Literature Review of Tree Crown Detection Based on Machine Learning and Remote Sensing Data

Forest monitoring is the first step to evaluating biodiversity distribution, biomass, carbon stocks, and health of the ecosystem. However, the traditional field survey remains labour-intensive and impractical for large-scale studies. The latest improvements in UAV-based remote sensing such as LiDAR, multispectral, hyperspectral, and RGB imagery associated with deep learning methodologies has shown feasibility of automatic Individual Tree Crown Detection (ITCD) with high precision and high resolution. The current review has summarized remote sensing platforms, workflow for data preprocessing, methods of crown delineation, tree species classification, and forest health application based on models of Mask R-CNN, YOLO, Faster R-CNN, Vision Transformers, and semi-supervised models. The importance of data fusion and the structural spectral combination for improving ITCD is highlighted. There are many issues remaining to address like annotation uncertainty, overlapping crown cover, class imbalanced dataset and limited applicability in real-time scenario. An integrated model with multiple functions and embedding into the process-driven work is proposed as a promising approach for future au