Research On Feature Recognition Method Of Weak Emission Line Of LAMOST Low-resolution Spectra

Feature recognition is one of the research hotspots of data mining and pattern recognition.At present,many feature recognition methods are applied to specific fields,such as astronomy,meteorology,finance,medicine,and so on.With the continuous development of technology and the launch of the sky survey program,the LAMOST astronomical telescope has observed tens of millions of low-resolution celestial spectra.The weak emission line characteristics that may exist in the spectra are important materials for us to further understand the Milky Way.Low-resolution spectral emission lines have characteristics such as rarity,complexity,and diversity,which greatly increases the difficulty of discovery and identification.This paper conducts an in-depth study on the measurement and identification methods of weak emission line features in LAMOST low-resolution spectra.The main research contents include:(1)A method for measuring the characteristics of low-resolution spectral emission lines is proposed.Aiming at the high-dimensional sparse features of low-resolution spectra,this paper proposes a low-resolution spectral emission line feature measurement method.This method first extracts the wavelength and flux of the low-resolution spectra,and then shape according to the peak value on the left and right sides is used to calculate the distance confidence and the waveform confidence,and takes the sum of the distance confidence and the waveform confidence as the low-resolution spectral emission line confidence.This method converts the abstract low-resolution spectral emission line into a specific confidence value.Experimental results show that this feature measurement method can effectively extract low-resolution spectral emission line feature,which lays the foundation for the subsequent low-resolution spectral weak emission line feature identification method.(2)A feature recognition algorithm for weak emission lines of low-resolution spectra is proposed.Aiming at the problem that it is difficult to identify weak emission lines in low-resolution spectra,this paper uses the weight ranking and momentum formula to propose a low-resolution spectral weak emission line feature identification algorithm WEDA.First,the low-resolution spectral emission line feature measurement method is used to preprocess the low-resolution spectral data.The feature of the specified emission line and the feature of the surrounding low-resolution spectral data emission line are extracted.The difference between the specific data is used to initialize the feature weights,and then momentum formula is used to continuously iteratively update the weights to identify the weak emission line characteristics of the low-resolution spectra.The experimental results show that compared with other similar classification algorithms,WEDA shows higher accuracy,and is not affected by the amount of data and the signal-to-noise ratio,and can effectively identify the characteristics of low-resolution spectral weak H? emission lines.(3)A parallel algorithm for feature recognition of weak emission lines in low-resolution spectra is proposed.Aiming at the low computational efficiency of the low-resolution spectral weak emission line recognition algorithm when processing massive spectra,this paper proposes a parallel algorithm SparkWEDA based on the parallel computing framework called Spark.First,the massive low-resolution spectra are partitioned into RDD,and then these partitions are allocated to each cluster node.The partitioned data in each node is processed in parallel through multiple threads,and then the results of each node are combined to obtain final result.The experimental results show that the SparkWEDA algorithm greatly shortens the running time while ensuring the accuracy and recall rate,and performs well in performance indicators such as speedup and scalability,which can meet the processing needs of massive low-resolution spectra.