Automatic Detection of Edge Coherent Modes in EAST Using a Modified YOLO Model

To enable efficient, automated detection of edge coherent modes (ECMs) in the Experimental Advanced Superconducting Tokamak (EAST) and thereby facilitate investigations of their underlying physics and support steady-state high-confinement-mode (H-mode) operation, we developed an ECM detector based on a modified YOLOv11s architecture. An ECM image dataset was constructed from helium-beam emission spectroscopy (He-BES) measurements, and its characteristic visual features were systematically analyzed. To accommodate the distinctive properties of ECMs, three modifications were introduced into the baseline model: a squeeze-and-excitation (SE) attention mechanism to enhance the extraction of informative channel features; a SlimNeck architecture to reduce computational redundancy; and a composite loss function, termed Inner-Focaler-CIoU, that integrates Inner-IoU and Focaler-IoU to improve bounding-box regression. The integrated model achieved AP@50 and AP@50–95 values of 0.965 and 0.665, respectively, on the test set. As a practical demonstration, the model was used to generate scatter plots relating detected ECM events to the key physical parameters q95 and βp, illustrating its utility for screening and visualizing experimental data. This study provides an accurate and efficient approach to automated ECM detection and establishes a reliable foundation for further physics-based analyses of ECM behavior.