Work

Learning a density-guided adaptive spiking network for remote sensing image dehazing

Year: 2026
Type: article
Abstract: Remote sensing images are often degraded by atmospheric haze, resulting in reduced visibility and structural information loss. Although spiking neural networks (SNNs) have shown potential for efficient image restoration, existing approaches have not explicitly incorporated spatial haze density into neuronal dynamics, limiting their adaptability to non-uniform haze degradation. In this paper, we propose DASNet, a Density-Guided Adaptive Spiking Network for remote sensing image dehazing. Different from existing SNN-based restoration approaches, DASNet introduces haze degradation awareness into neuronal dynamics through a lightweight Haze Density Estimation (HDE) branch, which predicts spatial haze density distributions from hazy observations. Based on the estimated density map, we design an Adaptive Haze-aware Leaky Integrate-and-Fire (AH-LIF) neuron that dynamically adjusts membrane potential evolution according to different haze levels, enabling region-specific temporal information aggregation. Specifically, heavily degraded regions can preserve more historical information through enhanced membrane integration, while lightly degraded regions maintain rapid responses for detail preservation. Furthermore, a pseudo haze density supervision strategy is introduced using paired hazy and clear remote sensing images to improve density estimation accuracy. Extensive experiments on public remote sensing image dehazing benchmarks demonstrate that DASNet achieves competitive restoration performance while preserving the efficiency advantages of spiking neural networks.
Language: English
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7236155162
FWCI:
Cites: 52
Cited by:
Related to: 10