Gamma instant radiation pulse event selecting and triggering method

Abstract

The invention discloses a method for selecting and triggering gamma instant radiation pulse events, and belongs to the technical field of nuclear radiation detection. The method comprises the following steps: carrying out normalization preprocessing on an original gamma pulse radiation time spectrum; s time-frequency transformation is carried out on the preprocessed data, and time-frequency features are extracted; optimizing parameters of the frequency threshold and the amplitude threshold after S time-frequency transformation by adopting a deep reinforcement learning model to obtain an optimal frequency threshold and an optimal amplitude threshold; performing filtering processing on the signal of the S time-frequency conversion result, and separating noise from an effective signal; and searching a pulse radiation event for the filtered signal by applying a morphology-based peak searching method, and positioning the position and the characteristic of a real gamma instant radiation pulse event. According to the method, time-frequency analysis is carried out through S time-frequency transformation, the problem that the frequency of an analysis window cannot be adjusted through short-time window Fourier transformation is solved, the method has better frequency resolution and phase retention characteristics, and gamma pulse signal characteristics can be represented more accurately.

Description

Technical Field

[0001] The present invention belongs to the technical field of nuclear radiation detection, and in particular relates to a method for selecting and triggering a gamma prompt radiation pulse event. Background Art

[0002] Gamma-ray prompt radiation events originate from nuclear physics phenomena and high-energy astrophysical processes. For example, gamma-ray bursts (GRBs) are the most energetic, transient explosive phenomena in the universe. Since their accidental discovery by the US Vela satellite in 1967, GRBs have been a key topic in high-energy astrophysics research. GRBs are believed to be produced by extreme astrophysical processes such as the collapse of massive stars (long GRBs) or the merger of neutron stars (short GRBs). Their short duration, unpredictability, and non-repeatability make their observation and identification extremely challenging.

[0003] Currently, the detection of gamma-ray prompt radiation pulses (GRBs) relies primarily on high-energy detectors aboard orbiting satellites, such as NASA's Swift satellite, the Fermi Gamma-ray Space Telescope (Fermi), and China's Insight-HXMT (HXMT). These detectors continuously monitor high-energy radiation events in the universe, generating massive amounts of observational data daily. However, within this vast amount of data, how to quickly and accurately identify GRBs and extract valuable scientific information from them remains a critical technical challenge in the international GRB research community.

[0004] Traditional methods for detecting gamma-ray prompt radiation pulse events rely primarily on simple time-series threshold triggering techniques. This involves triggering the recording of an event when the photon count rate recorded by the detector exceeds a certain threshold of the background count rate within a short period of time. This approach has the following drawbacks: 1) it is prone to missing detections of weak gamma-ray bursts with a low signal-to-noise ratio; 2) it is prone to false alarms in the face of complex interference sources such as the cosmic ray background and solar activity; 3) it cannot effectively identify gamma-ray bursts with complex temporal structures; and 4) its processing speed is slow, making it inadequate for real-time warnings. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for selecting and triggering gamma-ray prompt radiation pulse events in response to the above-mentioned problems, hoping to improve the problems of low recognition accuracy, difficult threshold setting and low processing efficiency of traditional methods in low signal-to-noise ratio environments.

[0006] The technical solution adopted by the present invention is as follows: a method for selecting and triggering a gamma prompt radiation pulse event, the gamma prompt radiation pulse event selection-triggering method comprising the following steps: Performing normalization preprocessing on the original gamma pulse radiation time spectrum to obtain preprocessed data; Perform S time-frequency transformation on the preprocessed data, extract the time-frequency features, and obtain the time-frequency matrix; The deep reinforcement learning model is used to optimize the frequency threshold and amplitude threshold parameters after S time-frequency transformation to obtain the optimal frequency threshold and optimal amplitude threshold; Based on the optimal frequency and optimal amplitude thresholds, the signal of the S time-frequency transform result is filtered to separate the noise and the effective signal; The morphological peak-finding method is applied to the filtered signal to search for pulse radiation events and locate the position and characteristics of the real gamma-ray prompt radiation pulse events.

[0007] Furthermore, the normalization processing includes amplitude normalization and time normalization. Amplitude normalization uses the maximum-minimum normalization method to map the signal amplitude to the [0,1] interval and converts the sampling sequence to a uniform length through linear interpolation; time normalization converts the sampling points into a uniform time interval to obtain the preprocessed data.

[0008] Furthermore, the formula for S time-frequency transform is: ; in, is the time parameter, is the frequency parameter, is the input signal, is the Gaussian window function, The imaginary unit is, To represent the original signal, The time, width and frequency of the Gaussian window function Inversely proportional.

[0009] Furthermore, the deep reinforcement learning model is used to optimize the parameters after the S time-frequency transformation, which specifically includes the following steps: The deep Q network DQN is used to extract time-frequency features from the results of the time-frequency transformation of S, and the time-frequency feature matrix is ​​obtained, that is, the state space. The action space is the adjustable range of the frequency threshold and the amplitude threshold, and the optimal frequency threshold and amplitude threshold parameters are predicted; Reward Function : ; in, is the harmonic mean of precision and recall, is the weight coefficient; by The value is used as the optimization target, and the frequency and amplitude threshold parameters are iteratively optimized through reinforcement learning.

[0010] Furthermore, the filtering process includes the following steps: The components below the frequency threshold in the result of S time-frequency transform are set to 0; The components below the amplitude threshold in the result of S time-frequency transform are set to 0; The filtered time-frequency domain is converted back to the time domain signal through the inverse S time-frequency transform.

[0011] Furthermore, the morphological peak finding method includes the following steps: Perform dilation on the filtered signal to enhance potential peaks; Perform corrosion operation on the expanded signal to remove small fluctuations; Calculate morphological gradients and locate the points where the signal changes sharply; Valid pulse events are screened using the peak screening conditions of minimum peak height, minimum peak width, and minimum peak spacing.

[0012] Furthermore, the deep Q network DQN includes a hybrid network consisting of a convolutional neural network and a long short-term memory network. The convolutional neural network is used to extract local time-frequency features, and the long short-term memory network is used to capture the dependencies of time series.

[0013] Furthermore, the deep reinforcement learning model is implemented by combining offline training and online fine-tuning. Offline training uses a labeled gamma pulse dataset, and online fine-tuning dynamically adjusts model parameters based on real-time data.

[0014] Furthermore, the gamma prompt radiation pulse event selection-triggering method adopts a sliding window mechanism for processing real-time data streams, and sets the window length and sliding step size.

[0015] It should be noted that the window length and sliding step size are adaptively adjusted according to the characteristics of the gamma pulse signal.

[0016] The method of the present invention is applicable to signals acquired by multiple types of gamma-ray detectors, including scintillator detectors, semiconductor detectors and gas detectors; and has strong robustness in low signal-to-noise ratio environments containing high background radiation, electronic noise or complex interference sources.

[0017] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. The present invention performs time-frequency analysis through S time-frequency transform, overcoming the problem that the short-time window Fourier transform cannot adjust the analysis window frequency, and has better frequency resolution and phase retention characteristics. At the same time, the present invention introduces multi-resolution analysis of wavelet transform and maintains a direct connection with the Fourier spectrum, thereby being able to more accurately characterize the characteristics of gamma pulse signals, especially in low signal-to-noise ratio environments.

[0018] 2. The present invention combines deep reinforcement learning technology to achieve automatic optimization of frequency thresholds and amplitude thresholds, avoiding the subjectivity and inadaptability of manually setting thresholds in traditional methods, and improving the generalization ability of the system.

[0019] 3. The present invention adopts a morphological peak-finding method to analyze the morphological characteristics of gamma pulse events. Compared with the simple threshold triggering method, it greatly improves the detection accuracy and reduces the false trigger rate, especially in the case of pulse overlap.

[0020] 4. The present invention is implemented through a sliding window mechanism and an efficient algorithm, which is suitable for real-time processing of large amounts of detection data and meets the requirements of high-throughput nuclear detection systems.

[0021] 5. The present invention has adaptive learning capabilities, can adjust parameters according to actual data characteristics, adapt to different detection environments and equipment conditions, and has strong versatility and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flow chart of the overall method of the present invention; Figure 2 This is a structural diagram of the deep reinforcement learning framework of the present invention. DETAILED DESCRIPTION

[0023] The present invention will be described in detail below with reference to the accompanying drawings.

[0024] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0025] With the development of artificial intelligence and deep learning technologies, it has become possible to automatically identify gamma-ray bursts using advanced signal processing and pattern recognition techniques. However, due to the diversity and complexity of gamma-ray burst signals and the limited computing resources in an on-orbit environment, it is necessary to develop a new method for selecting and triggering gamma-ray prompt radiation pulse events. This method can quickly and accurately identify gamma-ray burst events in massive amounts of X-ray and gamma-ray observation data and provide timely warning information for subsequent multi-band joint observations.

[0026] The present invention has important scientific value and application significance for accurately identifying the characteristics of gamma-ray prompt radiation events and precisely determining their arrival time series parameters, efficiently processing massive detection data, optimizing event screening algorithms, and significantly improving the response performance of real-time trigger systems.

[0027] like Figure 1 As shown, a method for selecting and triggering a gamma prompt radiation pulse event includes the following steps: Step S100: Obtain the original gamma pulse radiation time spectrum data and perform normalization preprocessing. The normalization operation includes: Step S101: Amplitude normalization, using the maximum-minimum normalization method to process the signal amplitude x: ; in, and are the maximum and minimum values ​​of the signal, respectively, so that all signal amplitudes are mapped to the [0,1] interval; Step S102: Time normalization, unifying the data obtained at different sampling rates to a standard time interval to ensure consistency in subsequent analysis.

[0028] The purpose of the normalization preprocessing in step S100 is to eliminate the effects of differences in the responses of different detectors and changes in measurement conditions, and to provide a standardized data basis for subsequent analysis.

[0029] Step S200: Perform S transform on the normalized signal to obtain time-frequency representation. S transform is a linear time-frequency analysis method that combines the advantages of short-time Fourier transform and continuous wavelet transform, and is defined as: ; Compared with wavelet transform, S transform has the following advantages: (1) Frequency-adaptive window: The width of the window function changes with frequency. The window is narrow at high frequencies (high time resolution) and wide at low frequencies (high frequency resolution). (2) Phase reference: Maintaining the same phase reference as the input signal facilitates phase analysis; (3) Linear frequency axis: The frequency axis is a linear scale, which is intuitive and easy to interpret.

[0030] The time-frequency matrix obtained after S-transformation clearly shows the energy distribution of the gamma pulse signal at different times and frequencies.

[0031] Step S300: Figure 2 As shown in the figure, a deep reinforcement learning framework is used to automatically optimize the frequency threshold and amplitude threshold parameters, which includes the following sub-steps: S301: Construct a hybrid deep neural network, including convolutional layers, pooling layers, and LSTM layers, to extract time-frequency features from the S-transform results.

[0032] S302: The network output layer generates normalized frequency threshold and amplitude threshold prediction values ​​in the range [0, 1], which are then converted to actual thresholds: in, and is the normalized prediction value output by the neural network, ranging from [0,1], 、 is the actual frequency threshold and amplitude threshold after denormalization, 、 The actual minimum and maximum values ​​of the frequency threshold, 、 The actual minimum and maximum values ​​for the amplitude threshold.

[0033] S303: Apply the predicted threshold to the filtering and morphological peak finding process of the S-transform result, filter the S-transform result x(t), and obtain , and reuse structural elements right Perform dilation and erosion to calculate morphological gradients , and finally the predicted threshold is applied to , perform peak detection. After the test results are completed, the performance indicators such as accuracy, recall rate and F1 value are calculated. The morphological gradient formula is as follows: ; in, Represents a signal (or image) that has been filtered. Structuring element is a parameter in morphological operations. Represents a dilation operation. Represents an erosion operation. Represents the morphological gradient results.

[0034] S304: Perform reinforcement learning optimization, wherein a deep Q network (DQN) is used to implement reinforcement learning, which is defined as follows: Status: S-transform characteristics of the current data window; Action: Adjust the frequency threshold and amplitude threshold (discretize into multiple adjustment steps); award: ; in, is the weight coefficient, which is set to =1.0; DQN stabilizes the training process through experience replay and target network technology, continuously optimizes Q-value estimation, and ultimately selects the action strategy that maximizes long-term rewards, that is, the optimal threshold parameter combination.

[0035] Step S400: Based on the optimal frequency threshold and amplitude threshold obtained in step S3, filter the S-transform result: Step S401: The frequency of the S transformation result that is below the optimal frequency threshold is converted The frequency components of are set to zero:

[0036] ,when

[0037] ,when

[0038]

[0039] in, is the original S-transform result, τ is the time parameter, f is the frequency parameter, This is the optimal frequency threshold obtained by optimization in step S3. This operation can effectively remove low-frequency background noise and retain the high-frequency components in the gamma pulse event.

[0040] Step S402: The S transform result is lower than the optimal amplitude threshold. The components of are set to zero:

[0041] , when

[0042] 0, when

[0043]

[0044] in, This is the optimal amplitude threshold obtained by optimization in step S3. This operation can further suppress the low-amplitude components in the S-transform result and highlight the main energy contribution of the gamma pulse event.

[0045] Step S403: Transform the filtered time-frequency domain into Convert back to time domain signal:

[0046] in, Represents the inverse S-transform operation, and the calculation formula is: ; The inverse S transform reconstructs the filtered time-frequency representation into a time domain signal. After double threshold filtering, the noise component is significantly reduced, while the characteristics of the gamma pulse event are preserved and enhanced. The time-domain reconstructed signal will serve as the input for the morphological peak-finding method in the subsequent step S5 to accurately identify the location and characteristics of the gamma pulse event.

[0047] Compared with traditional time-domain filtering or fixed threshold processing methods, this step significantly enhances the gamma pulse signal, while the noise component is effectively suppressed, providing a high-quality signal basis for subsequent gamma pulse event selection.

[0048] Step S500: Applying a morphological peak-finding method to the filtered signal to accurately locate the gamma pulse event, including the following sub-steps: S51: Use structural element B to signal Perform the expansion operation:

[0049] The dilation operation enhances the peak regions in the signal, making potential gamma pulse events stand out more.

[0050] S52: Perform corrosion operation on the expanded signal:

[0051] The erosion operation removes small fluctuations and subtle noise in the signal, retaining the main pulse shape.

[0052] S53: Calculate the difference between the dilation result and the erosion result to obtain the morphological gradient:

[0053] The morphological gradient highlights areas of sharp signal changes, corresponding to the start and end of gamma pulse events. edge.

[0054] S54: Apply a peak detection algorithm to the morphological gradient signal and combine it with the following filter criteria: Minimum peak height threshold: h_min, excludes low-amplitude fluctuations; Minimum peak width threshold: w_min, excludes sharp noise pulses; Minimum peak spacing threshold: d_min, separates adjacent pulse events.

[0055] When the detected peak meets all the above conditions, it is determined to be a valid gamma pulse event, and its characteristic parameters such as position, amplitude, and duration are recorded.

[0056] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for selecting and triggering a gamma prompt radiation pulse event, characterized in that: The gamma prompt radiation pulse event selection and triggering method includes the following steps: Performing normalization preprocessing on the original gamma pulse radiation time spectrum to obtain preprocessed data; Perform S time-frequency transformation on the preprocessed data, extract the time-frequency features, and obtain the time-frequency matrix; The deep reinforcement learning model is used to optimize the frequency threshold and amplitude threshold parameters after S time-frequency transformation to obtain the optimal frequency threshold and optimal amplitude threshold; Based on the optimal frequency and optimal amplitude thresholds, the signal of the S time-frequency transform result is filtered to separate the noise and the effective signal; The morphological peak-finding method is applied to the filtered signal to search for pulse radiation events and locate the position and characteristics of the real gamma-ray prompt radiation pulse event.

2. A method for selecting and triggering a gamma prompt radiation pulse event according to claim 1, characterized in that: Normalization processing includes amplitude normalization and time normalization. Amplitude normalization uses the maximum-minimum normalization method to map the signal amplitude to the [0,1] interval and converts the sampling sequence to a uniform length through linear interpolation; time normalization converts the sampling points to a uniform time interval to obtain preprocessed data.

3. A method for selecting and triggering a gamma prompt radiation pulse event according to claim 2, characterized in that: The formula for S time-frequency transform is: ； in, is the time parameter, is the frequency parameter, is the input signal, is the Gaussian window function, The imaginary unit is, To represent the original signal, The time, width and frequency of the Gaussian window function Inversely proportional.

4. A method for selecting and triggering a gamma prompt radiation pulse event according to claim 3, characterized in that: The optimization of the parameters after the time-frequency transformation of S using the deep reinforcement learning model specifically includes the following steps: The deep Q network DQN is used to extract time-frequency features from the results of the time-frequency transformation of S, and the time-frequency feature matrix is ​​obtained, that is, the state space. The action space is the adjustable range of the frequency threshold and the amplitude threshold, and the optimal frequency threshold and amplitude threshold parameters are predicted; Reward Function : ； in, is the harmonic mean of precision and recall, is the weight coefficient; by The value is used as the optimization target, and the frequency and amplitude threshold parameters are iteratively optimized through reinforcement learning.

5. A method for selecting and triggering a gamma prompt radiation pulse event according to claim 4, characterized in that: The filtering process includes the following steps: The components below the frequency threshold in the result of S time-frequency transform are set to 0; The components below the amplitude threshold in the result of S time-frequency transform are set to 0; The filtered time-frequency domain is converted back to the time domain signal through the inverse S time-frequency transform.

6. A method for selecting and triggering a gamma prompt radiation pulse event according to claim 5, characterized in that: The morphological peak finding method includes the following steps: Perform dilation on the filtered signal to enhance potential peaks; Perform corrosion operation on the expanded signal to remove small fluctuations; Calculate morphological gradients and locate the points where the signal changes sharply; Valid pulse events are screened using the peak screening conditions of minimum peak height, minimum peak width, and minimum peak spacing.

7. The method for selecting and triggering a gamma prompt radiation pulse event according to claim 4, wherein: The deep Q network DQN includes a hybrid network consisting of a convolutional neural network and a long short-term memory network. The convolutional neural network is used to extract local time-frequency features, and the long short-term memory network is used to capture the dependencies of time series.

8. The method for selecting and triggering a gamma prompt radiation pulse event according to claim 4, wherein: The deep reinforcement learning model is implemented through a combination of offline training and online fine-tuning. Offline training uses a labeled gamma pulse dataset, and online fine-tuning dynamically adjusts model parameters based on real-time data.

9. The method for selecting and triggering a gamma prompt radiation pulse event according to claim 1, wherein: The gamma prompt radiation pulse event selection and triggering method adopts a sliding window mechanism for processing real-time data streams, and sets the window length and sliding step size.

Images