Deep learning for the identification of tree pests, pathogens, and disease symptoms: A systematic review

Abstract

The increasing availability of high-resolution imagery and advances in deep learning have created new opportunities for the automated assessment of tree health. This systematic review synthesizes current research on the use of deep learning to identify pests, pathogens, and disease symptoms in individual trees, with a focus on long-rotation forest trees. Following the PRISMA framework, we reviewed 92 English-language journal articles identified through a systematic search of the Web of Science database. The studies were analyzed with regard to the tree health problems investigated, imaging technologies, annotation and preprocessing procedures, and the deep learning methods used for classification, detection, and segmentation. The literature is strongly dominated by UAV-based imagery, with pine wilt disease representing a particularly prominent research focus, especially in Southeast China. Object detection is the most commonly addressed task, and YOLO-based architectures are the most frequently used models. Despite the rapid growth of this research area, several limitations remain. The biological relationship between host trees and pathogens is rarely considered explicitly, while the generalizability of models across tree species, geographic regions, and acquisition conditions is often insufficiently evaluated. Comparisons between studies are further complicated by the lack of standardized datasets and evaluation protocols. Moreover, the strong emphasis on UAV-based remote sensing leaves other potentially valuable data sources, including in-situ observations and citizen science, comparatively underexplored. Overall, the findings highlight the need for more diverse datasets, standardized evaluation procedures, and greater consideration of biological and environmental variability to develop robust and transferable deep learning approaches for tree health monitoring.