Efficient Aerial Image Dehazing via Haze Region Aware Feature Learning and Hadamard Gating

Abstract

Remote sensing imagery is highly susceptible to haze, which can obscure visibility and limit the reliability of downstream analysis tasks, making aerial image dehazing critical for space and defense applications. Existing methods often fail to faithfully restore structural details and color fidelity under spatially varying dense haze and are parameter-intensive, limiting their deployment in resource-constrained environments. To address this issue, we propose a lightweight encoder–decoder framework for aerial image dehazing that jointly models anisotropic spatial characteristics and global contextual dependencies. The proposed multiscale directional feature fusion (MDFF) module explicitly encodes directional and enlarged receptive-field interactions to capture spatially varying haze distribution, while the haze region-aware refinement (HRAR) module estimates haze regions and refines degraded features through dual-stage attention for guided feature learning. In addition, the Hadamard-gated feature modulation (HGFM) module introduces parameter-efficient multiplicative feature recalibration to enhance fine textural details. Experimental results on benchmark datasets demonstrate that the proposed method achieves superior restoration performance while maintaining substantially fewer parameters and higher computational efficiency. The code will be publicly available at: https://github.com/Shiladityagit/Lightweight_Aerial_Dehazing.