RSICB: Remote Sensing Image Classification Benchmark with 45 Scene Categories

RSICB128 dataset contains 36,707 satellite and aerial images of 128x128 pixel resolution. The dataset covers 45 scene categories organized into a 2-level hierarchy with 6 super-categories, following China's landuse classification standard. It was created by Haifeng Li et al. and preprocessed for Meta-Album by Phan Anh Vu in March 2022.

Use Cases

• Training scene classification models based on 45 distinct land use categories.
• Developing hierarchical classification systems based on the 6 super-categories described.
• Benchmarking image classification algorithms on a standardized remote sensing dataset.
• Studying land cover patterns based on images from various locations around the world.

Strengths

• Contains 36,707 images, providing a substantial volume for training.
• Offers a structured 2-level label hierarchy with 45 classes and 6 super-categories.
• Images are preprocessed to a uniform 128x128 pixel resolution.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect geographic bias inherent to the source collection.

Provenance

Source

RSI-CB: A Large Scale Remote Sensing Image Classification Benchmark via Crowdsource Data

Collection Method

Images were assembled from various locations worldwide, following China's landuse classification standard.

Geography

Various locations around the world.