Computer vision-based characterization of large-scale jet flames using a synthetic infrared image generation approach

Abstract

Different kinds of fire accidents can occur during industrial activities that involve hazardous materials, such as jet fires, which are often involved in a process known as a domino effect that generates a sequence of other accidents of greater magnitude. Jet fires present specific features that can significantly increase the probability of this domino effect, so they become relevant from a risk analysis perspective, making their proper characterization a crucial task. Data acquisition of jet fires involves expensive experiments, especially when infrared imagery is necessary. Therefore, this paper proposes a method that uses Generative Adversarial Networks to produce plausible infrared images from visible ones, making experiments less expensive and allowing for other potential applications. As validation, the infrared images are used in a fire characterization approach that employs Deep Learning to segment radiation zones and extracts the jet fire’s geometrical information. A comparison is done between the measurements obtained from real and generated infrared images. The results suggest that, with the proposed approach, it is possible to realistically replicate the analysis obtained from experiments carried out using both visible and infrared cameras.