dataset_info:
features:
- name: audio
dtype: audio
- name: key
dtype: string
- name: user_obs_status_queen
dtype: string
- name: pred_status_queen
dtype: string
- name: pred_score
dtype: int64
- name: request_timestamp
dtype: timestamp[ns]
splits:
- name: train
num_bytes: 164022287
num_examples: 86
download_size: 163139798
dataset_size: 164022287
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
license: mit
task_categories:
- feature-extraction
- audio-classification
tags:
- audio
- classification
- biology
- beehive
size_categories:
- n<1K
Dataset Card for AI-Belha
Dataset Summary
AI-Belha is a dataset comprising audio recordings from beehives, collected to determine the presence and status of the queen bee. The dataset includes 86 mono WAV files, each approximately 60 seconds long and sampled at 16 kHz, totaling about 1 hour and 26 minutes of audio. Each recording is annotated with beekeeper observations and model predictions regarding the queen bee's status.
Supported Tasks and Leaderboards
- Audio Classification: Classify beehive audio recordings to determine the queen bee's status.
- Environmental Monitoring: Use acoustic data for non-invasive monitoring of beehive health.
Languages
Not applicable (non-verbal audio data).
Dataset Structure
Data Instances
Each instance in the dataset includes:
- audio: The WAV file containing beehive sounds.
- user_obs_status_queen: Beekeeper's observation of the queen bee's status, categorized as:
- unknown
- queen_present_original
- queen_absent
- queen_present_newly_accepted
- queen_present_rejected
- pred_status_queen: Model's predicted status of the queen bee.
- pred_score: Confidence score (0–100%) of the model's prediction.
- request_timestamp: Timestamp of when the audio was recorded.
Data Fields
| Field | Type | Description |
|---|---|---|
| audio | audio | WAV file of beehive sounds. |
| user_obs_status_queen | string | Beekeeper's observation of queen bee's status. |
| pred_status_queen | string | Model's predicted queen bee status. |
| pred_score | float32 | Confidence score of the prediction (0–100%). |
| request_timestamp | timestamp | Timestamp of audio recording. |
Dataset Creation
Curation Rationale
The dataset was created to facilitate research into non-invasive methods for monitoring beehive health, specifically focusing on detecting the presence and status of the queen bee through acoustic analysis.
Source Data
Initial Data Collection and Normalization
Audio samples were collected between April and December 2024 by beekeepers across various regions of Portugal. Recordings were made using smartphone microphones placed outside the beehives. All beekeeper information has been anonymized to comply with data privacy standards.
Annotations
Annotations include:
- Beekeeper observations regarding the queen bee's status.
- Model predictions generated by a fine-tuned YAMNet model.
Considerations for Using the Data
Social Impact
This dataset aims to support sustainable beekeeping practices by providing tools for non-invasive monitoring of hive health, potentially aiding in the conservation of bee populations.
Discussion of Biases
Potential biases may arise from:
- Variability in recording equipment and environmental conditions.
- Subjectivity in beekeeper observations.
- Limited to recordings from Portugal within a specific timeframe.
- Audio quality may vary due to differing recording devices and ambient noise.
Licensing Information
This dataset is released under the MIT License.