A detection method of Edge Coherent Mode based on improved SSD

• Published in: Fusion engineering and design 2022-06, Vol.179, p.113141, Article 113141
• Main Authors: Zeng, Fulin, Liu, Ying, Ye, Yang, Zhou, Jijun, Liu, Xiaotao
• Format: Article
• Language: English
• Subjects: Data analysis; Deep learning; EAST; Edge coherent mode; Flooding; Machine learning; Object detection; Target detection
• ISSN: 0920-3796; 1873-7196
• DOI: 10.1016/j.fusengdes.2022.113141

•Single-shot multibox detector (SSD) is studied to apply to detect ECM.•The positive-negative sample ratios of SSD are adjusted to improve detection accuracy.•The flooding loss is imported to suppress overfitting of SSD and improve model robustness and detection accuracy.•The proposed improved detection method applies for subsequent intelligent data analysis and provides reference for physical events detection in fusion experiments. It has become a trend to deeply analyze and mine diagnostic data using machine learning methods. Single-shot MultiBox detector (SSD) is an excellent target detection network in deep learning. In this paper, two improvements for SSD are proposed to accurately detect the edge coherent mode in the cross-power spectrum. One is adjusting the positive–negative sample ratios to balance the training dataset, and the other is importing the flooding loss to mitigate overfitting. After testing, the accuracy of the improved SSD reaches 92.12%. The result shows that the performance of the improved SSD for ECM detecting is better than that of the typical SSD, which can be applied to subsequent intelligent data analysis.