BoWFire Fire Detection Segmentation Dataset

The BoWFire dataset is an image dataset specifically for flame detection, which aims to improve the accuracy of fire detection and reduce false alarms. The dataset includes fire images in a variety of emergency situations, such as building fires, industrial fires, car accidents, and riots, as well as emergency situations without visible fire and fire-like areas (such as sunsets and images of red or yellow objects).

The characteristic of the BoWFire dataset is that it not only contains flame detection images, but also includes the segmentation of flame areas. In the dataset, the fire area is marked as white and the non-fire area is marked as black, which provides strong support for flame detection and segmentation tasks7. In addition, the dataset is divided into a training set and a test set. The training dataset contains 240 50×50 pixel images, of which 80 are classified as having fire and 160 are classified as not having fire. The test dataset contains 226 images of different resolutions, of which 119 contain flame images and 107 contain no fire images7.

The BoWFire dataset is used to develop and evaluate a new flame detection method that improves the accuracy of flame detection and reduces false alarms by combining pixel color and texture analysis. Experimental results show that this method can effectively reduce the false alarm rate while maintaining comparable accuracy to the state of the art.

The researchers also conducted detailed configuration and experiments on the BoWFire dataset, using a variety of algorithms including Naive-Bayes classifier and KNN classifier for pixel color classification, and superpixel algorithm and local binary pattern (LBP) for texture classification. The experimental results show that the BoWFire method reduces false alarms while maintaining a high flame detection accuracy.

This dataset was released by a research team from the University of São Paulo in 2015. The related paper is titled “BoWFire: Detection of Fire in Still Images by Integrating Pixel Color and Texture Analysis"