Power Transmission Corridors Wildfire Detection for Multi-Scale Fusion and Adaptive Texture Learning Based on Transformers

Power Transmission Corridors Wildfire Detection for Multi-Scale Fusion and Adaptive Texture Learning Based on Transformers

Ensuring the operational safety of power transmission corridors is critical for maintaining a stable electricity supply. However, their geospatial complexity makes them increasingly vulnerable to wildfires. Wildfires vary dramatically in scale—from pixel-level ignition to kilometer-wide s...

Full description

Saved in:
Bibliographic Details
Main Authors: Yicen Liu, Xi Liu, Zhikun Zheng, Shimin Luo, Yongjun Xiao, Shuxian Wang, Xiaojiang Liu, Zhenhong Sun
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Power transmission corridors wildfire detection
multi-scale fusion
adaptive texture
transformer
Online Access:https://ieeexplore.ieee.org/document/11079617/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Ensuring the operational safety of power transmission corridors is critical for maintaining a stable electricity supply. However, their geospatial complexity makes them increasingly vulnerable to wildfires. Wildfires vary dramatically in scale—from pixel-level ignition to kilometer-wide spread—and often occur in visually cluttered environments. These conditions challenge existing Transformer- and YOLO-based detectors, which struggle with multi-scale feature conflicts and fine-grained inter-class distinctions. To address this, we propose a unified wildfire detection framework incorporating two key modules: the Multi-scale Adaptive Texture Encoding (MATE) module and the Dynamic Feature Pyramid Integration (DFPI) module. MATE leverages heterogeneous receptive fields to capture wildfire patterns across spatial scales, while DFPI enhances feature discrimination via scale-aware attention and adaptive resolution fusion. Additionally, we construct a large-scale wildfire dataset with 5,000 annotated images covering diverse terrains, weather, lighting, and fire stages. Extensive experiments demonstrate that our method achieves superior performance, with an mAP50 of 72.9% and an mAP50-95 of 39.3%, significantly outperforming state-of-the-art baselines in detecting wildfires under real-world, cluttered conditions.
ISSN:2169-3536

Similar Items