EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

LWL-RT-DETR: A Lightweight Collaborative Model for Wildlife Detection

Huan Huang, Xiaodong Yang

IEEE Access · 2026 · 14 · 107969-107983

DOI: 10.1109/access.2026.3713506 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Huan Huang und 1 weiteren Autor:innen. Die Arbeit nennt 32 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Infrared camera technology is critical for wildlife diversity monitoring, but its inherent image characteristics, such as weak texture, low contrast, and severe scale variation, pose serious challenges to existing deep learning detectors. To address this problem, this study proposes a lightweight wildlife detection model, LWL-RT-DETR, for infrared images. The model focuses on detail compensation, scale adaptation, and lightweight modeling through a systematic reconstruction of RT-DETR. First, a Superficial Detail Fusion Module (SDFM) is designed to enhance the perception of weak edges and textures in the early stage of feature extraction. Second, the Attention Scale Sequence Fusion module (ASF) is introduced to achieve adaptive fusion of multi-scale features via a dynamic attention mechanism. Third, the context-guided module (CG) is used to reconstruct the backbone network, which significantly reduces the number of parameters while expanding the receptive field. Finally, a Scale-based Dynamic Loss (SD Loss) is proposed to dynamically adjust the optimization weight according to the target size. Experiments on a self-built infrared camera wildlife dataset show that the proposed model achieves mAP50 of 85.6%, precision of 90.7%, and recall of 82.1% with a 17.0% reduction in parameters, which are 3.0, 4.3, and 3.1 percentage points higher than the baseline RT-DETR-ResNet18, respectively. This work offers a promising direction for accurate wildlife monitoring with reduced computational complexity.

Zitieren

APA

Huang, H., & Yang, X. (2026). LWL-RT-DETR: A Lightweight Collaborative Model for Wildlife Detection. IEEE Access, 14, 107969-107983. https://doi.org/10.1109/access.2026.3713506

Vancouver

Huang H, Yang X. LWL-RT-DETR: A Lightweight Collaborative Model for Wildlife Detection. IEEE Access. 2026;14:107969-107983. doi:10.1109/access.2026.3713506

BibTeX
@article{Huang2026,
  author = {Huang, Huan and Yang, Xiaodong},
  title = {LWL-RT-DETR: A Lightweight Collaborative Model for Wildlife Detection},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {107969--107983},
  doi = {10.1109/access.2026.3713506}
}
Kostenlos starten →