Deep Learning-Based Smoke Detection in Petrochemical Scenes

Over the past few centuries, petrochemical disasters have occurred both internationally and domestically, with a large number of casualties. Due to the high purity and quantity of combustible materials, the consequences of petrochemical accidents are usually very serious. Smoke as a precursor to flames can be detected as an early warning for subsequent serious accidents. In this work, a smoke detector from image has been developed based on SSD-MobileNet in petrochemical scenes. A smoke dataset containing 3042 images captured in various environments downloaded from the Internet and 1968 synthesized images has been collected for evaluating the proposed model。 Experiment results show that the accuracy and efficiency of the proposed SSD-MobileNet-based deep model has achieved a relative good results, which shows that smoke can be detected accurately from images.