Improvement Of Feature Extraction And Intelligent Identification Method For The Edge Coherent Mode In East

The edge coherent mode (ECM) is a promising operational regime in the Experimental Advanced Superconducting Tokamak (EAST). In order to understand the physical mechanisms of ECM thoroughly, the automatic recognition of ECM is necessary. In this work, we define an adaptive denoising inequality based on the variance, and propose a dual-channel convolution feature extraction model to obtain a visualized ECM feature on which to perform ECM recognition. The new ECM recognition approach proposed in this paper provides better performance than the ECM recognition approach based on the feature of the row information points in the spectrogram, in terms of both precision and recall. In addition, the proposed approach is compared to four classical deep learning algorithms, and is found to provide competitive performance with a small amount of data.