A lightweight edge-based wild boar detection framework for wildlife monitoring and decision support in agroecosystems.
Wild boar (Sus scrofa) intrusions in agricultural environments cause severe crop damage and economic losses and have become an increasingly prominent human-wildlife conflict. Effective wildlife monitoring is therefore essential for sustainable agricultural ecosystem management. However, wild boar detection in real-world scenarios remains challenging due to frequent occlusion and complex backgrounds. Meanwhile, edge devices deployed in farmland environments typically have limited computational resources, imposing strict requirements on lightweight model design and inference efficiency. To address these challenges, this study proposes HED-YOLOv11, a lightweight wildlife object detection framework with three targeted architectural optimizations. First, the backbone network is reconstructed using HGNetv2 to enhance multi-scale feature extraction capability. Second, a novel C2EPSA module is designed to strengthen discriminative feature representation and suppress background interference in complex farmland scenes. Third, the DySample module is introduced into the neck network to improve feature reconstruction in challenging scenes involving occlusion. Furthermore, to facilitate potential deployment in field monitoring systems, a collaborative model compression strategy combining LAMP-based channel pruning and knowledge distillation is proposed, reducing computational overhead while maintaining detection performance. Experimental results show that HED-YOLOv11 achieves 92.6% precision and 92.2% [email protected], with 2.37M parameters. After model compression, the parameters is reduced to 1.27M. The compressed model achieves an inference speed of 30.7 FPS on the RK3568 edge device. These results demonstrate the feasibility of the proposed framework for edge-based wild boar detection and its potential to support wildlife monitoring and management decision-making in agricultural ecosystems.
Authors
- Degang Wang (ORCID: https://orcid.org/0000-0001-7524-1392)
- Zhang Junguo
- Mingyu Xu
Institutions
- Beijing Forestry University (CN)
Publication Details
- Journal
- PubMed
- Published
- 2026-11-12
- DOI
- https://doi.org/10.1016/j.jenvman.2026.131317
- Primary Topic
- Advanced Neural Network Applications
- Type
- article
- Field-Weighted Citation Impact
- 0.00