Early smoke detection of forest wildfire video using deep belief network

Abstract

This paper presents a novel approach for smoke detection to overcome forest wildfires based on machine learning technique (Deep Belief Network). Video smoke detection is applied on many surveillance and security applications. Smoke detection method should have a high detection rate to have a strong detector of smoke detection. Deep Belief Network which is a stacked layers of Restricted Boltzman Machine is the technique that we used for smoke detection. This technique extracts and classify smoke and no smoke regions simultaneously. The effectiveness of our implemented smoke detection method is evaluated after calculating smoke detection rate, time of pre-traing and time of fine-tuning. The higher is the detection rate the better is the smoke method and the lowest is the time of pre-training and fine-tuning the speeder is the method for smoke detection.