Peak-valley segmentation multi-size adaptive peak-searching method applied to collision light pulse radiation

Abstract

The application discloses a peak-valley segmentation multi-size adaptive peak searching method applied to collision light pulse radiation, relates to the field of light radiation measurement, and aims at the peak value structure size difference and asymmetric structure distribution of collision light radiation signal data and the like features, realizes adaptive extraction of different size peak characteristic quantities based on a peak-valley segmentation region and in combination with a logarithmic normal distribution fitting algorithm, and simultaneously realizes adaptive traversal of main existing regions of different scale peak value structures through screening of the peak and the valley, reduces the calculation amount of signal data processing and peak influence region screening, improves the peak searching process efficiency of different size peak value structures, and saves the calculation cost and time.

Description

Technical Field

[0001] This invention relates to the field of optical radiation measurement. More specifically, this invention relates to a peak-valley segmentation multi-size adaptive peak-finding method applied to collisional light pulse radiation. Background Technology

[0002] Radiometers measure pulse and spectral signals, requiring the extraction of radiation peak characteristics for further analysis. However, the large volume of data necessitates automated extraction, rather than manual identification. Generally, data processing involving peak signals relies on peak-finding methods, followed by multi-peak fitting and peak characteristic extraction and characterization based on the identified peaks. Peak-finding methods are widely used in elemental spectroscopy detection, gamma-ray spectroscopy analysis, and other fields.

[0003] Currently, there are various peak-finding algorithms, including amplitude methods, derivative methods, symmetric zero-area methods, wavelet transforms, and some novel peak detection methods such as the data trend accumulation peak detection algorithm without basis subtraction. These peak-finding methods are generally based on identifying features at the highest peak position of the signal, which is not suitable for region matching and segmentation of peak data, or for identifying peak signals with large size differences or asymmetry. In addition, many intelligent spectral data processing software programs have integrated peak detection algorithms suitable for peak detection of specific spectral signals, such as "centWave," "massifquant," and "matchedFilter." Users can find peaks by inputting parameters such as minimum peak area, peak height, or peak width; however, these methods require many parameters, have a high degree of subjectivity in application, require users to have certain professional background knowledge, and are not conducive to automated computer processing. Deep learning, due to its excellent feature learning capabilities, has shown performance close to or even surpassing that of the human eye in complex tasks such as image recognition and segmentation, and has been widely used in recent years for peak detection tasks in analytical chemistry. However, these methods require prior knowledge and training library support, making them unsuitable for analyzing signals with few samples.

[0004] The self-emission radiation signal from high-temperature materials generated by collisional sputtering exhibits multiple asymmetric pulsed radiation peaks. Since these peaks originate from different processes and mechanisms, their peak widths vary significantly. Fitting and extracting feature quantities from peak structures of different sizes requires traversing the expected range of peak structure width variations to achieve good feature extraction results. However, this approach requires prior knowledge of the peak structure variation range, and the traversal process incurs substantial computational costs. Therefore, a new peak-finding processing method is needed for feature extraction and analysis.

[0005] While some peak-finding methods have been disclosed in the prior art, such as the automatic peak-valley finding method, device, medium, and electronic device with application number 202110102714.3, this method mainly determines the initial peak position and the initial valley positions on both sides of the initial peak based on the second derivative of the curve to be integrated, preset peak position parameters, and preset search range parameters; it then corrects the initial peak position according to the initial valley position and preset peak correction rules to obtain the corrected peak position; it further corrects the initial valley position according to the corrected peak position, preset search range, and second derivative to obtain the corrected valley position; and within the corrected valley position interval, if a point on the curve to be integrated and the corrected peak are located on the same side of the dividing line, then the corrected peak position is determined to be the optimal peak position, and the corrected valley position is determined to be the optimal valley position. It is mainly used to find specific peak structures with predictable peak position parameters and search range. However, the light pulse signal generated by the collision radiation intensity signal exhibits the characteristics of multiple superimposed peak structures. The width of the pulse peak structure covers the range from tens of microseconds to hundreds of milliseconds, with large differences in size and structure and the peak position is not fixed. It is difficult to pre-set parameters such as peak position and search range for peak structures of different sizes of light pulses. At the same time, a matching peak finding strategy is required for the asymmetric pulse radiation peak characteristics. Therefore, this method is not suitable for the adaptive peak finding process of collision light pulse radiation. Summary of the Invention

[0006] One object of the present invention is to solve at least the above-mentioned problems and / or defects, and to provide at least the advantages described below.

[0007] To achieve these objectives and other advantages of the present invention, a multi-size adaptive peak-finding method for peak-valley segmentation applied to collisional light pulse radiation is provided, comprising:

[0008] S1. Perform wavelet denoising on the collision light pulse radiation signal obtained from experiments or numerical simulations to obtain filtered data;

[0009] S2. Based on the superposition characteristics of the collision light radiation peak structure, the local minimum value is obtained from the filtered data by multi-point fitting to calculate the first derivative, and the local minimum value is used as the initial selection of the segmentation point.

[0010] S3. Fit the peak features of the asymmetric peak structure using the log-normal distribution fitting algorithm, extract the feature peak data segments that meet the feature selection conditions, and make a merging judgment on the segmentation domains, and delete the segmentation points according to the judgment results.

[0011] S4. Determine if the number of segmentation points has been updated. If so, proceed to S3; otherwise, proceed to S5. Adaptively extract peak features of the size from the filtered data in a loop.

[0012] S5. Extraction ends, and the peak finding and extraction results are output.

[0013] Preferably, in S3, the log-normal distribution fitting algorithm is characterized by the following formula:

[0014]

[0015] In the above formula, a 1. a 2. a 3 is the log-normal distribution fitting coefficient. a 4. a 5 represents the corresponding linear fit coefficient. G It follows a log-normal distribution, and , x The sampling time of the light pulse radiation signal. y These are the values ​​at the sample points;

[0016] a 1. a 2. a 3. a 4. a The initial value for iteration 5 can be given according to the following formula:

[0017]

[0018] In the above formula, w To fit the interval width, and w = x r - x l , x l , x r These are the left and right dividing points, respectively. x mid , x max These represent the midpoint and maximum values ​​within the fitted interval, respectively. y max The maximum sampled value within the fitting interval. y l , y r These are the sampled values ​​at the left and right dividing points of the fitted interval, respectively;

[0019] Then, the peak height H, peak half-width L, and peak position among the peak structure characteristics are... P x Characterized by the following fitting parameter expression:

[0020] .

[0021] Preferably, the processing flow of the fitting and screening algorithm includes:

[0022] S30. Set peak filtering conditions;

[0023] S31. Determine whether the peak feature quantity obtained in S3 meets the filtering conditions. If it does, extract the data within the corresponding segmentation area; otherwise, delete the segmentation point.

[0024] S32. Merge adjacent regions, and based on the peak-valley segmentation principle, filter and delete segmentation points and update segmentation domains in the current segmentation point set, so as to find peak structures that meet the requirements again when extracting peak structure features in the new segmentation region.

[0025] Preferably, in S31-S32, the method for selecting peak-valley dividing points is as follows:

[0026] Set a logical judgment flag with an initial value of 1 for all split points in the current split point set;

[0027] When judging peak characteristics, the amplitude of the two endpoints of the region that does not meet the screening criteria is compared, and the endpoint with the higher amplitude is set to a logical flag of 0.

[0028] Once the peak features of all segmented regions have been determined, the segmentation points with a logical identifier of 0 are deleted, completing the screening of peak and valley segmentation points and obtaining new segmented regions for the next peak determination.

[0029] The present invention has at least the following beneficial effects:

[0030] Firstly, this invention establishes a log-normal distribution fitting algorithm for the asymmetric peak structure of collision light radiation, thereby realizing the extraction of peak characteristic quantities of the asymmetric peak structure;

[0031] Secondly, this invention provides an adaptive peak-finding method based on peaks and valleys, which can greatly reduce the amount of data and peak influence region analysis, improve the efficiency of peak-finding process for peak structures of different sizes, and save computational costs and time.

[0032] Third, this invention can adaptively find peak structures of asymmetry and different scales. Furthermore, by changing the type of fitting function, the peak-valley segmentation method can also be used to find symmetrical peak features.

[0033] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of wavelet noise reduction multilayer signal decomposition according to the present invention;

[0035] Figure 2 This is a flowchart illustrating the quantization extraction of radiation peak features using the peak-valley segmentation multi-size adaptive peak-finding method of this invention.

[0036] Figure 3 The image shows the collision light radiation intensity signal curve (black) and the adaptive fitting extraction result for experimental condition 1 in the verification example of this invention.

[0037] Figure 4 Experimental condition 1 for the verification example of this invention, the radiation peak structure diagram extracted from the collision light radiation intensity signal curve;

[0038] Figure 5 This is a graph showing the collision light radiation intensity signal curve and adaptive fitting extraction result corresponding to experimental condition 2 in the verification example of this invention.

[0039] Figure 6 Experimental condition 2, a verification example of the present invention, shows the radiation peak structure diagram extracted from the collision light radiation intensity signal curve.

[0040] Figure 7 This is a graph showing the collision light radiation intensity signal curve and adaptive fitting extraction result corresponding to experimental condition 3 in the verification example of this invention.

[0041] Figure 8 Experimental condition 3, a verification example of the present invention, shows the radiation peak structure diagram extracted from the collision light radiation intensity signal curve.

[0042] Figure 9 This is a graph showing the collision light radiation intensity signal curve and adaptive fitting extraction result corresponding to experimental condition 4 in the verification example of this invention.

[0043] Figure 10 Experimental condition 4, a verification example of the present invention, shows the radiation peak structure diagram extracted from the collision light radiation intensity signal curve.

[0044] Figure 11 This is a graph showing the collision light radiation intensity signal curve and adaptive fitting extraction result corresponding to experimental condition 5 in the verification example of this invention.

[0045] Figure 12 Experimental condition 5, a verification example of the present invention, shows the radiation peak structure diagram extracted from the collision light radiation intensity signal curve.

[0046] Figure 13 This is a graph showing the collision light radiation intensity signal curve and adaptive fitting extraction result corresponding to experimental condition 6 in the verification example of this invention.

[0047] Figure 14Experimental condition 6, a verification example of the present invention, shows the radiation peak structure diagram extracted from the collision light radiation intensity signal curve.

[0048] Figure 15 This is a graph showing the collision light radiation intensity signal curve and adaptive fitting extraction result corresponding to experimental condition 7 in the verification example of this invention.

[0049] Figure 16 Experimental condition 7, a verification example of the present invention, shows the radiation peak structure diagram extracted from the collision light radiation intensity signal curve.

[0050] Figure 17 This is a graph showing the collision light radiation intensity signal curve and adaptive fitting extraction result corresponding to experimental condition 8 in the verification example of this invention.

[0051] Figure 18 Experimental condition 8, used as a verification example of the present invention, shows the radiation peak structure diagram extracted from the collision light radiation intensity signal curve. Detailed Implementation

[0052] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0053] This invention leverages the peak-valley formation mechanism in signal data, namely, the contribution of the peak-valley positions between peaks to the fitting results of adjacent peak structures is locally minimal. By using peak-valleys as the dividing points between adjacent peaks, a peak-valley segmentation region peak-finding method is proposed. This method adaptively establishes segmentation fitting intervals based on the peak-valley characteristics of peak structures of different sizes in signal data. Through fitting and screening methods, segmentation intervals are extracted and merged to achieve adaptive extraction of feature quantities of peaks of different sizes.

[0054] Example:

[0055] This embodiment is mainly used for extracting peak feature quantities of collision light radiation intensity time evolution signals, and can also be used for other spectral peak detection processes.

[0056] The main function of this embodiment is to achieve adaptive fitting and extraction of features such as peak height, peak width, and peak position for peak structures of different sizes in any signal curve. While ensuring the accuracy of physical properties, it saves computation time and cost. Specifically, for example... Figure 2 As shown, the operation steps include:

[0057] Step 1: Wavelet noise reduction;

[0058] Wavelet transform is used to transform and filter noise from signals obtained from experiments or numerical simulations. Wavelet analysis is based on wavelet transform.

[0059] (1)

[0060] In the formula: For the mother wavelet, a The scaling factor. b This is the translation factor. By changing... a and b It can perform scaling and translation transformations of the mother wavelet to realize time-domain signals. f Time-frequency localization analysis. Take... and This is transformed into a discrete wavelet transform form. For equally spaced time-domain acquired signals, a binary wavelet transform form is used, i.e. and For the mother wavelet function There are various wavelet waveforms, including Haar, Marr, Morlet, Daubechies (DbN), and Mallat. This invention uses the orthogonal fast algorithm Mallat to implement low-frequency filtering of time-domain signals. The specific decomposition and filtering process can be found in [link to relevant documentation]. Figure 1 Here, S is the signal to be decomposed. An orthogonal decomposition algorithm is applied, decomposing each layer into a low-frequency component A and a high-frequency component D. In the first layer, v is set to 1. The next layer continues to decompose the low-frequency component A of the signal, and this process continues as the number of decomposition layers increases. m The value also increases. For binary wavelet decomposition, the relationship between the frequency of the decomposed signal and the number of decomposition levels is: The relationship, and the number of signal points will also be shown. Decreasing relationship. For example... Figure 1 As shown, the relationship between the final decomposed signal S and the signals of each decomposed layer is as follows: .

[0061] For the time evolution signal of ultra-high-speed collision light radiation, noise is mostly high-frequency. Wavelet transform analysis is used to eliminate high-frequency noise while retaining useful low-frequency information. When the signal has strong fluctuations and the number of sampling points is too sparse, too many decomposition layers will lead to the filtering out of deterministic fluctuation information in the signal, causing signal distortion. Therefore, under normal circumstances, a comprehensive approach is to decompose the signal into 3 or 4 layers to obtain the filtered data A. m .

[0062] Step 2: Peak-valley segmentation of the filtered signal;

[0063] The derivative method is used to detect local peaks and valleys in the filtered data through slope changes, serving as segmentation points. For multi-peak superposition distributions in the signal data, this algorithm assumes that the peaks and valleys formed between peaks are local minima. A first-order derivative fitting method is used to find the inflection points where the derivative changes from negative to positive, and these inflection points are taken as local minima. To reduce derivative oscillations caused by noise in the acquired signal data, a multi-point slope fitting method is used to calculate the first derivative during processing. kThe multi-point fitting formula is as follows:

[0064] (2)

[0065] In the formula, and These represent the fitted point data and the average data, respectively, with q being the number of fitted points, typically chosen as 4. The filtered data is segmented by finding local minima as initial peak and valley locations.

[0066] Step 3: Fitting the peak structure with initial values ​​provided;

[0067] After determining the segmentation points, the extracted peak structures are further fitted and extracted. Since the collisional radiation peak structures in the signal often exhibit asymmetric distribution characteristics, this method uses a log-normal distribution for fitting, with the distribution function as follows:

[0068] (3)

[0069] in, G It follows a log-normal distribution:

[0070] (4)

[0071] In the formula, , , These are the fitting coefficients for the log-normal distribution. Since the peak radiation structure is mostly superimposed on the background signal or other radiation peaks, a single function cannot be used for fitting. A linear distribution function needs to be added to the log-normal distribution function to linearly approximate the superimposed radiation signal intensity. , These are the corresponding linear fitting coefficients.

[0072] The function fitting process is a least squares problem, and this algorithm uses the Levenberg-Marquardt (LM) algorithm for iterative calculation. The LM algorithm is similar to the Gauss-Newton iterative algorithm, but it introduces adjustment parameters to avoid ill-conditioned conditions caused by the singularity of the Jacobian matrix in the Gauss-Newton algorithm, and it can maintain quadratic convergence. However, the LM algorithm is a local algorithm and depends on the selection of initial values. The initial values ​​for iteration can be given according to the physical meaning of the parameters:

[0073] (5)

[0074] In the formula, w To fit the interval width, and w = x r - x l ,x l , x r These are the left and right dividing points, respectively. x mid , x max These represent the midpoint and maximum values ​​within the fitted interval, respectively. y max The maximum sampled value within the fitting interval. y l , y r These are the sampled values ​​at the left and right dividing points of the fitted interval, respectively.

[0075] Step 4: Extraction of peak structure features;

[0076] The peak structure characteristics of a time-domain signal are contained within the fitted logarithmic distribution function. Through fitting, the corresponding fitting parameters can be obtained, and quantized features such as radiation peak position, peak height, and peak width can be extracted from them. Based on the log-normal distribution parameter relationship, peak height H, peak half-width L, and peak position... P x This can be expressed as an expression for the fitting parameters:

[0077] (6)

[0078] Step 5: Determine whether to merge the filtered data segmentation domains;

[0079] The region formed by the aforementioned peak-valley segmentation is fitted, and filtering conditions are set to judge the extracted peak features. In practical applications, the peak structure filtering conditions can be set more leniently by combining the segmented region. For example, in a signal segment, only peak structures that meet the preset peak conditions can be extracted by setting filtering conditions.

[0080] If the peak features extracted from the corresponding region meet the screening criteria, the data within the corresponding segmented region is extracted; otherwise, the segmentation point is deleted and adjacent regions are merged. Here, based on the peak-valley segmentation principle, further screening and deletion of segmentation points and updating of segmentation domains are performed in the current set of segmentation points to find possible peak structures.

[0081] The specific method for filtering and deleting segmentation points is as follows: Add a logical judgment flag to all segmentation points, with an initial value of 1; compare the amplitudes of the two endpoints of regions that do not meet the peak extraction parameters, and then set the logical flag of the endpoint with the higher amplitude to 0. After performing peak feature judgment on all segmented regions, obtain the logical judgment value corresponding to each segmentation point, and then delete the segmentation points with a logical flag of 0, thereby achieving the deletion and filtering of peak and valley segmentation points and obtaining the peak judgment region for the next division.

[0082] Step 6: Extract peak features from the filtered data in a loop;

[0083] Following step five, after the segmentation points have been filtered and deleted, the peak feature extraction and segmentation point filtering and deletion processes are repeated for the divided regions. This process is repeated in a loop. See the flowchart for details. Figure 2 The cyclic filtering process continues until the number of segmentation points no longer changes, and then the peak structure found in the loop is output.

[0084] After completing steps one through six above, adaptive peak finding and feature extraction are achieved based on signal data.

[0085] This example addresses the characteristics of peak structure size differences and asymmetric structure distribution in collision light radiation signal data. Based on a peak-valley segmentation region-finding method combined with an asymmetric peak fitting algorithm, it achieves adaptive extraction of peak feature quantities for peaks of different sizes. This invention adaptively traverses the main existence regions of peak structures at different scales by deleting peak-valley segmentation points and merging neighboring regions, reducing the computational load of signal data processing and peak influence region screening, improving the efficiency of the peak-finding process for peak structures of different sizes, and saving computational costs and time.

[0086] Verification example:

[0087] like Figures 3-18 As shown, this invention provides collision light radiation intensity detection signal data for eight experimental conditions. The method described in this scheme is used to extract the pulse peak structure from the light radiation signal, and the adaptive extraction result is plotted on the original data curve. The specific extraction process for the signal curve is as follows: Figure 3 , Figure 5 , Figure 7 , Figure 9 , Figure 11 , Figure 13 , Figure 15 , Figure 17 As shown, through Figure 4 , Figure 6 , Figure 8 , Figure 10 , Figure 12 , Figure 14 , Figure 16 , Figure 18 It can be seen that the peak finding method of the present invention can be used to extract the peak structure features of complex signal curves.

[0088] The above solution is merely an illustration of a preferred example and is not limited thereto. When implementing this invention, appropriate substitutions and / or modifications can be made according to the user's needs.

[0089] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. It can be applied to various fields suitable for the present invention. Other modifications can be readily made by those skilled in the art. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and examples shown and described herein.

Claims

1. A peak-valley segmentation multi-size adaptive peak-searching method applied to a collision light pulse radiation, characterized in that, The method comprises the following steps: S1, wavelet denoising is performed on the collision light pulse radiation signal obtained through experiment or numerical simulation to obtain filtered data; S2, based on the superposition characteristics of the peak structure of the collision light radiation, a first derivative is calculated by using a multi-point fitting method to obtain a local minimum value from the filtered data, and the local minimum value is taken as a preliminary segmentation point; S3, a lognormal distribution fitting algorithm is used to fit the peak characteristic quantity of the asymmetric peak structure, a characteristic peak data segment that meets the characteristic quantity screening condition is extracted, and a merging judgment is performed on the segmentation domain, and the segmentation points are deleted according to the judgment result; S4, it is judged whether the number of segmentation points is updated, if yes, S3 is entered, otherwise S5 is entered; the peak characteristic quantity is adaptively extracted from the filtered data in a loop manner; S5, the extraction is completed, and the peak searching and extraction results are outputted; In S3, the lognormal distribution fitting algorithm is characterized by the following formula: In the above formula, a 1、 a 2、 a 3 is a logarithmic normal distribution fitting coefficient, a 4、 a 5 is a corresponding linear fitting coefficient, G is a logarithmic normal distribution, and , x is a sampling point time of the optical pulse radiation signal, y is a sampling point value; a 1、 a 2、 a 3、 a 4、 a The iteration initial value for 5 can be given according to the following formula: In the above formula, w is the fitting interval width, and w = x r - x l , x l , x r are left and right split points, respectively, x mid , x max are the midpoint value and the position of the maximum value in the fitting interval, respectively, y max is the maximum sampling value in the fitting interval, y l , y r are the sampling values at the left and right split points of the fitting interval, respectively; Then, the peak height H, the peak half-height width L, and the peak position P in the peak structure feature quantity P x characterized by the following fitting parameter expression: 。 2. The peak valley segmentation multi-size adaptive peak finding method applied to the colliding light pulse radiation according to claim 1, wherein, The processing flow of the fitting screening algorithm comprises the following steps: S30, setting a peak screening condition; S31, judging whether the peak characteristic quantity obtained in S3 meets the screening condition, if yes, the data in the corresponding segmentation region is intercepted, otherwise the segmentation point is deleted; S32, merging adjacent regions, and based on the peak-valley segmentation principle, the segmentation points in the current segmentation point set are screened and deleted, and the segmentation region is updated, so that the peak value structure characteristic is cyclically extracted in the new segmentation region, and the peak value structure meeting the requirement is found again.

3. The peak valley segmentation multi-size adaptive peak finding method applied to the colliding light pulse radiation according to claim 2, characterized in that, In S31-S32, the screening method of the peak-valley segmentation point is as follows: An initial value of 1 is set for all segmentation points in the current segmentation point set as a logical judgment identifier; When the peak characteristic judgment is performed, the amplitude of the two end points in the region that does not meet the screening condition is compared, and the end point with higher amplitude is set as a logical identifier of 0; After the peak characteristic judgment of all segmentation regions is completed, the segmentation points with the logical identifier of 0 are deleted, the screening of the peak-valley segmentation point is completed, and the new segmentation region for the next division peak judgment is obtained.

Images