VHR-10: Ten-Class Geospatial Object Detection in Satellite Imagery

NWPU VHR-10 is a ten-class geospatial object detection dataset containing 800 very-high-resolution optical remote sensing images. The dataset was created by the Northwestern Polytechnical University, with 715 color images from Google Earth and 85 pansharpened color infrared images from Vaihingen data. It was mirrored to Hugging Face by the user 'satellite-image-deep-learning' in May 2024.

Use Cases

• Training object detection models for ten specific geospatial object classes based on the dataset description.
• Benchmarking model performance on very-high-resolution optical imagery with spatial resolutions from 0.08 to 2 meters.
• Developing algorithms for detecting objects in both color and color infrared remote sensing images.
• Researching domain adaptation between different satellite image sources (Google Earth and Vaihingen data).

Strengths

• Contains 800 annotated very-high-resolution remote sensing images.
• Includes imagery from two distinct sources with spatial resolutions ranging from 0.5-2 meters and 0.08 meters.
• Focuses on a challenging ten-class object detection task for geospatial analysis.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset is divided into two parts, but the split details are not provided in the available metadata.

Provenance

Source

Northwestern Polytechnical University (NWPU), mirrored from GitHub.

Collection Method

Images were acquired from Google Earth and the Vaihingen dataset.

Freshness

Last updated 2024-05-28 07:58:33; freshness should be verified.