Research On The Pulsar Signal Recognition Based On Wavelet Kernel Extreme Learning Machine

Because of the complex electromagnetic environment of the universe,X ray pulsar signal recognition as one of the key technologies of pulsar navigation is the foundation of the spacecraft attitude determination positioning,and has very important research significance.The usual method of pulsar identification is to analyze the signal,and the correlation features are extracted,and comparison.However,these methods are based on the accumulated pulse profile,and some pulsars accumulate identifiable profiles in a shorter time,while some other pulsars can still not get satisfactory profiles after a long period of accumulation.The method of machine learning is widely used and effective.Some scholars have introduced the method of machine learning into the identification of pulsars,and the pulsars can be successfully identified under certain conditions,and the method of machine learning is extremely fast and suitable for space navigation whose resources are limited.The pulsar identification method based on machine learning is preliminary,there is still a lot of research space to get better results.Extreme learning machine(ELM)has the advantages of easy to use,fast and effective,but classification effect of the pulsar identification based on the ELM is susceptible to noise,that will lead to a significant reduction in the classification accuracy.The classification accuracy of the pulsar signal can be improved by using the kernel ELM,which can improve the classification accuracy of the recognition algorithm under the noise condition and have the remarkable research value to achieve better classification performance.The kernel function is the core part of the kernel ELM classifier.For the selection of kernel function and the performance of different kernel function are worthy of discussion.Gaussian kernel function and wavelet kernel function are now widely used and their effect is better.Because of the large number of wavelet functions,the wavelet function which can be used as kernel function is also more.This paper only uses Morlet as the research object and simulates the experiment.At the same time,the choice of kernel function parameters is also an important factor affecting the performance of the kernel ELM classifier.Therefore,it is necessary to select the appropriate kernel parameter optimization algorithm and the optimal kernel parameters to obtain better identification effect.Considering that the pulsar signal is a weak signal,and the electromagnetic environment of the universe is complex,firstly,the pulse signal is analyzed,and the pulsar signal noise and the corresponding denoising method are used to make the pulsar identification more effective on the basis of guarantee defficiency.The key technology of kernel ELM is the choice of kernel function and kernel parameter.In this paper,the selections of kernel function and kernel parameter are discussed,and the influence of different kernel function on the identification of pulsar is compared.The influence of the selection of the kernel parameters on the identification of the pulsar is experimentally analyzed,and the results show that the maximum performance of the kernel ELM cannot be obtained with the parameters set according to the experience.In this paper,a wavelet kernel ELM based on particle swarm optimization to identify the pulsars is proposed.It can be seen from the experiment that the particle swarm optimization algorithm can optimize the kernel parameters more quickl y than the grid search method.The efficiency and accuracy of the identification of pulsar signals in different situations are greatly improved.Experiments show that the pulsar identification algorithm based on wavelet kernel ELM can quickly and effectively perform pulsar identification.