Construction and Verification of Image Datasets for Fire Hazards in Cultural Relics Buildings

Abstract

The fire safety of cultural relic building(CRB) is an important topic in the field of cultural relic protection. In recent years, more and more researchers have applied technologies such as image processing and machine learning to the early detection and alarm of CRB fires. However, the image data of fire and interference sources in CRB scenes is scarce. This article proposes a scheme for constructing a fire hazard image dataset based on the characteristics of CRB scenes. On this basis, in order to meet the requirements of timeliness, accuracy, and reliability for fire detection in CRBs, a lightweight FireNet fire detection network was used to train the FireNet dataset. The obtained training parameters were applied to the CRB Fire Hazard Dataset for testing, and the recognition accuracy reached 70.78% without training. The above results indicate that the network ensures both lightweight and high level of accuracy in fire detection of CRBs. At the same time, it also proves that there is a significant difference in the image fire detection effect between CRB scenes and other building scenes, and the construction of a CRB fire hazard image dataset is of great significance.