Fire Scene Thermal Imaging Segmentation Dataset

In the current rapid urbanization process, the frequency of fire accidents is high and the impact range is broad, while existing fire detection methods have deficiencies in accuracy and real-time response capability, leading to missed opportunities for optimal fire fighting. Traditional visual detection methods are easily affected by smoke occlusion, thus there is an urgent need to introduce more reliable detection technology. The Fire Scene Thermal Imaging Segmentation Dataset addresses this issue by providing image data based on thermal imaging, focusing on accurately segmenting fire source locations to achieve faster fire response and handling. The dataset was built using thermal imaging camera equipment to collect image data in various simulated fire environments, ensuring data diversity and authenticity. In terms of quality control, multiple rounds of expert annotation and consistency checks were adopted to ensure the accuracy and consistency of annotations. The annotation team has rich backgrounds in firefighting and thermal imaging technology, with a scale of 20 people. Data preprocessing includes noise filtering, image enhancement, and corresponding label generation, and is finally stored in hierarchical folders for easy retrieval and use. The Fire Scene Thermal Imaging Segmentation Dataset features high annotation accuracy (up to 98%) and strong data integrity (covering various fire scenarios), and also employs innovative multi-modal data enhancement technology to improve the accuracy of model fire source recognition. Compared to traditional visible light image datasets, this dataset performs superiorly in smoke-obscured environments, providing up to a 35% recognition performance improvement. Additionally, the dataset's data scarcity lies in its real fire scene coverage and detailed segmentation annotations, capable of handling diverse fire detection needs, with high scalability and industry versatility.