dcskycam/helicopter-classification

DCSkyCam Helicopter Classification Dataset

This dataset contains cropped images of helicopters and non-helicopter objects captured by the DCSkyCam - a Raspberry Pi-based webcam in Washington, DC. The images were used to train binary and multi-class helicopter classification models using transfer learning from EfficientNet.

Dataset Description

The DCSkyCam system used a three-stage detection pipeline:

1. Object Detection (SSD MobileNet V3) identifies candidate objects in the sky
2. Binary Classifier determines if a candidate is a helicopter or not
3. Multi-class Classifier identifies the specific helicopter type

This dataset contains the cropped images used for stages 2 and 3.

Image Format

• Format: JPEG (cropped from full-frame)
• Typical size: Pixel sizes vary depending on MobileNet bounding box output. Most have at least 1 axis >= 100px.
• Total images: 8,100

Metadata Columns

Column	Description
file_name	Image filename (relative path within train/)
helicopter_type	"helicopter" or "not_helicopter"
heli_subtype	Helicopter type code (e.g., "UH1N", "AS350") or "none" for non-helicopters

Class Distribution

Helicopter Types: 5,153 images

Type	Count	Type	Count
UH1N	1,920	MH65	717
AS350	895	B412	451
H60	451	EC35	126
VH3D	194	VH92	125
S76	124	AW139	59
UH72	39	EC45	26
MD52	14	B206	5
AH6	2	V22	2

Non-helicopter: 2,947 images

Usage Notes

• The test/train/validation split is left to the user — no pre-defined splits are included.
• Data augmentation was applied during model training but is not reflected in this dataset. Users are free to apply their own augmentation.
• A small number of additional copyrighted images (<100) were used for training but are also excluded from this dataset for licensing reasons.

License

This dataset is licensed under the Creative Commons Attribution 4.0 International License.

Citation

@misc{dcskycam_classification_dataset,
  title = {DCSkyCam Helicopter Classification Dataset},
  author = {DCSkyCam contributors},
  year = {2026},
  url = {https://huggingface.co/datasets/dcskycam/helicopter-classification},
  license = {CC-BY-4.0}
}

Project Information

The DCSkyCam project was an AI-enabled sky monitoring system built on Raspberry Pi that automatically detected and identified helicopters using TensorFlow Lite models.