Toke T. Høye

BioCluster uses foundation models to cluster and zero-shot label ecological images, achieving 84% and 83% species-level accuracy on moth and bird datasets.

InsectDCT detects, classifies, and tracks insects in camera-trap footage using deep learning, achieving high accuracy in complex environments.

This paper describes an open-source automated insect monitoring system using AI and camera traps to track nocturnal insects non-lethally and scalably.

flatbug, a YOLOv8 instance-segmentation model with multi-scale tile-pyramid inference and polygon-based deduplication, segments terrestrial arthropods across 23 image subdatasets and transfers to held-out subdatasets with a modest F1 penalty.

This paper presents a workflow to harmonize digitized herbarium data, resulting in a more accurate checklist of Greenland's flora with increased species richness and occurrence records.

This paper presents an image processing pipeline for analyzing insect camera trap data, enabling detection, tracking, and classification of nocturnal insects, with evaluation of edge computing devices for real-time processing.

This paper presents a deep learning pipeline to monitor arthropod abundance and floral cover using time-lapse camera data, showing a correlation between flower cover and arthropod activity.

This paper develops deep learning models to extract micrometeorological data from wildlife camera images, showing improved accuracy and revealing ecological insights.

This paper presents an algorithm for hierarchical insect classification using multitask learning and anomaly detection, improving accuracy at higher taxonomic ranks while identifying classification uncertainty.