The National Spherical Torus Experiment (NSTX) at the Princeton Plasma Physics Laboratory (PPPL) is a magnetic fusion device based on the spherical tokamak concept. It explores a novel structure for the magnetic field used to contain the hot ionized gas, called plasma. The success of devices, such as NSTX, is dependent on two key factors - first, the plasma has to be hot enough so the particles can fuse, and second, it has to be confined long enough so that the particles do fuse. This can be challenging for many reasons, one of which is the presence of fine-scale turbulence which causes leakage of plasma and particles from the center of the device to the edge. This leakage could result in a significant heat loss from the plasma, loss of confinement of the particles, as well as erosion or vaporization of the containment wall of the reactor.

To understand the physical mechanisms behind this fine-scale turbulence, physicists are using 2-D optical imaging technology to obtain images of the edge turbulence in experimental reactors such as NSTX. The images are taken using a ultra high-speed, high resolution camera, with each sequence consisting of 300 frames taken at 250,000 frames per second. Each 16 bit frame is 64x64 pixels. These images have indicated the presence of coherent structures, which retain their geometry over many characteristic lengths of motion.

Our work focuses on the use of image analysis techniques to identify these coherent structures (or 'blobs') in the images, extract characteristics for them, and track them over time. This is challenging as the images are noisy, there is no definition of what constitutes a coherent structure, and there is no ground truth to verify our analysis techniques. Our early work focused on denoising the images, removing the background (or quiescent) intensity, and experimenting with various segmentation algorithms to separate the blobs from the background (see Figure 1).

Figure 1. Processing of a single image from a 300 frame sequence: from left to right - raw image; after denoising; the background plasma; the image after the background has been divided out.
Figure 1. Processing of a single image from a 300 frame sequence: from left to right - raw image; after denoising; the background plasma; the image after the background has been divided out.

While it is possible to segment the blobs in a single image, it is harder to find a single algorithm which can identify the blobs in all images in a sequence using a fixed set of parameters. This is because there is a wide variation in the images of a sequence (see Figure 2).

Figure 2. Twenty consecutive images from a sequence (after de-noising and removal of the background) illustrating the diversity within a sequence.
Figure 2. Twenty consecutive images from a sequence (after de-noising and removal of the background) illustrating the diversity within a sequence.

Our current work focuses on identifying segmentation techniques that are robust so they can be applied to all images in a sequence and finding techniques for identifying the background for longer sequences composed of thousands of frames.

This work is a collaboration with Drs. Ricardo Maqueda and Stewart Zweben from the Princeton Plasma Physics Laboratory.

Publications

For papers and presentations related to this blob tracking project, see StarSapphire Publications.