X-Ray Pulsar Signal Recognition Based On Extreme Learning Machine

Due to the high reliability,the strong autonomy and the wide range of application,X-ray Pulsar-based Navigation(XPNAV) has become the new great significant methods of celestial navigation. The key step of XPNAV is to recognize the X-ray pulsar fastly and accurately by using the detected X-ray pulsar photon signals. Therefore, a fast and effective recognition algorithm has a great significance in promoting the study of XPNAV.The traditional recognition algorithms of pulsar signal have two key steps. The first step is to acquire the translation invariant and scale-invariant of X-ray signal which is based on the higher order spectrum of integrated pulsar profile. The second step is to use the invariant got form the first step to match the invariant of standard pulsar profile. The traditional recognition algorithms have made a good recognition results. However, integrated pulsar profile is built after processing the arriving X-ray photon signals. And the calculation of higher order spectrum need amount of time and storage space. Therefore, the traditional recognition algorithm has great limitations in the practical application. To solve the above issues, this paper presents a new X-ray signal recognition algorithm based on the Kernel Extreme Learning Machine which processes the X-ray photon signals directly. Firstly, generated the non-homogeneous Poisson model of the X-ray pulsar photon signals; Secondly, selected the number of photons which is from different sub-intervals within the same period constitute a set of vectors, then normalization as the feature vector; Finally, recognized the X-ray signals with K-ELM.ELM is a new machine learning algorithm based on single hidden layer feed-forward networks. The input weights and the hidden layer biases of ELM can be randomly assigned if the activation functions in the hidden layer are infinitely differentiable.Once random values have been assigned to these parameters in the beginning of learning, all of the parameters of ELM are not need to be adjusted and the output matrix H can actually remain unchanged. Thus, this algorithm can just simply get the output value of the hidden layer by one step. Simulation results show that the K-ELM can achieve the similar recognition,at the same time, reduce the computational complexity and improve the system recognition rate which is compared with the traditional recognition algorithm. The K-ELM owns the higher research value and practical value.