NVST Interference Fringes Classification Recognition And Removal Based On Deep Learning

The Sun is very important for human life,and it is constantly sending energy to our earth all the time.Solar activity has a huge impact on the Earth's environment.Therefore,it is particularly important to study the internal structure of the sun and predict the solar activity by observing the sun.One meter New Vacuum Solar Telescope(NVST)is the largest vacuum solar telescope in China.Due to the CCD used in the telescope system for data imaging,equal thickness interference fringes caused by film interference are produced,which has a great impact on the quality of NVST observation data products.In order to improve the quality of NVST data,this paper analyzes the fringe features of interference fringes in NVST observation data,combined with the advantages of automatic feature extraction based on deep learning,aims to develop a method of classification,recognition and removal of interference fringes based on deep learning.Firstly,according to the characteristics of interference fringes in NVST observation data,the periodicity and shape characteristics of interference fringes in observation data are studied.Combined with deep learning method,the application of deep learning in image classification and solar physics is studied,and a suitable method for NVST interference fringes classification and removal is found.Then,based on the common convolutional neural network for classification and recognition,the interference fringe enhancement image decomposed by Adaptive Wavelet Transform(AWT)is used to make the NVST interference fringe recognition data set.For different network structures,through a series of evaluation parameters to evaluate the network structure,select the appropriate network structure and the appropriate network structure parameters,and get the training model with high classification and recognition accuracy.The model is used to recognize NVST short exposure data with fringes,and has high classification accuracy for NVST interference fringes.Then,the NVST interference fringe removal model based on residual learning is used as the interference fringe removal tool,the NVST short-exposure data without interference fringe is used as the input of the NVST interference fringe removal model based on residual learning,and the data corresponding to superimposed interference fringe is used as the output of the NVST interference fringe removal model based on residual learning.The Peak Signal-to-Noise Ratio(PNSR)is used to evaluate the influence of different network structure parameters on different network structures,and the NVST interference fringe removal model with high PNSR and optimal network structure parameters is obtained.The model is used to remove the interference fringes of NVST short-exposure data,and good results are obtained.Finally,the automatic processing of NVST data products is formed by combining the stripe removal model and recognition model trained by deep learning.In the experiment of automatic processing of NVST data products using NVST short-exposure data,the NVST interference fringes recognition and removal methods based on deep learning are used respectively.The results show that the proposed method is reasonable and effective for NVST short exposure data.