Measuring species of lichen in forestrystands using detection algorithms
2025
Enare, Alexander
Hanging lichen plays a crucial role both as a winter food source for reindeer and as a sensitivebioindicator of environmental change, including rising nitrogen levels linked to climate change.Declines in lichen populations can serve as early warning signals of ecological imbalance.However, current methods for detecting hanging lichen species, such as Usnea spp. (US), Alec-toria sarmentosa (AS), and Bryoria spp. (BS), rely on ocular tools like relascopes, which intro-duce human bias and offer limited spatial insight. In this thesis, we propose a machine learningbased method for detecting and monitoring two key lichen groups, US and AS/BS, in forest en-vironments using image analysis. While machine learning has previously been applied to detectground lichen and berry shrubs, its application to hanging lichen represents new ground.We created a custom dataset from high-resolution forest imagery, as no suitable dataset existed.Using this dataset, we trained a You Only Look Once (YOLO) object detection model. Themodel achieved an F1-score of 0.79 for US and 0.62 for AS/BS, the latter affected by lowvisual contrast in complex forest scenes. To support practical deployment, we developed a userinterface (UI) for streamlined inference visualization and data export. We standardized imagecollection using a custom-built tool that captured images in the four cardinal directions.Our approach delivers a practical and reproducible alternative to manual surveys, laying thegroundwork for automated long-term lichen monitoring. Beyond lichen detection, the samesegmentation-based framework could be adapted to broader ecological applications, such asassessing fertilization impacts on tree growth and surrounding fauna. While this would requirecustomized models, the underlying pipeline remains transferable.
显示更多 [+] 显示较少 [-]