Open AccessVideo fire recognition based on multi-channel convolutional neural network
Author(s) -
Chen Zhong,
Yu Shao,
Hongjun Ding,
Ke Wang
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1634/1/012020
Subject(s) - convolutional neural network , computer science , artificial intelligence , fire detection , channel (broadcasting) , pattern recognition (psychology) , segmentation , feature (linguistics) , computer vision , engineering , telecommunications , architectural engineering , linguistics , philosophy
A video fire detection method based on multi-channel convolutional neural network(CNN) is presented in this paper. With improved OTSU and connectivity analysis, an adaptive threshold segmentation for flame image is implemented. According to the flame combustion characteristics, the feature of flame area is extracted using parameters such as colour, roundness and area change. Using TensorFlow platform, the CNN and the video database is formed. In order to improve the ability of dynamic fire detection, the method of combining random gradient descent and momentum correction is used to train CNN. Experimental results show that the accuracy of CNN fire detection method is 86%, compared with 68.2% of the traditional k-nearest algorithm (KNN).