Estimation of Damaged Regions by the Bark Beetle in a Mexican Forest Using UAV Images and Deep Learning

Abstract

Sustainable forestry for the management of forest resources is more important today than ever before because keeping forests healthy has an impact on human health. Recent advances in Unmanned Aerial Vehicles (UAVs), computer vision, and Deep Learning (DL) models make remote sensing for Forest Insect Pest and Disease (FIPD) possible. In this work, a UAV-based remote sensing process, computer vision, and a Deep Learning framework are used to automatically and efficiently detect and map areas damaged by bark beetles in a Mexican forest located in the Hidalgo State. First, the image dataset of the region of interest (ROI) is acquired by a UAV open hardware platform. To determine healthy trees, we use the tree crown detection prebuilt Deepforest model, and the trees diseased by pests are recognized using YOLOv5. To map the area of the damaged region, we propose a method based on morphological image operations. The system generates a comprehensive report detailing the location of affected zones, the total area of the damaged regions, GPS co-ordinates, and both healthy and damaged tree locations. The overall accuracy rates were 88% and 90%, respectively. The results obtained from a total area of 8.2743 ha revealed that 16.8% of the surface was affected and, of the 455 trees evaluated, 34.95% were damaged. These findings provide evidence of a fast and reliable tool for the early evaluation of bark beetle impact, which could be expanded to other tree and insect species.