Abstract:In pulsar astronomy, detecting effective pulsar signals among numerous pulsar candidates is an important research topic. Starting from space X-ray pulsar signals, the two-dimensional autocorrelation profile map (2D-APM) feature modelling method, which utilizes epoch folding of the autocorrelation function of X-ray signals and expands the time-domain information of the periodic axis, is proposed. A uniform setting criterion regarding the time resolution of the periodic axis addresses pulsar signals without any prior information. Compared with the traditional profile, the model has a strong anti-noise ability, a greater abundance of information and consistent characteristics. The new feature is simulated with double Gaussian components, and the characteristic distribution of the model is revealed to be closely related to the distance between the double peaks of the profile. Next, a deep convolutional neural network (DCNN) is built, named Inception-ResNet. According to the order of the peak separation and number of arriving photons, 30 data sets based on the Poisson process are simulated to construct the training set, and the observation data of PSRs B0531+21, B0540-69 and B1509-58 from the Rossi X-ray Timing Explorer (RXTE) are selected to generate the test set. The number of training sets and the number of test sets are 30,000 and 5,400, respectively. After achieving convergence stability, more than 99$\%$ of the pulsar signals are recognized, and more than 99$\%$ of the interference is successfully rejected, which verifies the high degree of agreement between the network and the feature model and the high potential of the proposed method in searching for pulsars.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to effectively detect real pulsar signals among a large number of pulsar candidate signals. Specifically, the paper focuses on how to quickly and accurately identify effective low signal - to - noise ratio (S/N) pulsar signals from numerous pulsar candidate signals in the X - ray band. The importance of this problem lies in that traditional feature extraction methods have problems such as weak noise resistance and insufficient information when dealing with low signal - to - noise ratio signals. Especially in space observations, due to the limited receiving area and sparse data statistical information, traditional pulsar search algorithms are difficult to be effectively applied. To solve the above problems, the author proposes a method that combines two - dimensional autocorrelation profile map (2D - APM) and deep convolutional neural network (DCNN). By introducing the superposition of autocorrelation and folded profile map, this method can enhance the consistency of signal features and the ability to retain details without relying on prior information, while suppressing noise interference and shape distortion. In addition, the paper also constructs a deep convolutional neural network named Inception - ResNet to learn the spatial information of 2D - APM, so as to achieve efficient classification of pulsar candidate signals. ### Specific solutions: 1. **Feature representation**: - **2D - APM**: By performing autocorrelation processing on the X - ray photon arrival time (TOA) and performing superposition of folded profile maps near the test period, a two - dimensional autocorrelation profile map (2D - APM) is generated. This method not only enhances the consistency of signal features, but also retains more detailed information and improves the noise resistance ability. - **Time resolution setting criterion**: In order to deal with pulsar signals with unknown periods and masses in practical applications, a time resolution setting criterion for the period axis is proposed to ensure the consistency of the extracted features. 2. **Feature perception**: - **Deep convolutional neural network (DCNN)**: A deep convolutional neural network is constructed by using Inception and ResNet modules to learn the spatial information of 2D - APM. By using the simulation method of non - uniform Poisson process to generate the training set, the network can identify pulsar candidate signals with high sensitivity. - **Network structure optimization**: The Inception module and the residual module (ResNet) are adopted, and the network structure is optimized by techniques such as batch normalization and dimension reduction of 1×1 convolutional layer, which improves the training efficiency and generalization ability. ### Experimental verification: - Tests were carried out using the data of Rossi X - ray Timing Explorer (RXTE). The results show that the network can identify pulsar signals with an accuracy of more than 99% and successfully reject more than 99% of interference signals, verifying the effectiveness and high potential of the proposed method. In conclusion, this paper solves the problem of efficiently identifying pulsar signals in the X - ray band by proposing a new feature representation method and a deep - learning model, providing new tools and methods for pulsar research.

A Method for Pulsar Searching: Combining a Two-dimensional Autocorrelation Profile Map and a Deep Convolutional Neural Network

Search for Gravitational Wave Probes - A Self-Supervised Learning for Pulsars Based on Signal Contexts

Adaboost-DSNN: an adaptive boosting algorithm based on deep self normalized neural network for pulsar identification

Application of Artificial Intelligence in the Selection of Pulsar Candidate

Pulsar Candidate Classification Using Generative Adversary Networks

Searching for pulsars using image pattern recognition

Pulsar candidate identification using advanced transformer-based models

Pulsar Candidate Selection Using Ensemble Networks for FAST Drift-Scan Survey

Pulsars Detection by Machine Learning with Very Few Features

Selection of radio pulsar candidates using artificial neural networks

Pulsar Candidate Classification Using A Computer Vision Method Combining with Convolution and Attention

The Quick Search Algorithm of Pulsar Period Based on Unevenly Spaced Timing Data

A novel single-pulse search approach to detection of dispersed radio pulses using clustering and supervised machine learning

MFPIM: A Deep Learning Model Based on Multimodal Fusion Technology for Pulsar Identification

PEACE: Pulsar Evaluation Algorithm for Candidate Extraction -- A software package for post-analysis processing of pulsar survey candidates

Hierarchical signal fusion network for pulsar detection with phase-correlation and signal attentions

Identifying pulsar candidates in interferometric radio images using scintillation

An energy-based pulsar period estimation method using Hilbert curve and double CNNs

Progress on Radio Pulsar Survey and Estimate the Neighbor Pulsar Population

A Novel Pulsar Profile Extraction Approach for Synchronization Measurement and Its Hardware Implementation