EDGE DETECTION AND MACHINE LEARNING FOR AUTOMATIC FLOW STRUCTURES DETECTION AND TRACKING ON SCHLIEREN AND SHADOWGRAPH IMAGES

Visual information in experimental fluid dynamics is enlarging now to the level of big scientific data due to usage of digital cameras. Flow images and animations are the objects of digital image processing where different algorithms can be applied. Schlieren, shadow, and other refraction-based techniques have been often used to study gas flow. They can capture strong density gradients, such as shock waves. Shock detection is an important task in analyzing unsteady supersonic gas flows. High-speed cameras are widely used to record large arrays of shadow images. In this paper, two computer software systems based on the edge detection and machine learning with CNN were developed to process datasets of the shadow images and automatically detect shock waves, plumes, and other gas flow structures. The edge-detection software uses image filtering, noise removing, background image subtraction, and edge detection based on the Canny algorithm. The machine learning software is based on CNN. A full object detection network was trained to identify two different types of objects on the schlieren or shadow images: plumes and shock waves. We applied transfer learning to decrease learning time and number of images for training. Shock detection was tested on a flat shock wave moving in a shock tube. Plume detection was tested on the shadow flow images, initiated by the pulsed surface gas discharge. It was shown that both edge detection and machine learning may be successfully applied for gas flow structures tracking and measurements. The dynamics of the original and the reflected shock waves and the position and size of the plume were automatically measured by the developed software.