BDF-18K Dataset(ForestFireDataset)

BDF-18K Dataset (BoWFire - Dataset for Fire and Smoke - Forest FireDataset)

This dataset is designed to support forest fire detection tasks, combining public and proprietary datasets, and is named BDF-18K. It contains images from the BoWFire and DFS public datasets, as well as a self-collected dataset captured in a forest area in Shifeng District, Zhuzhou City, Hunan Province, China. This dataset covers various forest fire scenarios and features.

Dataset Composition:

Public Datasets:

BoWFire: Includes 122 images, mainly for building fires, industrial fires, etc.

DFS Dataset: Includes 9462 images, involving static images of flames, smoke, and background.

Self-collected Dataset:

ForestFireDataset: Captured using drones and handheld devices, emphasizing dynamic scenes, multiple perspectives, and complex backgrounds. This dataset provides valuable data for simulating real-world forest fire rescue scenarios.

Capture Equipment:

DJI Mavic 2 Dual (Low altitude capture, covering different perspectives, lighting conditions)

DJI Matrice 300 RTK with Zenmuse H20T camera (High-altitude capture, supports infrared and visible light modes)

Moto Edge X30 phone (Simulating the perspective of a firefighter, capturing close-up fire details)

Capture Location: A typical forest area in Zhuzhou, Hunan Province, representing the natural environment and climate of Southern China.

Dataset Content:

Flames (small and large scale), smoke (light and dense smoke), firefighters, volunteers, and environmental backgrounds (vegetation, bare ground).

Image resolution: 1920x1080, frames extracted at 30 FPS.

Dataset Labeling:

The dataset uses the "fire" label for marking fire spots, suitable for training models such as YOLO.

Dataset Splits:

The dataset is split into 70% for the training set, 20% for the validation set, and 10% for the test set. Each subset is balanced in terms of class distribution to ensure fair training and evaluation.

Public Datasets Download:

The BoWFire and DFS datasets are public and can be downloaded independently. Specific download links can be found in the relevant documentation or release channels.

BoWFire Dataset: Includes building fires, industrial fires, and other scenarios, as well as images with no obvious flames (e.g., sunsets or red/yellow objects), making it useful for fire detection.

DFS Dataset: Contains flames, smoke, and background images, suitable for fire and smoke detection tasks.