Aluminum Droplet, Oxide Cap and Flame Segmentation in Burning Al/Ap Propellants by Combining Yolov7 and Two-Stage Cluster

Abstract

Aluminum additives pose complex influences on propellant combustion, and visualizing the process by high-speed microscopic imaging is one of the most powerful tools to explore the effect. However, the noise out of focus and the grayscale fluctuation between images impede the structure information extraction. This work proposes a method to segment the oxide cap, aluminum droplet and enveloping flame combining YOLOv7 detection and two-stage cluster segmentation, which incorporates position information. The method is operated with metric evaluation; the 86% Mean Intersection over Union (MIoU) and a concentrated distribution in the confusion matrix diagonal strongly prove the outstanding performances. The position, shape, and motion information of the structures can be extracted from the accurate segmentation results, which are essential for combustion mechanism revelation including reaction nonuniformity, combustion rate and motion impetus.

Keywords: Al/AP propellant, Flame segmentation, YOLOv7 detection network, Two-stage cluster