InsectDCT

Intended use. YOLO models for processing time-lapse images from insect camera traps. It contains code to detect, classify, and track insects with various backgrounds of plants and flowers.

Architecture and training

• Architecture: YOLOv11
• Task Object tracking.
• Trained on: several datasets of insects on (semi-)natural backgrounds, mainly flowers and plants.
• Taxonomic coverage: mainly pollinators, see the Zenodo record for the class list.
• Framework: Ultralytics YOLOv11

Inputs and outputs

• Input size: author reports that images are resized to 1920x1080 for detection with YOLOv11, while insect crops are resized to either 128x128 or 224x224, depending on the classification model.
• Channels: standard RGB, pipeline also extracts motion-enhanced image representation to improve detection (done with YOLOv11).
• Outputs: classification/box (-CL.csv) and hierarchical-detail (-HI.csv) tables; see the verified profile below.
• Input mode: standalone or time-lapse images, or video.

The documented InsectDCT CSV profile describes the full detector/classifier pipeline at revision e459ae8, using the default CSVformat=tracking and ConvNextBase V6 classifier. The retained CPU probe processed 35 consecutive author frames and produced 40 rows each in the -CL.csv classification/box table and -HI.csv hierarchical-detail table. These tables contain classifier confidence values, not the native YOLO detector output. Tracking is a subsequent stage.

Performance

Multiple, complex metrics reported in the pre-print…

Limitations

Lifted from the pre-print discussion section:

The taxonomic classifier is currently trained on Northern and Central European
pollinator taxa, with a large part of the dataset focused on vegetation dominated by plant species of the *Sedum* genus, which implies that the pipeline will require re-training for direct transferability to other geographic regions or habitat and vegetation types. Researchers deploying the pipeline in new contexts are encouraged to supplement the existing dataset with locally collected and annotated images, following the incremental re-training workflow described above, which has been specifically designed to accommodate this kind of extension. The classes with few training samples, such as rare solitary bees and some folded-wing Lepidoptera, show reduced classification accuracy, and this imbalance could be mitigated in future versions by re-training the algorithm by users who require improved performance in those insect groups. Regarding the tracking algorithm, it can produce fragmented or merged trajectories when small insects cross paths or are temporarily occluded, and tracks should therefore be interpreted as activity proxies rather than records
of unique individuals over extended periods.

How to obtain the weights

From the Github repository insectDCT, all the various model versions trained weights are in run/detect, saved as .pt checkpoint files.

Citation

@article {Bjerge2026.07.07.736939,
	author = {Bjerge, Kim and Wogram, Simon F. A. and Serra-Marin, Pau Enric and Sakhiashvili, Otar and H{\o}ye, Toke T.},
	title = {InsectDCT: A generalized pipeline for detection, taxonomic classification, and tracking of insects in camera-trap recordings},
	elocation-id = {2026.07.07.736939},
	year = {2026},
	doi = {10.64898/2026.07.07.736939},
	publisher = {Cold Spring Harbor Laboratory},
	abstract = {Automated monitoring of insect pollinators in natural environments with insect camera traps and trained deep learning algorithms provides novel data for insect ecological studies. However, efficient and accurate image recognition analysis of the recorded images or videos is challenging, particularly for images containing small insects against complex backgrounds with diverse vegetation communities. Even when insects can be detected in images, identifying their taxonomy remains difficult, particularly in footage with low image resolution, light conditions, and distances from the plants, and in cases where insects appear blurry or only partially visible.In this work, we present InsectDCT, an AI-based pipeline for automated detection, hierarchical classification, and tracking of insects in footage of natural vegetation tested in different environments. The InsectDCT pipeline consists of three levels: insect Detection and localization, hierarchical taxonomic Classification, and spatio-temporal Tracking. In the first stage, insects are detected in time-lapse images or video recordings using the You Only Look Once (YOLO11) object detection architecture. Detection performance is improved using motion-enhanced images, which improve robustness in cluttered and 3 dimensional environments. The detector is trained on an extensive dataset that contains more than 60,000 images collected using camera traps deployed across a wide range of plant families and floral habitats. In the second stage, detected insects are classified using a hierarchical taxonomy-aware classification framework that covers 80 taxonomic groups. Classification is performed at multiple taxonomic levels, including order, family, and genus/species, allowing coarse and fine-grained ecological analyzes while accounting for varying levels of visual ambiguity. In the third stage, a multi-object tracking module is applied to high temporal-resolution image sequences and video data to associate detections of the same individual across time. InsectDCT code and all datasets are made publicly available.Competing Interest StatementThe authors have declared no competing interest.},
	URL = {https://www.biorxiv.org/content/early/2026/07/18/2026.07.07.736939},
	eprint = {https://www.biorxiv.org/content/early/2026/07/18/2026.07.07.736939.full.pdf},
	journal = {bioRxiv}
}