RSICB: 800 Remote Sensing Images for Land Use Classification

A collection of 800 satellite and aerial images from locations around the world, preprocessed to 128x128 pixel resolution. The dataset covers 20 scene categories organized under a 2-level hierarchy with 6 super-categories, including agricultural land and construction land. It was created by Haifeng Li et al. and released as part of the Meta-Album benchmark in March 2022.

Use Cases

• Training land cover classification models based on the 20 scene categories.
• Benchmarking few-shot learning algorithms using the standardized Meta-Album format.
• Developing models to recognize hierarchical land-use categories based on the described 2-level label structure.
• Preprocessing and augmentation studies based on the 128x128 pixel image resolution.

Strengths

• Includes 800 images across 20 distinct classes, providing a defined scope for model training.
• Images are preprocessed to a consistent 128x128 pixel resolution using an anti-aliasing filter.
• Features a 2-level hierarchical label structure with 6 super-categories, adding semantic depth.

Limitations

• Row count is unknown, which may limit suitability assessment beyond the stated 800 images.
• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

RSI-CB: A Large Scale Remote Sensing Image Classification Benchmark via Crowdsource Data (GitHub: lehaifeng/RSI-CB)

Collection Method

Images were selected from various global locations and classified according to China's landuse classification standard.

Geography

Various locations around the world.