Convolutional Neural Network for Fire Video Image Detection in the Thermal Power Plant

In this paper, the mobility features of fire to eliminate the interference of similar fire scenes such as lighting by using the change of fire coordinates before and after the fire video in the thermal power plant were proposed to address the issues of interference lookalike fire scenes in the recognition approach. The structure used in this paper for training and testing was the Caffe framework after considering a lot of open-source frameworks for deep learning. After images were taken from several thermal videos, 92% accuracy of performance was obtained. The system was able to differentiate between the false positive fire and non-fire regions with high accuracy. The experiment's outcome indicated that this proposed system could identify, locate and recognize images of fire. The method identifies and localizes fire images for unlike fire situations with good generalization and anti-interference ability.