Search

CN-121997195-A - EDXRF spectrum intelligent analysis method, system, equipment and medium

CN121997195ACN 121997195 ACN121997195 ACN 121997195ACN-121997195-A

Abstract

The application relates to an EDXRF spectrum intelligent analysis method, system, equipment and medium. The method comprises the steps of obtaining a digital pulse waveform, carrying out multi-dimensional feature extraction on the digital pulse waveform to generate a standardized multi-dimensional feature vector, inputting the vector into a pre-trained gradient lifting decision tree model to obtain an event classification result of a current pulse event, using the event classification result as an event classification judgment result of a preset decision object, associating the digital pulse waveform to obtain a structured decision object, carrying out differentiation processing according to the event classification judgment result in the structured decision object to obtain an effective X-ray photon energy value, mapping the energy value to a corresponding energy spectrum channel according to an energy-channel mapping relation, and carrying out counting accumulation to generate a high-fidelity energy spectrum. The method improves the accuracy and the energy spectrum fidelity of pulse signal processing and enhances the accuracy and the stability of EDXRF spectrum analysis under a complex detection scene through multidimensional feature extraction and pre-training model reasoning.

Inventors

  • ZHANG WEN
  • Xiao Liu Chengsai

Assignees

  • 深圳市莱雷科技发展有限公司

Dates

Publication Date
20260508
Application Date
20260129

Claims (9)

  1. 1. An EDXRF spectrum intelligent analysis method, which is characterized by comprising the following steps: the method comprises the steps of obtaining a digital pulse waveform, carrying out multidimensional feature extraction on the digital pulse waveform, and generating a standardized multidimensional feature vector; Inputting the standardized multidimensional feature vector into a pre-trained gradient lifting decision tree model for reasoning to obtain an event classification result of a current pulse event, and associating the event classification result with the digital pulse waveform to obtain a structured decision object by taking the event classification result as an event classification decision result of a preset decision object; And acquiring a preset energy-channel mapping relation, mapping the effective X-ray photon energy value to a corresponding energy spectrum channel according to the energy-channel mapping relation, and counting and accumulating to generate a high-fidelity energy spectrum.
  2. 2. The method of claim 1, wherein said performing multi-dimensional feature extraction on said digital pulse waveform to generate a normalized multi-dimensional feature vector comprises: Extracting time domain features of the digital pulse waveform to obtain a time domain feature subset, wherein the time domain feature subset comprises pulse amplitude, rising edge time, falling edge time, pulse width, pulse area, overshoot parameters and ringing parameters; Performing discrete wavelet transformation processing on the digital pulse waveform, and calculating the energy of detail coefficients and the statistical characteristics of approximation coefficients under different decomposition scales to obtain a frequency domain characteristic subset; Carrying out morphological feature extraction on the digital pulse waveform to obtain a morphological feature subset, wherein the morphological feature subset comprises pulse symmetry, top flatness and normalized cross correlation coefficient with a standard single pulse template; Carrying out fusion processing on the time domain feature subset, the frequency domain feature subset and the morphological feature subset to obtain an initial feature vector; And carrying out standardization processing on the initial feature vector based on a preset feature mean value and standard deviation to generate the standardized multidimensional feature vector.
  3. 3. The method according to claim 1, wherein the inputting the standardized multidimensional feature vector into a pre-trained gradient lifting decision tree model to perform reasoning to obtain an event classification result of a current pulse event, and associating the event classification result with the digital pulse waveform to obtain a structured decision object by using the event classification result as an event classification decision result of a preset decision object, includes: Inputting the standardized multidimensional feature vector into the pre-trained gradient lifting decision tree model for reasoning to obtain probability vectors for representing that the current pulse event belongs to each preset category event; According to the probability vector, carrying out judgment processing according to a maximum probability principle to obtain an event classification result, wherein the event classification result comprises effective monopulses, resolvable pile-up pulses and invalid events; Creating the preset decision object, assigning the event classification result as an event classification decision result of the preset decision object to obtain an initial decision object, and when the event classification result is the resolvable pile-up pulse, performing subcomponent prediction processing based on the standardized multidimensional feature vector to obtain predicted subcomponent information, and storing the predicted subcomponent information into the initial decision object to obtain an optimized decision object; carrying out association binding on the digital pulse waveform and the optimization decision object; And when the event classification result is not the resolvable pile-up pulse, carrying out association binding on the digital pulse waveform and the initial decision object to obtain the structured decision object.
  4. 4. A method according to claim 3, wherein said performing a corresponding differentiation process based on said event classification decisions in said structured decision object results in an effective X-ray photon energy value, comprises: When the event classification judgment result is the effective monopulse, the digital pulse waveform is called from the structured decision object; Performing pulse amplitude or pulse area calculation processing on the digital pulse waveform to obtain a corresponding amplitude value or area value; According to a preset system energy scale curve, converting the amplitude value or the area value into a corresponding effective X-ray photon energy value; When the event classification judgment result is the resolvable pile-up pulse, the digital pulse waveform and the estimated subcomponent information are called from the structured decision object; Acquiring a preset standard single-pulse response function, and establishing a convolution superposition model according to the standard single-pulse response function; Taking the estimated energy and the estimated relative intensity in the estimated subcomponent information as the initial value of the sub-pulse parameter of the convolution superposition model, adjusting the parameter of the convolution superposition model through a nonlinear optimization algorithm, and obtaining an optimized convolution superposition model when the residual error between the synthesized waveform output by the convolution superposition model and the digital pulse waveform meets the preset convergence condition; Inputting the digital pulse waveform into the optimized convolution superposition model, performing waveform fitting processing to obtain a fitting result containing parameters of each sub-pulse, analyzing energy values of each sub-pulse from the fitting result, and taking the energy values of each sub-pulse as the corresponding effective X-ray photon energy values; And when the event classification judgment result is an invalid event, executing elimination processing on the structured decision object and the corresponding associated data.
  5. 5. The method of claim 1, wherein the obtaining a preset energy-channel mapping relationship, mapping the effective X-ray photon energy value to a corresponding energy spectrum channel according to the energy-channel mapping relationship, and performing count accumulation to generate a high-fidelity energy spectrum, comprises: Acquiring the preset energy-channel mapping relation, wherein the energy-channel mapping relation is the corresponding relation between the energy of the X-ray photon and an energy spectrum channel; Carrying out channel matching processing on each effective X-ray photon energy value according to the energy-channel mapping relation to obtain a target energy spectrum channel corresponding to each effective X-ray photon energy value; Performing accumulation update processing on the count of each target energy spectrum channel to obtain accumulated counts of each energy spectrum channel; and constructing an energy spectrum histogram according to the accumulated count of each energy spectrum channel, and generating the high-fidelity energy spectrum.
  6. 6. The method of claim 4, wherein the mathematical expression of the standard single impulse response function is: Wherein, the Is the time for the pulse to rise to 99% peak and , ; Is a time variable; Is the pulse rising edge time constant; is the pulse falling edge time constant; Is pulse flat top width; is the pulse peak amplitude; The standard deviation of the noise of the detector; mean value 0 and variance 0 For simulating the actual noise characteristics of the detector.
  7. 7. An EDXRF spectroscopy intelligent resolution system, the system comprising: The system comprises a multi-dimensional feature extraction module, a multi-dimensional feature extraction module and a multi-dimensional feature extraction module, wherein the multi-dimensional feature extraction module is used for acquiring a digital pulse waveform, and the digital pulse waveform is obtained by preprocessing an original digital pulse waveform; The classification reasoning and object construction module is used for inputting the standardized multidimensional feature vector into a pre-trained gradient lifting decision tree model for reasoning to obtain an event classification result of a current pulse event, and associating the event classification result with the digital pulse waveform to obtain a structured decision object as an event classification decision result of a preset decision object; the differentiation processing and energy spectrum generation module is used for executing corresponding differentiation processing according to the event classification judgment result in the structured decision object to obtain an effective X-ray photon energy value, acquiring a preset energy-channel mapping relation, mapping the effective X-ray photon energy value to a corresponding energy spectrum channel according to the energy-channel mapping relation, and counting and accumulating to generate a high-fidelity energy spectrum.
  8. 8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 6 when the computer program is executed.
  9. 9. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 6.

Description

EDXRF spectrum intelligent analysis method, system, equipment and medium Technical Field The invention belongs to the technical field of computers, and particularly relates to an EDXRF spectrum intelligent analysis method, an EDXRF spectrum intelligent analysis system, EDXRF spectrum intelligent analysis equipment and EDXRF spectrum intelligent analysis medium. Background Energy dispersive X-ray fluorescence (EDXRF) spectroscopy has been widely used in many fields such as geological exploration, alloy analysis, environmental monitoring, cultural relic identification, etc. because of its advantages of rapidness, no damage, portability, where hand-held EDXRF spectrometers are important application devices in industry by virtue of flexible field detection capabilities. The method comprises the following steps of irradiating a sample with primary X-rays emitted by a miniature X-ray tube, exciting sample atoms to generate characteristic X-ray fluorescence, receiving fluorescent photons by a detector, converting the fluorescent photons into original analog electric pulse signals, amplifying the original analog electric pulse signals by a preamplifier, obtaining original digital pulse waveforms through pretreatment operations such as pulse forming and baseline restoration, further preprocessing the original digital pulse waveforms to obtain digital pulse waveforms which can be used for subsequent analysis, and then processing the digital pulse waveforms by a digital pulse processor, mapping the digital pulse waveforms to corresponding channels according to energy, generating energy spectrum histograms, and providing a basis for element qualitative and quantitative analysis. However, the conventional EDXRF spectrum analysis method still has significant technical defects in practical application, namely, firstly, the processing flow of a digital pulse waveform is stiff, the conventional method mostly adopts a mode of combining linear filtering with fixed threshold screening, and only depends on single characteristics such as pulse peak value, simple width and the like to judge, system extraction and standardization processing are not carried out on time domain, frequency domain and morphological multidimensional characteristics of the digital pulse waveform, so that essential attributes of pulse signals are not fully described, secondly, the model for classifying pulse events lacks a definite adaptive training flow, the conventional model mostly adopts a general algorithm, no pulse sample characteristics based on EDXRF spectrum detection are subjected to targeted training and optimization, classification accuracy is insufficient, effective single pulses, resolvable accumulated pulses and invalid events are difficult to effectively distinguish, thirdly, the conventional method lacks a structured decision and association mechanism, a classification result and an original digital pulse waveform are not organically bound to form a structured decision object, the differential processing lacks a definite basis, and only adopts a simple discarding mode for complex situations such as pulse accumulation, and the like to cause effective counting loss, meanwhile, peak and interference peak value, low-free peak value, low-level, high-level stability and high-stability, and high-quality analysis performance, and no relative stability and no influence on the characteristics are generated when the method is used for detecting and analyzing the characteristics. Disclosure of Invention Based on the above, it is necessary to provide an EDXRF spectrum intelligent analysis method, system, device and medium for solving the above technical problems, so as to improve the comprehensiveness and accuracy of digital pulse waveform processing, and enhance the pertinence and reliability of pulse event classification, so as to meet the application requirements of complex detection scenes on site. In a first aspect, the application provides an EDXRF spectrum intelligent analysis method, comprising: the method comprises the steps of obtaining a digital pulse waveform, carrying out multidimensional feature extraction on the digital pulse waveform, and generating a standardized multidimensional feature vector; inputting the standardized multidimensional feature vector into a pre-trained gradient lifting decision tree model for reasoning to obtain an event classification result of a current pulse event, and taking the event classification result as an event classification judgment result of a preset decision object, and correlating digital pulse waveforms to obtain a structured decision object; and acquiring a preset energy-channel mapping relation, mapping the effective X-ray photon energy value to a corresponding energy spectrum channel according to the energy-channel mapping relation, and counting and accumulating to generate a high-fidelity energy spectrum. In one embodiment, the multi-dimensional feature extraction is performed on the digital pu