A Learnable Joint Spatial and Spectral Transformation for High Resolution Remote Sensing Image Retrieval

Geometric and spectral distortions of remote sensing images are key obstacles for deep learning-based supervised classification retrieval, which worsened by cross-dataset applications. A learnable geometric transformation model imbedded in a learning has been used as tool handling to process close-range with different view angles. However, model, is more noteworthy image processing, not yet designed explored up now. In this paper, we propose joint spatial (JSST) retrieval (RSIR), composed three modules: parameter generation network (PGN); conversion module; module. The PGN adaptively learns the parameters simultaneously from input content, these then guide conversions produce new modified correction. Our JSST front-end deep-learning-based network. spectral-modified inputs provided endow better generalization adaptation ability RSIR. experiments on four open-source RSIR datasets confirmed that our proposed embedded outperformed state-of-the-art approaches comprehensively.