys, observed significant influence of deep learning methods, mainly due to their ability to automatically extract spatial characteristics and design flexibility, as in the case of different Neural Networks (NN) architectures. Other algorithms such as Random Forest (RF) and Support Vector Machine (SVM) also occupy a prominent position in this context, having a proven ability to deal with complex problems. However, reduction of computational load, consolidation of analysis protocols, automated selection of predictors, use of proprietary software, and expansion of the research scope are some of the challenges that still need to be overcome to disseminate and improve the use of RPA data sets. Thus, computer vision through deep learning networks emerges as a promising approach in object-based image analysis, considering its ability to recover patterns, even when operating in less complex databases, such as RGB compositions.