AI Insight
This study presents a deep learning method for detecting both visible and obscured wildfire flames by analyzing smoke movement patterns in RGB video footage. The researchers developed a frame-wise transformer architecture that examines temporal patterns across consecutive video frames while maintaining spatial information, achieving high accuracy rates of over 92% in fire segmentation and detection tasks. The method was tested on the FLAME2 dataset using only RGB video input, making it compatible with low-cost commercial drones equipped with standard cameras.
Why it matters
This approach could significantly improve early wildfire detection capabilities using affordable drone technology, as it can identify fires even when flames are hidden by smoke or vegetation. The reliance on standard RGB cameras rather than expensive infrared equipment makes the technology accessible for widespread deployment in wildfire monitoring and management.
Understand the Science
⚠️ Preprint – Noch nicht peer-reviewed
Dieser Artikel wurde noch nicht von unabhängigen Experten begutachtet. Die Ergebnisse sind vorläufig und sollten mit Vorsicht interpretiert werden.
Abstract: In this paper, we propose a new paradigm in wildfire management for detecting both visible and obscured fire using Deep Learning (DL). The essence of our method is the joint spatial and temporal analysis of RGB videos, embracing the fact that temporal patterns of smoke motions in consecutive video frames can indicate obscured fire flames. This perspective contrasts the traditional approach of independently analyzing images and individual video frames. To this end, we propose a frame-wise transformer architecture that applies an attention mechanism between the consecutive frames while preserving spatial information within frames as opposed to regular attention mechanisms like the non-local block, which combines spatial and temporal information into a single vector. We also optimize the temporal pattern lengths under different spatial resolutions to achieve the highest performance while reducing the computational cost. To investigate the applicability of our method in aerial image processing, we applied it to a curated version of the FLAME2 dataset that includes dual-mode RGB/IR videos. The IR video feed is used to extract ground truth, while the model is trained only on RGB videos to maintain applicability to lowcost commercial drones (which are equipped only with regular cameras) and facilitate processing existing RGB-only datasets. Our proposed method achieves a Segmentation Foreground Dice score of 92.61%, an Object Detection Precision and Recall rate of 93.21% and 91.73%, and clip-level classification accuracy of 98.02%, respectively. We also investigated the efficacy and generalizability of our method through ablation study