Automatic detection of snow breakage at single tree level using YOLOv5 applied to UAV imagery

Stefano Puliti, Rasmus Astrup

Abstract: Highlights•Trained a YOLOv5 object-detection model for forest snow damage detection.•Across all classes, the model had a precision of 78% and a recall of 70%•The model had best accuracy for damaged trees despite this being a minority class.•While well transferable across seasons, the best performance was for summer data.•The model was robust to weather and sun elevation angle variations.