Identifying Southern Pine Beetle Damage in High Spatial Resolution Drone Imagery Using Multiscale Image Segmentation and Random Forest Classification

Abstract

Southern pine beetles (Dendroctonus frontalisi) pose an existential ecological and economic threat to pine dominant forest communities, such as the Pine Barrens of Eastern Long Island. This research combines multiscale image segmentation methods, image texture metrics, and random forest classification to develop a novel approach for identification of individual trees damaged by southern pine beetles using ultra-high spatial resolution imagery. In 2022 an unmanned aerial system (UAS) collected 7cm imagery over a mixed canopy forest in Long Island, NY. This imagery was then processed using multi-scale image segmentation to identify individual tree crowns (ITC). These crowns were used to develop a variety of spectral and texture metrics as input for a random forest classification. Model performance metrics such overall accuracy (93.57%) and kappa (0.839) indicate a reasonable overall performance, while individual class metrics for the damaged class such as producers accuracy (75%) and users accuracy (90%) highlight potential for model refinement. This research represents one of several projects in a growing body of literature attempting to utilize relatively low cost multispectral remote sensing drones to map and monitor invasive insect species such as the southern pine beetle.

Keywords: Unmanned Aerial SystemsRemote SensingSouthern Pine BeetleRandom ForestMultiscale Image Segmentation