Enhancing wildfire detection: a novel algorithm for controllable generation of wildfire smoke images
Overview
This work develops a GAN-based, controllable smoke-image generator that improves background consistency and expands training data for image-based wildfire detection.
Core Contributions
- Uses multi-scale feature fusion to improve consistency between synthesized smoke and complex forest backgrounds.
- Introduces local feature matching in the discriminator to guide realistic smoke-texture generation.
- Generates 13,400 wildfire-smoke images using forest backgrounds and early-fire simulations from the Fire Dynamics Simulator.
Publication Information
- Authors: Yinuo Huo, Qixing Zhang, Chong Wang, Haihui Wang, Yongming Zhang
- Journal: International Journal of Wildland Fire
- Publication date: 2024-11-11
- DOI: 10.1071/WF24068
- Publisher: CSIRO Publishing
- Status: Published
Citation
Yinuo Huo, Qixing Zhang, Chong Wang, Haihui Wang, Yongming Zhang. “Enhancing wildfire detection: a novel algorithm for controllable generation of wildfire smoke images.” International Journal of Wildland Fire, vol. 33, no. 11, WF24068 (2024). https://doi.org/10.1071/wf24068
Keywords
wildfire detection, smoke synthesis, generative adversarial network, data augmentation, Fire Dynamics Simulator
还没有评论，欢迎发表第一条讨论。