Convolutional Neural Network based Fire Detection for Surveillance Environment

Considering the severity and the loss occurred by the types of damage that can occur due to fire disaster, there is always a demand of continuous progression in fire detection capabilities. This paper proposes fire detection model using Convolution neural networks inspired from ShuffleNet. Usually it has been observed a trade-off between model’s performance and its size, restricting its deployment on memory constrained devices. In this paper we have attempt to achieve descent accuracy which is comparable to some existing work without compromising on its model size. Experimental results on benchmark dataset verify its accuracy and justify its applicability for effective fire disaster management on small portable devices with limited memory.