Solar Filament Recognition Based on Improved VNet

The Solar filaments is a common solar activity, and the efficient and accurate detection of the solar filaments is of great significance to the study of the solar magnetic field. Due to the fact that the full-plane image has the characteristics of dimming, uneven brightness, and the inaccuracy of the labeled data set, the result of dark strip segmentation is not accurate, weak and small filaments are missed, and darker background parts are falsely detected as filaments, and with other issues. Therefore, a method for detecting solar filaments based on improved VNet network is proposed. First, the solar dark stripe data set is made by combining the solar magnetic map; secondly, the inception module is used to fuse the features of different scale feature maps in the down-sampling part of the VNet network, and the attention mechanism is added to enhance the semantic information of the dark stripe part of the feature map; Finally, a deep supervision module is introduced in the up-sampling part to retain more detailed features of the sun filaments. In order to verify the performance of the algorithm, we use a data set of 191 full-disk solar images, which contains a total of 3 372 filaments. The average accuracy of the algorithm on the test data set is 0.988 3, the F₁ value is 0.838 5. The experimental results prove that this method can effectively identify the filaments in the full-disk solar images.