Solar radio spectrogram segmentation algorithm based on improved fuzzy C-means clustering and adaptive cross filtering

Abstract

Abstract Solar radio spectrograms contain essential information, such as the duration type; therefore, recognizing and detecting solar radio spectrograms are significant for the further study of solar radio. With the upgrading of solar radio observation, considering the equipment that has already generated amounts of data, researchers have begun to use machine learning methods to recognize and detect solar radio spectrograms to resolve the weaknesses of manual identification, such as time consumption. However, the spectrograms are characterized by noise or insignificant outburst features, which affect the recognition and detection of solar radio spectrograms. In contrast, extracting the burst region separately and the more distinctive spectrogram features will help identify and detect it. Therefore, to remove the burst domain of the radio spectrogram better, this paper combines the idea of image segmentation and proposes a solar radio spectrogram segmentation algorithm based on improved fuzzy C-means (FCM) clustering and adaptive cross filtering for the extraction of the burst domain of solar radio spectrograms. This algorithm has multiple processing steps. The first step is solar radio spectrogram segmentation with the improved FCM based on the kernel-induced distance by incorporating spatial constraints combined with random walk and adaptive affiliation linking (RWAKFCM_S). The second step is adaptive cross filtering, eliminating the noise clustered in bursts. The results show the following. (1) The RWAKFCM_S proposed in this paper has better anti-noise and segmentation performance than other methods in the synthetic, natural, and solar radio spectrogram segmentation experiments; it can also overcome the problems of noise sensitivity when segmenting spectrograms by traditional FCM. (2) The RWAKFCM_S can satisfy the high accuracy and rate of solar radio spectrogram segmentation demands. (3) The adaptive cross filtering proposed in this paper can eliminate noise clustered in the eruption domain. (4) The proposed method enables burst region extraction.