Published October 10, 2025 | Version v1
Dataset Open

[Dataset] Non-invasive Honeybee Colony Monitoring via Robotic Mapping of Combs in Observation Hives

  • 1. Faculty of Electrical Engineering, Czech Technical University in Prague
  • 2. Artificial Life Lab, Department of Zoology, Institute of Biology, University of Graz
  • 3. University of Manchester
  • 4. Durham University, Computer Science Department
  • 5. Durham University

Description

Dataset Overview

This dataset was developed as part of the research study:

Janota et al., "Non-invasive Honeybee Colony Monitoring via Robotic Mapping of Combs in Observation Hives", Computers and Electronics in Agriculture, 2025. DOI: 10.1016/j.compag.2025.111031

Details regarding data collection procedures are available in the paper referenced above. Due to the large volume of data generated during the study, only part of the datasets is publicly released. The remaining datasets can be provided upon reasonable request. The datasets included in this release are:

  • Map Checkpoints: Sets of cells with belief values over considered states, along with a list of all integrated observations in the map.
  • Trained Models: Classification models for egg-laying, open brood, capped brood, and cell detection model.

  • Annotated Cell Sequences: A collection of 531 cell sequences containing individual cell images and corresponding annotations.

The code for the paper can be found at: https://gitlab.sensorbees.eu/janota/2025_cea_mapping.git


1. Map Checkpoints

The map folder contains an image of the final map, .npz files with the individual map checkpoints and a .csv file with all the integrated observations in the map (open cell, capped brood, egg-laying event).

Individual Map Checkpoints

The individual .npz files contain the following attributes:

Attribute Description
centers metric positions of the cells (in meters)
counts a number of detections of the cell
sizes estimated radius of the cell (in pixels)
last_prediction_timestamp UTC timestamp (seconds) of the last prediction step (default=-1 if the cell was not observed yet)
last_update_timestamp UTC timestamp (seconds) of the last update step with observation (default=-1 if the cell was not observed yet)
cell_states belief over the states (43-dimensional vector of values in range (0, 1))

List of all integrated observations

The .csv file with all integrated observations has the following columns:

Column name Description
cell_id ID of the cell in the map that the observation belongs to
det_id -2 for egg-laying obs., -1 for capped brood obs., >= 0 for open cell obs.
u_x  x-position in image (pixels) for capped brood obs., metric x-position (meters) for open cell obs.
v_y_conf y-position in image (pixels) for capped brood obs., metric y-position (meters) for open cell obs., position confidence for egg-laying obs.
timestamp UTC timestamp (seconds) of the obs.
obs confidence of open brood/capped brood/egg-laying classifier
bag_name identifier for the images (not relevant without raw data)
scan_id identifier for the images (not relevant without raw data)
img_id identifier for the images (not relevant without raw data)

 

2. Trained Models

We provide the following trained models in the folder models:

  • Cell detection (YOLOv5s6)
  • Open brood classification (ResNet-9)
  • Capped brood classification (ResNet-9)
  • Egg-laying event classification (ResNet-9)


3. Annotated Cell Image Sequences

We provide both the test and train cell sequence images in respective folders test_sequences_2024, train_sequences_2024. The folders contain subfolders with names denoting the ID of the cells. Each cell folder contains images (both open cell detections and occluded/capped cell images) and a .csv file with the following columns:

Column name Description
cell_id ID of the cell in the map that the observation belongs to
det_id -1 for occluded/capped brood observation, >= 0 for open cell observation
x x-position (pixels) of the cell in the corresponding raw image
y y-position (pixels) of the cell in the corresponding raw image
timestamp UTC timestamp (nanoseconds) when the corresponding image was taken
capped_brood 1 if observation is capped brood, 0 otherwise
labels -2 for capped brood, -1 for unknown, 0 for other, 1 for any open brood, 2 for egg, 3 for larva
bag_name  identifier for the images (not relevant without raw data)
scan_id  identifier for the images (not relevant without raw data)
img_id  identifier for the images (not relevant without raw data)

 

Citation

To attribute this dataset in your research, please cite the corresponding paper:

Janota et al., "Non-invasive Honeybee Colony Monitoring via Robotic Mapping of Combs in Observation Hives", Computers and Electronics in Agriculture, 2025. DOI: 10.1016/j.compag.2025.111031

Files

map.zip

Files (5.2 GB)

Name Size Download all
md5:0a68fe8127cc0b58de05dc3098cea841
196.2 MB Preview Download
md5:19237412d729d13ce0e6df6f21e774fd
28.3 MB Preview Download
md5:8db32ef3a65bf36413590aef73a5eeec
1.7 GB Preview Download
md5:e2f7a46454efbe6aa6cbcafe3627a612
3.3 GB Preview Download

Additional details

Related works

Is supplement to
Dataset: 10.1016/j.compag.2025.111031 (DOI)

Funding

European Commission
SENSORBEES - Sensorbees are ENhanced Self-ORganizing Bio-hybrids for Ecological and Environmental Surveillance 101130325
UK Research and Innovation
SENSORBEES: Sensorbees are ENhanced Self-ORganizing Bio-hybrids for Ecological and Environmental Surveillance 10109956
European Commission
RoboRoyale - ROBOtic Replicants for Optimizing the Yield by Augmenting Living Ecosystems 964492
Ministry of Education Youth and Sports
Robotics and advanced industrial production CZ.02.01.01/00/22 008/0004590
University of Graz
Field of Excellence COLIBRI
Czech Technical University in Prague
Student Grant SGS22/168/OHK3/3T/13

Dates

Collected
2024-08-15/2024-09-09