Deforestation areas are increasing yearly, and the current reforestation efforts are not enough to mitigate this, with alarming consequences such as altered water cycle, loss of biodiversity, and global warming. Therefore, the recent drone reforesting projects provide a new solution where faster reforestation can be done at a lower cost. However, drone reforesting has a low survival rate for the spread of seeds due to several factors such as depredation, soil quality, seed dormancy, and wrong spread area. This paper describes how to identify deforested areas from drone imagery using convolutional neural networks (CNN) to do multi-class semantic segmentation, so the drone can spread seeds on favorable sites and increase the seeds' survival rate. It also describes how forest regeneration can be monitored over the years by identifying the forest area in a drone image. In addition, the paper examines how to handle the dataset to obtain the best prediction model possible based on some training and validation metrics.
