| 其他摘要 | In recent years, the spectral data of celestial bodies observed have achieved a dramatic increase thanks to the successive implementation of various projects of spectral sky survey. Therefore, higher requirements for the automatic classification and analysis of spectrum are proposed for large-scale projects of spectral sky survey. In this paper, we design a deep learning method for MK classification of stellar spectra.In the classification of spectral type, the classification problem is transformed into a regression one in this paper, and a method of spectral category regression based on the depth residual network is put forward to conduct a prediction of MK spectral subtype on stellar spectrum. Convolution layers, maximum pooling layer, average pooling layer, full connection layer and residual structures are used to make up this model. The maximum pooling layer is used to filter features, the convolution layer to extract features, and the average pooling layer to reduce parameters and improve efficiency. Residual structure can prevent the degradation of network, extract high-dimensional abstract features by deepening the network and improve training speed. Considering the non-zero probability of data with false labels and corrupted data, Log-Cosh is adopted as loss function in this paper to reduce the negative impact of bad samples. The experimental data are randomly extracted from LAMOST, SDSS and other spectral libraries. The spectra are divided according to the proportion of 7:1:2 after eliminating the bad value and normalizing the flow. The experiment was divided into two parts. In the first part, the network is adopted to carry out prediction on the spectral subtype, and the maximum absolute error, the average absolute error and the standard deviation are used to compare the performance of convolution kernels with different shapes. The predicted value is taken as the abscissa and the label as the ordinate, and the second-order nonlinear fitting is used for all sample points in the test set, a straight line that is coincident with y = x is obtained, proving that the model can predict the spectral subtype well. The second part is concerning the internal analysis of the model. The main characteristics of the model in predicting four types of spectra, A, F, G, K, are mainly explored with the method of category activation mapping, thus endowing the model with interpretability. In text data set, the average absolute error of the prediction is 0.4 spectral subtypes. It is shown that the method proposed in this paper can better predict spectral subtypes with faster speed and higher accuracy according to the comparison of same data set with nonparametric regression, AdaBoost regression tree and K-means.In the classification of luminosity level, As the luminosity level classification depends on the subtle changes of spectral lines, this paper uses multi-scale convolution model to extract spectral line features, mainly including convolution layer, ReLU activation layer, batch normalization layer, softmax layer and full connection layer. Due to the scarcity of labeled data, this paper uses the idea of transfer learning to pre train the model on the spectral data set, and then fine tune it on the luminosity data set. Due to the imbalance of data sets, this paper uses focal loss as the loss function to improve the training effect. |
修改评论