To address the identification of internal component damage to the first wall during the EAST discharge process, the YOLOv8 model is introduced for hotspot recognition, presenting an improved EAST-YOLOv8 algorithm based on the traditional YOLOv8 method. This method can effectively assist manual detection and improve the efficiency of damage detection of internal components in EAST discharge. YOLOv8 behaves poorer in early warning of internal component damage during EAST discharges. According to engineering and physical characteristics of this problem, targeted improvements have been made to YOLOv8,including adding a small object detection layer and incorporating the Convolutional Block Attention Module(CBAM). Then a network suitable for detecting internal component damage under the conditions of long-pulse,high-parameter discharges in EAST has been constructed. Experimental results demonstrate the significant advantages of EAST-YOLOv8 over the original YOLOv8 in terms of loss function and recall rates.EAST-YOLOv8 achieves a detection accuracy of 93.6%, mAP@0.5 of 93.5%, and a recall rate of 93.9%, which offers more efficient and reliable support for the safety and stability of EAST operation.

YOLOv8, EAST, Tokamak, Small object detection

针对 EAST 装置放电过程中第一壁内部部件损伤的识别问题，首次引入 YOLOv8 模型进行热斑识别，提出一种基于改进 YOLOv8 的 EAST-YOLOv8 算法。该方法可有效协助人工检测，提升 EAST 装置放电内部部件损伤检测效率。传统 YOLOv8 网络在正常目标检测任务中表现优越，但对于 EAST 装置放电初期内部部件损伤的预警方面存在不足。根据该问题的工程与物理特点，对 YOLOv8 进行了具有针对性的改进，包括增加小目标检测层和加入 CBAM 注意力机制。从而构建了适用于 EAST 装置长脉冲高参数放电条件下内部部件损伤检测的网络模型。实验结果表明，EAST-YOLOv8 比传统YOLOv8 在损失函数和召回率上有明显优势，检测精确率达到 93.6%、mAP@0.5 达到 93.5%、召回率达到 93.9%。这一创新性的方法为 EAST 装置运行的安全性和稳定性提供了更为高效和可靠的支持。

YOLOv8, EAST, 托卡马克, 小目标检测