Tree Counting with Deep Learning Algorithm and Carbon Evaluation using Pleiades Satellite Imagery Data in Kulon Progo, Yogyakarta, Indonesia

Trees are important living creatures on earth where their existence is very important for the health of the Earth’s atmosphere, for the exchange of oxygen and carbon dioxide, etc. That’s why the number of existing trees must be maintained as much as possible. In this case, there are many ways to maintain the existence of the tree. To maintain the existence of trees, there are several ways to maintain them such as counting or mapping the trees that exist in the area. Since it’s too hard to count the trees by hand, luckily some algorithms can be used to count the trees themselves, which can help to count or map the trees, known as deep forest algorithms. The deep forest algorithm is a package in python that can be used to train, predict, and evaluate the individual tree crowns from aerial RGB imagery using a deep learning network for object detection. The reason for using the algorithm itself is that, as of 2019, there are not many tree counting experiments using this algorithm, only 14 experiments based on the official website. Not only tree counting as the goal of this experiment, evaluating carbon for every detected tree itself can also help so that carbon can be maintained and not disposed of as a hazardous substance. The main purpose of conserving the carbon itself is to protect the environment from sustainable development. As for the result of this experiment, the trees detected by deep learning deep forest algorithms successfully detected 613 trees from a total of 788 trees, making it an accuracy of 77.97% in the Kulon Progo area. As the carbon evaluation from counting all parts of an oil palm tree, know that for single biomass of oil palm tree is 354,8641 kg/ tree and it makes a total of 613 detected trees by Deep Forest Algorithms is 217531.6933 kg carbon in the study area.