Answers are grounded in this paper and cite their sources.
Extreme binary compression encoding is essential for achieving efficient, large-scale, and fast retrieval of remote sensing (RS) images. In this study, we introduce a condensed and discriminative feature encoding (CDFE) designed for RS image retrieval. Specifically, the visual mamba schema is used to perform feature extraction on RS images, providing subsequent binary encoding with representative and salient features that preserve crucial semantic information. Due to the critical role of feature polarity in binary encoding, i.e. positive and negative values, we design a discriminative attention module (DAM) to adaptively emphasize values located close to zero. By assigning more weight to these central features, DAM enhances the discriminative capacity of the encoded features, thereby improving retrieval performance. To further support robust encoding, the training process employs a composite loss function comprising three complementary components: cosine loss, Euclidean loss, and classification loss. Together, these components guide the model toward producing compact and discriminative binary codes that are crucial for efficient retrieval across diverse image classes. Extensive experiments are conducted on three widely used RS datasets, which encompass various land-cover classes and imaging conditions. The experimental results confirm that the proposed CDFE method achieves superior performance and excels in extreme binary compression tasks, where maintaining discriminative power under high compression is critical. These results highlight the effectiveness of CDFE in advancing the capabilities of RS image retrieval.
No takes yet. Share an insight, caveat, or question.
Han et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: