Remote monitoring has become an important tool for recognizing land and ground objects through sensor data analysis. The use of Machine Learning (ML) algorithms for classification of remote monitoring images has increased in recent years. ML-based image classifiers have played an important role in detecting deforestation, illegal mining or fire. However, the precise classification of land use is a huge challenging task, especially in remote tropical regions, due to the complex biophysical environment and the limitations of the remote monitoring infrastructure. This work aims at studying the trade-offs between performance and accuracy of classification systems for the Brazilian Amazon rainforest, taking into account different computing platforms (server and edge), ML algorithms and images sizes. Although there is a direct relationship between image accuracy and quality, our experimental study shows that it is possible to use low-cost computational environments to perform image classification. The results indicate that Amazon rainforest can be monitored with affordable computing resources such as drones.