Video Smoke Detection Based on Convolution Neural Network

In this paper, a new method of video smoke detection based on convolution neural network and Mixture Gaussian background model is proposed for the low smoke detection rate in complex scene. The CNN model is trained by GoogLeNet, and the recognition rate is limited in the process of model recognition. The experimental results show that GoogLeNet can obtain good results with limited training data and limiting recognition rate of the model can reduce false positives effectively. The method proposed in this paper can meet the requirements of real-time smoke detection.