HAFM-Net: Hierarchical Alignment Fusion and Mapping for UAV-Based Misaligned RGB-T Salient Object Detection.
Saved in:
| Title: | HAFM-Net: Hierarchical Alignment Fusion and Mapping for UAV-Based Misaligned RGB-T Salient Object Detection. |
|---|---|
| Authors: | Zhang, Zhijie1 (AUTHOR), Chen, Kaihong1,2 (AUTHOR), Yang, Chen2,3 (AUTHOR), Zhang, Shanwen1,4 (AUTHOR), Wang, Zhen1,3,4 (AUTHOR) wangzhen4013@163.com |
| Source: | Remote Sensing. Jun2026, Vol. 18 Issue 12, p2039. 30p. |
| Subjects: | Drone aircraft, Multisensor data fusion |
| Abstract: | Highlights: HAFM-Net predicts saliency directly from unaligned UAV RGB-T inputs without explicit pixel-level registration. MRCF, HASI, and SPCE jointly improve misalignment-robust fusion, cross-scale representation, and small-object localization. Experiments on UAV RGB-T 2400 and weakly aligned benchmarks demonstrate robustness under blur, illumination variation, small-object, and foggy scenes. In unmanned aerial vehicle (UAV) scenarios, RGB-T salient object detection faces several challenges, including cross-modal spatial misalignment, redundant multi-scale features, and weak responses of small objects in cluttered backgrounds, which together degrade fusion effectiveness and localization stability in complex environments. To address these issues, we propose a Hierarchical Alignment Fusion and Mapping Network (HAFM-Net), a misalignment-robust fusion framework, for unaligned RGB-T salient object detection. The proposed method does not rely on explicit pixel-level preregistration. Instead, it replaces registration-first preprocessing with implicit feature-domain alignment and misalignment-robust fusion, enabling saliency prediction from unregistered RGB-T inputs. Specifically, we design a hierarchical adjacent-scale interaction mechanism to enhance multi-scale contextual modeling while suppressing cross-scale redundancy. We further develop a Misalignment-Robust Correlation Fusion module to explore cross-modal correlations and enable robust feature interaction under positional variations. In addition, a semantic–spatial complementary enhancement is introduced to promote collaboration between high-level semantic cues and low-level spatial details, thereby improving the representation and boundary localization of small salient objects. Experimental results on the UAV RGB-T 2400 dataset and an additional weakly aligned benchmark demonstrate that HAFM-Net achieves competitive performance and exhibits strong robustness in challenging scenarios, such as blur, illumination variation, small-object cases, and foggy conditions. [ABSTRACT FROM AUTHOR] |
| Copyright of Remote Sensing is the property of MDPI and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| Database: | Engineering Source |
|
Full text is not displayed to guests.
Login for full access.
|
|
| Abstract: | Highlights: HAFM-Net predicts saliency directly from unaligned UAV RGB-T inputs without explicit pixel-level registration. MRCF, HASI, and SPCE jointly improve misalignment-robust fusion, cross-scale representation, and small-object localization. Experiments on UAV RGB-T 2400 and weakly aligned benchmarks demonstrate robustness under blur, illumination variation, small-object, and foggy scenes. In unmanned aerial vehicle (UAV) scenarios, RGB-T salient object detection faces several challenges, including cross-modal spatial misalignment, redundant multi-scale features, and weak responses of small objects in cluttered backgrounds, which together degrade fusion effectiveness and localization stability in complex environments. To address these issues, we propose a Hierarchical Alignment Fusion and Mapping Network (HAFM-Net), a misalignment-robust fusion framework, for unaligned RGB-T salient object detection. The proposed method does not rely on explicit pixel-level preregistration. Instead, it replaces registration-first preprocessing with implicit feature-domain alignment and misalignment-robust fusion, enabling saliency prediction from unregistered RGB-T inputs. Specifically, we design a hierarchical adjacent-scale interaction mechanism to enhance multi-scale contextual modeling while suppressing cross-scale redundancy. We further develop a Misalignment-Robust Correlation Fusion module to explore cross-modal correlations and enable robust feature interaction under positional variations. In addition, a semantic–spatial complementary enhancement is introduced to promote collaboration between high-level semantic cues and low-level spatial details, thereby improving the representation and boundary localization of small salient objects. Experimental results on the UAV RGB-T 2400 dataset and an additional weakly aligned benchmark demonstrate that HAFM-Net achieves competitive performance and exhibits strong robustness in challenging scenarios, such as blur, illumination variation, small-object cases, and foggy conditions. [ABSTRACT FROM AUTHOR] |
|---|---|
| ISSN: | 20724292 |
| DOI: | 10.3390/rs18122039 |
