多场景烟雾火焰目标检测数据集(S-Firedata)

该团队提出了基于多维度场景的烟雾火焰检测数据集，数据来源整合了公开数据集（如BoWFireDataset、D-Fire等）与自制视频抽帧影像，囊括室内外环境、不同光照条件（如夜间低光、强逆光）、遮挡场景（如树木遮挡、机械干扰）以及动态背景（如人群移动、车辆穿行）。例如，自制数据通过发电厂监控视频抽帧，捕捉了电力设施爆炸引发的黑色烟雾与高温火焰的交互特征；而森林火灾影像则包含远距离烟雾扩散与近景火焰跳变的复合目标。通过高精度标注（边界框交并比≥0.85），该数据集为算法提供了高可靠性基准，可有效支持智慧安防系统的误报率优化、无人机巡检的实时火情识别，以及卫星遥感监测中的大范围火灾动态追踪，推动多场景火灾预警技术的实用化进程。

This research team proposes a multi-dimensional scene-based smoke and flame detection dataset. The dataset integrates data from public sources including BoWFireDataset, D-Fire, etc., as well as frame-extracted footage from self-made videos. It covers various scenarios such as indoor and outdoor environments, different lighting conditions (e.g., low-light nighttime, strong backlighting), occlusion situations (e.g., tree occlusion, mechanical interference), and dynamic backgrounds (e.g., crowd movement, vehicle passing). For example, the self-made data is extracted from power plant surveillance videos, capturing the interactive characteristics between black smoke and high-temperature flames caused by power facility explosions; while forest fire footage contains composite targets of long-distance smoke diffusion and close-up flame flicker. With high-precision annotations where the Intersection over Union (IoU) of bounding boxes is ≥ 0.85, this dataset provides a highly reliable benchmark for algorithms. It can effectively support the optimization of false alarm rates in intelligent security systems, real-time fire recognition for UAV inspections, and large-scale dynamic fire tracking in satellite remote sensing monitoring, thereby promoting the practical advancement of multi-scenario fire warning technologies.

搜集汇总

背景与挑战

背景概述

该数据集是一个专门用于烟雾火焰目标检测的多场景数据集，整合了公开数据集和自制视频抽帧影像，覆盖室内外、不同光照、遮挡及动态背景等复杂环境。数据集经过高精度标注，边界框交并比(IoU)≥0.85，标注一致率≥95%，确保了数据质量可靠。它适用于火灾预警、智慧安防、无人机巡检和卫星遥感监测等深度学习应用，为算法提供了高可靠性基准。