2026· IEEE Transactions on Geoscience and Remote Sensing· Vol 64, pp. 5636712-5636712· 0 citations· 55 references
Abstract
Small object detection in remote sensing (RS) imagery remains fundamentally challenging due to severe information degradation caused by limited spatial resolution and complex background interference. In deep neural networks, such degradation is further exacerbated by irreversible information loss during conventional downsampling, leading to weak and ambiguous feature representations. To address this issue, we propose a frequency–spatial joint decoupling and adaptive perceptual aggregation module (FSD-APAM), which explicitly separates signal-level detail preservation and perceptual-level feature discrimination within a unified framework. Specifically, a frequency-domain detail decoupling (FDD) unit leverages discrete wavelet transform to construct reversible feature pathways, enabling the recovery of high-frequency edge information suppressed during downsampling. Complementarily, a spatial salience decoupling (SSD) unit introduces a lateral inhibition mechanism to enhance isolated target responses while suppressing structured background interference. To further ensure global contextual consistency with low computational overhead, an adaptive contextual perceptual aggregation (ACPA) unit is designed to facilitate efficient interaction between sparse target cues and dense semantic representations. Extensive experiments on AI-TODV2, LEVIR-Ship, and VisDrone demonstrate that the proposed method consistently improves detection performance across multiple mainstream architectures without requiring substantial architectural modifications. In particular, FSD-APAM achieves superior accuracy in detecting extremely small objects while maintaining competitive efficiency, highlighting its practical value for large-scale RS applications. The source code is available at https://github.com/cskkx1/FSD-APAM
Small object detection in remote sensing imagery remains challenging due to the extremely limited pixel footprint of targets and the severe loss of fine-grained spatial details caused by multistage downsampling. To address these issues, we propose WE-DETR, a wavelet-enhanced multiscale feature compensation detection tr...
Yu Song, W. Zuo, Rong-Hua Shang et al.· IEEE Transactions on Geoscie...· 0 citations
The feature mapping stage is reconstructed to incorporate physical edge priors and frequency-domain attention-based denoising, and a Spatial Attention Wavelet Transform Feature Decomposer (SA-WTFD) is introduced in the frequency branch to adaptively suppress high-frequency background clutter and enhance discriminative...
Qi-Yuan Zhang, Jian-Shun Liu· Italian National Conference...· 1 citation
Small object detection in remote sensing images (RSIs) is challenging because imaging degradation weakens object textures, reduces contrast, and blurs boundaries. These effects are further aggravated by hierarchical feature extraction, where repeated downsampling weakens shallow spatial cues before they reach deeper se...
Wei He, Yun-Tao Xu, Qi Qi et al.· IEEE Transactions on Geoscie...· 0 citations
To address the attenuation of shallow fine-grained structural information during deep feature propagation and the representational conflict between classification and regression in remote sensing tiny object detection, this paper proposes a dual asymmetric detection framework, termed A2-Det. Built upon YOLO11n, the det...
A frequency-decoupled enhancement and large kernel cross-scale fusion network (FDLNet) is proposed, which jointly exploits spatial and frequency representations to disentangle genuine changes from interference and achieves state-of-the-art performance.
Yan Xing, Yu-Nan Jia, Jia-Li Hu et al.· IEEE Journal of Selected Top...· 0 citations
The results demonstrate that MWAE-YOLO can effectively enhance frequency-aware small-object feature representation, suppress background interference, and improve scale-sensitive localization accuracy in complex remote sensing scenes.