The Detection And Extraction Of Weak Signal In Low Frquency Radio Sky

With the development of radio technology,low frequency radio band will be the key direction of observational cosmology in the future 10 to 20 years.According to the big bang theory,universe entered into dark age and reionization age after that.In this age,the signal which called 21 cm signal was emitted at the wavelength 21 cm because of the ionization of neutral hydrogen,and it will be detected weakly with redshift to low frequency radio band.There are some other foreground radiations added together such as galactic synchrotron radiation,galactic free-free radiation,cluster,and discrete sources outside galaxy.Among these foreground radiations,cluster and 21 cm signal are the weakest components.Cluster is the biggest astrophysical object which is full of dark matter,stars and hot gas.It predicts that amounts of radio halos or sites will observed in low frequency radio band which hold important information of cluster evolution.And the extraction of 21 cm signal is also significant for studying the evolution process of early universe and the generation of first astronomical objects.The research content is divided into two parts in this dissertation.The first part is the detection and extraction of the clusters.As for the extraction of the clusters on the high-precision simulation of low-frequency radio image,an approach based on independent component analysis,hough transform and support vector machine was proposed.Firstly,we used the ICA to pre-separate the image into 4 parts.Then we focus on the clusters and point sources.We use Hough transformed algorithm to localize the clusters and point sources,and utilize local-ICA to separated these two components more accurately.As for the detection of clusters after extraction,feature samples was extracted from them such as surface brightness and statistical information to train SVM and classify them.Because of the number of point source is much more than cluster,the traditional SVM is not applicable and we use OCSVM to train the model and get good performance of cluster detection.Result shows the detection rate is improved with our method.The second part is for the extraction of 21 cm HII signal,the temperature of which is typically 4 or 5 orders of magnitude less than foreground radiation,so it is quite difficult to extract it.The usual method was to fit the foreground signal at the line-of-sight using polynomials,and subtracted them from the original combined signal,the remained component,i.e.the residual was see as the 21 cm HII signal.In this paper,we take the idea of the popular method as consideration,and proposed a new foreground fitting approach.We use the wavelet reconstruction algorithm to estimate the foreground components,and then get the residual 21 cm HII signal.Compared with the curve fitting method using polynomial,the reconstruction of the signal using our method performs better.