Tree Species Inventory Through Forest Segmentation and Deep Learning Techniques

Forest management holds paramount importance, necessitating the preservation of its ecosystems through comprehensive inventory and precise identification of tree species and their distributions. Recent advancements in deep learning have notably enhanced the accuracy of object detection and counting. This paper introduces a novel pipeline for segmenting forests into distinct plots and subsequently identifying the species within each plot. Leveraging the AirSim/Unreal Engine simulator, renowned for its provision of realistic physical and visual attributes, we present our approach. We employed a simple and computationally inexpensive method to prevent the counting of duplicate trees that overlap between two adjacent images. Our results demonstrate a robust capacity for tree identification and underscore the significance of precise forest segmentation.