Meta-Album APL Extended: 9,625 Satellite Images of 21 Aircraft Types

9,625 remote sensing images of airplanes, sourced from Google Earth satellite imagery and labeled by seven remote sensing specialists. The dataset includes 21 different aircraft types from around 36 airports, with each image cropped and resized to 128x128 pixels. It was created by Wu, Zhize in 2019 and later adapted for the Meta-Album benchmark in 2022.

Use Cases

• Training few-shot image classification models based on the 21 distinct aircraft classes.
• Benchmarking meta-learning algorithms on a standardized, preprocessed remote sensing dataset.
• Developing object detection or recognition models for aircraft in satellite imagery.
• Studying domain adaptation for computer vision tasks between satellite and other image domains.

Strengths

• Contains 9,625 labeled images, providing a substantial sample size for model training.
• Images were carefully labeled by seven specialists in remote sensing, suggesting high annotation quality.
• Includes 21 distinct aircraft classes, offering diversity for classification tasks.
• All images are preprocessed to a uniform 128x128 pixel size, ensuring consistency.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The original variable resolutions have been standardized, which may have removed some detail.
• Data may reflect geographic bias inherent to the selection of the 36 source airports.

Provenance

Source

Muti-type Aircraft of Remote Sensing Images (MTARSI) from Zenodo, created by Wu, Zhize.

Collection Method

Images were acquired from Google Earth satellite imagery and manually labeled by specialists.

Geography

Images were acquired from airports around the world, but specific locations are not detailed.