The compact torus (CT) injection device, widely known as a magnetized coaxial plasma gun, creates self-contained magnetic field structures, known as plasmoids, which exhibit high densities and velocities. Owing to its remarkable energy density, the CT injection device holds immense potential for tokamak core fueling, rendering it promising for future fusion reactor applications. This paper presents a novel algorithm, comprising a segmentation module based on the UNet neural network and a tracking module leveraging the simple online and real-time tracking (SORT) algorithm, developed for detecting and tracking plasmoids in visible images. The algorithm is specifically designed for the recently manufactured CT injection system of the EAST tokamak, known as EAST-CTI [Kong et al., Plasma Sci. Technol. 25(6), 065601 (2023)]. Our analysis reveals the presence of multiple plasmoids within the plasma flow ejected by the EAST-CTI system. The UNet convolutional neural network successfully detects these plasmoids, achieving a dice coefficient of 0.813 on the test dataset, indicating high accuracy. Meanwhile, a modified version of the SORT algorithm successfully tracks these plasmoids, demonstrating robust performance without false tracking or identity assignment errors. Overall, the developed algorithm offers critical insights into the evolution characteristics of CTs and meets the requirements of the EAST-CTI system's visible imaging diagnostics. This advancement creates a favorable environment for extensive data analysis using imaging data in future research endeavors.
© 2024 Author(s). Published under an exclusive license by AIP Publishing.