LwST-FSD: Efficient forest smoke detection via lightweight spatio-temporal modeling
Overview
LwST-FSD combines a temporal-shift-enhanced CSPDarknet backbone with a partial temporal aggregation neck for efficient spatio-temporal forest-smoke detection.
Core Contributions
- Develops TS-CSPDarknet, a lightweight spatio-temporal backbone built from CSPDarknet and Temporal Shift Modules.
- Introduces PTANeck to fuse temporal dynamics with multi-scale spatial features.
- Evaluates the detector on FIgLib, SKLFS-WildFire, and SKLFS-WildFire2025, the latter containing 8,363 video segments and 152,700 manually labeled frames.
Publication Information
- Authors: Chong Wang, Hai Song, Zhong Wang, Zhilin Shan, Qixing Zhang
- Journal: Information Sciences
- Publication date: 2025-11-01
- DOI: 10.1016/j.ins.2025.122481
- Publisher: Elsevier BV
- Status: Published
Citation
Chong Wang, Hai Song, Zhong Wang, Zhilin Shan, Qixing Zhang. “LwST-FSD: Efficient forest smoke detection via lightweight spatio-temporal modeling.” Information Sciences, vol. 719, 122481 (2025). https://doi.org/10.1016/j.ins.2025.122481
Keywords
forest smoke detection, spatio-temporal modeling, Temporal Shift Module, lightweight detector, wildfire
还没有评论，欢迎发表第一条讨论。