An Improved Algorithm Based on Convolutional Neural Network for Smoke Detection

Abstract

As an essential method of fire prevention and disaster control, smoke detection is of great significance to production and life. At present, the convolutional neural network (CNN) has achieved good results in the research of smoke detection. However, the detection accuracy is not high for some scenes. For example, the wind speed is tremendous, and the shape of the smoke changes rapidly. In order to deal with this problem better, this paper proposes an improved algorithm based on cascading classification and deep convolutional neural network. In the cascading classification part, we improve the cascading structure and make it select the appropriate parameter threshold for the smoke generated in different scenes. The convolutional neural network structure is trained to extract the variation characteristics of smoke better. Also, we optimize the parameters on the target data set. The experimental results show that the algorithm has achieved excellent results in accuracy and speed on the selected smoke detection data sets.