Answers are grounded in this paper and cite their sources.
Remote sensing change detection (RSCD) elucidates spatiotemporal land cover dynamics through bitemporal image analysis, thereby providing critical infrastructure for urban expansion monitoring, forest degradation assessment, and disaster response management. Despite advancements in deep learning-based RSCD frameworks, three pivotal limitations persist: first, suboptimal global context modeling, second, insufficient temporal feature encoding, and third, monolithic decision architectures lacking cognitive stratification. To mitigate these challenges, we propose the human cognitive decision-inspired network—a novel RSCD paradigm that reengineers feature representation and information fusion through a biologically plausible four-stage pipeline: perception–memory–observation–decision. In the perception stage, we leverage a pretrained vision foundation model to extract hierarchical multiscale image features, simulating initial human perceptual processing of visual scenes. The memory stage incorporates a bidirectional long short-term memory architecture augmented with channel–spatial attention modules, enabling bidirectional temporal context encoding while selectively reinforcing feature associations in salient change regions. The observation stage introduces a Dense Mamba module that integrates multiscale convolutional kernels with dynamic weight allocation strategies, facilitating efficient global context perception through parallel feature extraction and adaptive fusion. Finally, the decision stage constructs an observation–memory fusion decider that synthesizes multievidence representations through cross-modal attention mechanisms, emulating human collaborative decision-making processes to generate high-fidelity change masks. Notably, we compared our work with 19 state-of-the-art works, including both recent and classical works, showing improved accuracy (via five evaluative metrics) over five benchmark datasets.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: