Empowering tree-scale monitoring over large areas: Individual tree delineation from high-resolution imagery

Accurate individual tree delineation (ITD) is essential for forest monitoring, biodiversity assessment, and ecological modeling. While remote sensing (RS) has significantly advanced forest ITD, challenges persist, especially in complex forest environments. The use of imagery data for ITD is compelling given the rapid increase in available high-resolution aerial and satellite imagery data, the increasing need for image-based analysis where reliable 3D data are unavailable, the widening gap between data supply and processing capabilities, and the limited validation of state-of-the-art (SOTA) methods across diverse real-world conditions. This study aims to advance ITD research by evaluating SOTA instance segmentation approaches, including both recently developed and established methods. 

Supervised approaches, both Machine Learning (ML) and Deep Learning (DL) methods, are data-driven and depend on high-quality datasets with extensive annotations. An image-based ITD dataset by a collaborative effort from numerous contributors worldwide has been built and been utilized in the ISPRS ITD Contest 2024.

The dataset covers 11 study sites across 9 countries including Australia, Canada, China, Germany, Kenya, Malaysia, Norway, Panama, and United States, spanning tropical, subtropical, and temperate climate zones. The dataset includes diverse forest types, such as rainforests, montane forests, moist forests, and savanna in tropics; evergreen broadleaf and mixed forests in subtropics; and broadleaf and mixed forests in temperate zone. In addition, the dataset includes both natural and urban forests. As the reference for the training and evaluation, 600,000 manually annotated ITC masks are provided for over 11,000 images. Annotations of the visible dominant and co-dominant individual tree crown (ITC) on images were conducted through visual interpretation and organized by the standardized MS COCO data format.

The dataset consists of training, validation, and testing sets. The training and validation sets include both images and labels for ITD model development. The testing set includes only images. The performance on the testing set can be evaluated by submitting the ITD results to the online benchmark platform, where the ranking and scores are displayed on the leaderboard.