TGMNet: Temporal-Guided Mamba Network for Moving Infrared Dim and Small Target Detection.
Saved in:
| Title: | TGMNet: Temporal-Guided Mamba Network for Moving Infrared Dim and Small Target Detection. |
|---|---|
| Authors: | Li, Kemian1 (AUTHOR), He, Xiaoyu1 (AUTHOR), Liu, Tianjin1 (AUTHOR) tianjinliu@buaa.edu.cn |
| Source: | Remote Sensing. Aug2026, Vol. 18 Issue 15, p2578. 21p. |
| Subjects: | Temporal integration, Signal detection, Remote sensing, Computer performance, Artificial neural networks, Real-time computing |
| Abstract: | Highlights: What are the main findings? A Temporal-Guided Mamba Network (TGMNet) is proposed for infrared small target detection by integrating spatial features with temporal dependency modeling via Mamba. Experimental results on DAUB and IRDST datasets demonstrate that TGMNet achieves state-of-the-art detection performance, reaching 94.71% mAP50 and 97.85% F1 on DAUB while requiring only 4.64M parameters and maintaining competitive inference speed. What are the implications of the main findings? The strategy of spatial enhancement before temporal modeling effectively preserves weak infrared target cues, enabling more reliable temporal dependency modeling against cluttered backgrounds. Lightweight design and competitive inference speed indicate the potential of the proposed method for real-time surveillance and remote sensing applications. Infrared dim and small target detection has drawn increasing attention due to its importance in both military and civil applications. Multi-frame approaches exploit temporal information across consecutive frames and generally achieve better performance than single-frame methods. However, effective exploitation of temporal information while maintaining computational efficiency remains challenging. In this paper, we propose a Temporal-Guided Mamba Network (TGMNet) for moving infrared dim–small target detection. Specifically, a Spatially-Enhanced Temporal Mamba (SETM) module is designed to model temporal dependencies across consecutive frames. Prior to temporal modeling, Center Difference Convolution and a Multi-Scale Block are introduced to enhance local contrast and multi-scale spatial contextual information. A Feature Aggregation Module (FAM) is developed to integrate temporal motion features from SETM with complementary spatial information from the key and the reference frames, enabling effective spatiotemporal feature interaction. Extensive experiments on the DAUB and the IRDST datasets demonstrate that the proposed TGMNet achieves superior detection performance compared with state-of-the-art methods while maintaining high computational efficiency, validating its effectiveness for moving infrared dim–small target detection. [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: What are the main findings? A Temporal-Guided Mamba Network (TGMNet) is proposed for infrared small target detection by integrating spatial features with temporal dependency modeling via Mamba. Experimental results on DAUB and IRDST datasets demonstrate that TGMNet achieves state-of-the-art detection performance, reaching 94.71% mAP50 and 97.85% F1 on DAUB while requiring only 4.64M parameters and maintaining competitive inference speed. What are the implications of the main findings? The strategy of spatial enhancement before temporal modeling effectively preserves weak infrared target cues, enabling more reliable temporal dependency modeling against cluttered backgrounds. Lightweight design and competitive inference speed indicate the potential of the proposed method for real-time surveillance and remote sensing applications. Infrared dim and small target detection has drawn increasing attention due to its importance in both military and civil applications. Multi-frame approaches exploit temporal information across consecutive frames and generally achieve better performance than single-frame methods. However, effective exploitation of temporal information while maintaining computational efficiency remains challenging. In this paper, we propose a Temporal-Guided Mamba Network (TGMNet) for moving infrared dim–small target detection. Specifically, a Spatially-Enhanced Temporal Mamba (SETM) module is designed to model temporal dependencies across consecutive frames. Prior to temporal modeling, Center Difference Convolution and a Multi-Scale Block are introduced to enhance local contrast and multi-scale spatial contextual information. A Feature Aggregation Module (FAM) is developed to integrate temporal motion features from SETM with complementary spatial information from the key and the reference frames, enabling effective spatiotemporal feature interaction. Extensive experiments on the DAUB and the IRDST datasets demonstrate that the proposed TGMNet achieves superior detection performance compared with state-of-the-art methods while maintaining high computational efficiency, validating its effectiveness for moving infrared dim–small target detection. [ABSTRACT FROM AUTHOR] |
|---|---|
| ISSN: | 20724292 |
| DOI: | 10.3390/rs18152578 |
