Image processing to calculate flame velocity based on deep learning

Flame velocity is one of the main characteristics of combustion. One of the methods to measure flame velocity is to determine the flame area by approximating the shape of the flame to a flat, conical, or truncated cone surface. To increase the precision in the calculation of the flame area and thus improve the calculation of the flame velocity, the use of semantic segmentation of the image is proposed to determine the flame image contour and then obtain the flame area with an integral calculation. Semantic segmentation was performed using the U-net structure. The results were compared with other classical image processing techniques showing the advantages and disadvantages. The results show that semantic segmentation based on Deep Learning segments the flame image with high precision even under changing conditions such as flame color variations and background imperfections. Key words. Laminar flames, burning velocity, image processing, deep learning.