Multi-Spectral Remote Sensing Image Retrieval Using Geospatial Foundation Models

Image retrieval enables an efficient search through vast amounts of satelliteimagery and returns similar images to a query. Deep learning models canidentify images across various semantic concepts without the need forannotations. This work proposes to use Geospatial Foundation Models, likePrithvi, for remote sensing image retrieval with multiple benefits: i) themodels encode multi-spectral satellite data and ii) generalize without furtherfine-tuning. We introduce two datasets to the retrieval task and observe astrong performance: Prithvi processes six bands and achieves a mean AveragePrecision of 97.62% on BigEarthNet-43 and 44.51% on ForestNet-12,outperforming other RGB-based models. Further, we evaluate three compressionmethods with binarized embeddings balancing retrieval speed and accuracy. Theymatch the retrieval speed of much shorter hash codes while maintaining the sameaccuracy as floating-point embeddings but with a 32-fold compression. The codeis available at https://github.com/IBM/remote-sensing-image-retrieval.