FF-Mamba-YOLO: An SSM-Based Benchmark for Forest Fire Detection in UAV Remote Sensing Images.

Area of Science:

• Remote Sensing
• Artificial Intelligence
• Forestry

Background:

• Accurate forest fire detection via UAVs is critical but challenged by multiscale targets and environmental interference.
• Existing methods struggle with complex UAV remote sensing image data.

Purpose of the Study:

• To develop an advanced UAV remote sensing target detection framework for improved forest fire identification.
• To address limitations in detecting multiscale targets and handling complex environmental interference in forest fire images.

Main Methods:

• Introduced FF-Mamba-YOLO, integrating Mamba and YOLO principles with novel modules (MFEBlock, MFFBlock, CFEBlock, MGBlock).
• Utilized state space models (SSMs) for global dependency capture and feature enhancement modules for local processing.
• Enhanced Path Aggregation Feature Pyramid Network (PAFPN) for superior feature fusion and DySample for efficient resolution enhancement.

Main Results:

• Achieved 67.4% mAP@50, 36.3% mAP@50:95, and 64.8% precision on a custom forest fire dataset.
• Demonstrated superior performance compared to existing state-of-the-art methods in forest fire detection tasks.
• FF-Mamba-YOLO effectively captured global dependencies and improved local feature processing and adaptive capabilities.

Conclusions:

• FF-Mamba-YOLO offers a robust and accurate solution for forest fire detection using UAV remote sensing.
• The novel architecture and modules significantly enhance detection performance in challenging conditions.
• This framework shows great potential for real-time forest fire monitoring and management.