Comparative Analysis of Traditional and Deep Learning Techniques for Industrial and Wildfire Smoke Segmentation

Abstract

Smoke is the first sign of ignition of fire because smoke becomes visible when the fire starts. At this stage, fire can be effectively controlled by locating the smoke at the earliest. Smoke causes several health issues such as skin allergies and breathing problems in humans and animals. One of the biggest smoke emission sources is the industrial smoke. For environmental safety, various harmful gases emitting from industrial chimneys need to be monitored constantly. Further, increasing incidents of wildfire have also resulted in severe environmental degradation in recent years. Thus, detection of smoke and finding its location at early stage can help in mitigating fire hazards. Several vision based techniques have been proposed by researchers using traditional image processing techniques in the past to identify and segment smoke in images. In recent years, deep learning techniques have shown promising performance in smoke detection. In this paper, we present a comparative analysis of traditional image processing and recent deep learning based smoke segmentation techniques with focus on industrial and wildfire smoke.