Pulse waveform classification method for multi-layer optimization mapping

Abstract

The invention discloses a pulse waveform classification method based on multilayer optimization mapping. Firstly, a normalized orthogonal function set is established according to waveform features; then solving feature vectors of all samples in the pulse waveform training set after mapping in the function space; secondly, performing twice projection (namely nonlinear projection and optimal direction linear projection) on the feature vector to obtain a critical judgment point; and finally, identifying the unknown waveform based on the critical judgment point to obtain a final attribution class. According to the method, after complex and diversified pulse waveforms are classified, a subsequent digital waveform forming method and a waveform parameter identification method can be conveniently developed based on categories, and the high sensitivity of the waveform forming and parameter identification method to pulse order and waveform complex diversity is overcome; and finally, guarantee is provided for accurate measurement of the energy spectrum under complex conditions.

Description

Technical Field

[0001] The invention relates to a pulse waveform classification method of multi-layer optimization mapping. Background Art

[0002] The front-end analog system of a nuclear radiation detector is composed of a detector, a preamplifier, a CR / RC network, an amplifier circuit, and a conditioning circuit. The front-end analog system is composed of a complex and diverse structure. Therefore, the waveform of the pulse signal output by the system may be diverse and complex, which is manifested as different orders in the s domain, that is, the number of components contained in each pulse and the component composition fractions are also different. For example, in a complex radioactive environment (for example, when multiple rays coexist), multiple types of detectors are required for coordinated measurement, and the measured pulse signal waveform may be complex and diverse. Due to the complex and diverse pulses output by the front-end analog system, the mathematical model has different orders (some may be infinitely high), that is, the number of components contained in a single pulse and the component composition fractions are also different; therefore, when the pulse signal is subsequently digitized and waveform-shaped (for example, standard Gaussian, quasi-Gaussian SK, trapezoidal, bullet-shaped and other emerging shaping methods, etc.) to obtain the pulse amplitude, the corresponding algorithm must be designed based on the order of the pulse signal model and the specific parameters of each component, otherwise it will lead to reduced accuracy or even failure of energy spectrum measurement. For example, in the case of misjudgment of the pulse signal order, the shaping method based on the mathematical model will cause the top of the trapezoidal shaping to be distorted due to coefficient mismatch, which will ultimately lead to reduced accuracy of the acquired pulse amplitude and energy spectrum. In radioactivity measurement, for the case where pulse waveforms have multiple orders and multiple waveform shapes at the same time (for example, using multiple detectors to measure radiation scenes with multiple rays), pulses must be classified (for example, based on order or waveform characteristics), and then various corresponding shaping and identification methods are used to ensure the accuracy of the acquired pulse waveform amplitude and energy spectrum. The present invention adopts a multi-layer optimized mapping pulse waveform classification method to classify pulses in the case of multiple types of pulse waveforms, so as to facilitate the subsequent qualitative and quantitative analysis of pulses and adopt corresponding matching algorithms, and ultimately provide guarantee for accurate radioactivity measurement. Summary of the invention

[0003] The purpose of the present invention is to disclose a multi-layer optimized mapping pulse waveform classification method to classify the complex and diverse pulse waveforms output by the front-end analog system of a nuclear radiation detector, so as to facilitate the subsequent digital waveform shaping and waveform parameter identification methods to be carried out based on the categories, thereby overcoming the high sensitivity of waveform shaping and parameter identification methods to the complexity and diversity of pulse orders and waveforms, and ultimately providing a guarantee for the accuracy of radioactivity measurements (for example, energy spectrum measurements) under complex conditions.

[0004] The present invention relates to a multi-layer optimized mapping pulse waveform classification method, which is realized through the following steps ① to ④.

[0005] Step ① Design M basis functions and obtain the normalized orthogonal function set. The function space spanned by the orthogonal function set is denoted as H M .

[0006] Step ② Find all samples x in the pulse waveform training set i (t)(a≤t≤b) in the function space H M The eigenvectors in X i ; The set of feature vectors of the lth and jth samples is denoted as A lj ; Initialize l=2 and j=1; and define an L-dimensional array NumClass and initialize it to 0.

[0007] Step ③ uses the ω in the training set l With ω j Samples in the function space H M The feature vector in , and based on the nonlinear projection method, the unknown sample x′(t)(a≤t≤b) is roughly classified and identified (that is, roughly identified as ω l Class or ω j Class), follow the steps below 3S 1 ~3S 4 accomplish: Step 3S1 Find ω l The feature vector of the class sample is in the set A lj The average value of the nonlinear mapping in Step 3S 2 Ask for ω j The feature vector of the class sample is in the set A lj The average value of the nonlinear mapping in Step 3S 3 Ask for ω l Class and ω j Two types of critical decision points w 0 ; Step 3S 4 The unknown sample x′(t)(a≤t≤b) to be classified and identified is roughly classified (that is, roughly identified as ω l Class or ω j Class), follow the steps (1) to (3): (1) Find x′(t)(a≤t≤b) in the function space H M The eigenvector X′ in (2) The eigenvector X′ in the set A lj After nonlinear mapping on Project again on to get the projection value y: (3) Based on the projection value y of the feature vector X′, the unknown sample x′(t) is classified into two categories, that is, x′(t) is judged as ω l Class or ω j Class, as follows: If y>w 0 , then x′(t) belongs to ω l class, and correct NumClass(l)=NumClass(l)+1; otherwise, x′(t) belongs to ω j class, and correct NumClass(j)=NumClass(j)+1. If j<l-1, then modify j=j+1 and return to step 3S 1 Continue execution; otherwise, go to step ④.

[0008] Step ④ If l<L, reset l=l+1 and j=1, and return to step ③ to continue the loop execution; otherwise, the sequence number corresponding to the maximum element in the array NumClass is used as the final class of x′(t). For example, if NumClass(k) is the maximum value, the pulse waveform x′(t) is classified as ω k kind.

[0009] In summary, through steps ① to ④, a multi-layer optimized mapping pulse waveform classification method is adopted to achieve the classification of complex and diverse pulse waveforms.

[0010] The beneficial effects of the present invention are: In radioactivity measurements, for situations where pulse waveforms have multiple orders and multiple waveform shapes at the same time (for example, using multiple detectors to measure radiation scenarios with multiple rays), the pulses must be classified (for example, based on order or waveform characteristics) and then the corresponding shaping and identification methods must be used to ensure the accuracy of the acquired pulse waveform amplitude and energy spectrum information. The present invention adopts a multi-layer optimized mapping pulse waveform classification method to classify pulses in the case of multiple types of pulse waveforms. The beneficial effects of this method are: (1) according to the characteristics of various pulse waveforms in the pulse data set, a matching orthogonal normalization function set is designed, and waveform mapping is performed based on the function set to obtain a feature vector, thereby effectively extracting the pulse waveform feature information; (2) the high-dimensional time domain waveform features are converted into finite-dimensional feature vectors, and at the same time, the "inconspicuous differences" between pulse waveforms in the time domain are converted into "obvious differences" in feature vectors, so as to facilitate the accurate separability of pulse waveforms; (3) the feature vectors of the pulse waveforms are nonlinearly projected to achieve accurate separability of pulse waveforms; The further separation between the characteristic vectors of various types of pulse waveforms is achieved, laying a solid foundation for the accurate identification of pulse waveform types; (4) the characteristic vectors after nonlinear mapping are re-projected in the optimal direction to ensure the accurate identification of pulse waveform types; (5) after accurate classification of complex and diverse pulse waveforms, it is convenient for subsequent digital waveform shaping and waveform parameter identification methods to be carried out based on categories, and then the corresponding matching algorithms are adopted for the qualitative and quantitative analysis of pulses, thus overcoming the high sensitivity of waveform shaping and parameter identification methods to the complexity and diversity of pulse orders and waveforms, and finally providing a guarantee for the accuracy of radioactivity measurements (for example, energy spectrum measurements) under complex conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0012] The following is a detailed description of an embodiment of the present invention in conjunction with the accompanying drawings. This embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation method and process are given, but the protection scope of the present invention is not limited to the following embodiment.

[0013] The present invention relates to a multi-layer optimized mapping pulse waveform classification method, which is realized through the following steps ① to ④.

[0014] Step ① Design M basis functions and obtain a normalized orthogonal function set, which can be implemented as follows:

[0015] (1) Taking the pulse waveform classification and recognition whose s-domain expression is as shown in formula (1) and whose poles are real numbers as an example, the designed basis function is f 1 (t),f 2 (t),f 3 (t),...,f M (t); the number of basis functions is M, and the expression of each basis function is shown in formula (2): The parameter τ m and M are set according to the pulse waveform characteristics.

[0016] (2) Obtain a set of normalized orthogonal functions according to the following formulas (3) to (6): Right now: In formula (3), φ k (t) is calculated according to the following formula (5): <f m (t),φ k (t)> is calculated according to the following formula (6); By φ 1 (t),φ 2 (t),φ 3 (t),...,φ M (t) constitutes a set of normalized orthogonal functions, and the function space spanned by them is denoted as H M .

[0017] Step ② Find all samples x in the pulse waveform training set i (t)(a≤t≤b) in the function space H M The eigenvectors in X i , implemented as follows: where x i (t)(a≤t≤b) is the i-th pulse waveform sample in the training set, and the total number of samples is N; the feature vector X i The elements of are shown in formula (8); Where X i =[X i (1),X i (2),...,X i (M)],X i (m) is X i The mth element of Assume that there are L types of samples in the pulse waveform training set, and the category of the lth (l=1,...,L) type of sample is ω l Indicates that ω l The number of class samples is N l ,but: ω l The characteristic vector of the pth sample in the class sample is recorded as: The set of feature vectors of samples of the lth and jth categories is denoted as A lj ; In order to execute the subsequent steps, l=2 and j=1 are initialized; and an L-dimensional array is defined as NumClass=[NumClass(1), NumClass(2), ..., NumClass(L)] and initialized to 0.

[0018] Step ③ uses the ω in the training set l With ω j Samples in the function space H M The feature vector in , and based on the nonlinear projection method, the unknown sample x′(t)(a≤t≤b) is roughly classified and identified (that is, roughly identified as ω l Class or ω j Class), follow the steps below 3S 1 ~3S 4 accomplish:

[0019] Step 3S1 Find ω l The feature vector of the class sample is in the set A lj The average value M of the nonlinear mapping in l : Where N′=N l +N j , is an N′×1 matrix, Represents ω l All feature vectors of class samples in set A lj The average value after nonlinear mapping on the qth vector; the nonlinear mapping here As shown in formula (12).

[0020] Step 3S 2 Ask for ω j The feature vector of the class sample is in the set A lj The average value of the nonlinear mapping in Where N′=N l +N j , is an N′×1 matrix, Represents ω j All feature vectors of class samples in set A lj The average value after nonlinear mapping on the qth vector; the nonlinear mapping here As shown in formula (14).

[0021] Step 3S 3 Ask for ω l Class and ω j Two types of critical decision points w 0 , according to the following formula (15); In formula (15) Calculate according to the following formulas (16) to (17): in, ω l The average value vector of class samples after nonlinear mapping In the best direction projection on ω j The average value vector of class samples after nonlinear mapping In the best direction Projection on. The optimal direction in formula (17) It is given by the following formula (18): In formula (18), I is the unit matrix, and υ is the matrix that makes B υ is the additional quantity of the positive definite matrix, T represents the matrix transpose, D l N l ×N l The matrix, whose elements are D j N j ×N j The matrix, whose elements are C l and C j As shown in formulas (19) and (20);

[0022] Step 3S 4 The unknown sample x′(t)(a≤t≤b) to be classified and identified is roughly classified (that is, roughly identified as ω l Class or ω j Class), follow the steps (1) to (3):

[0023] (1) Find x′(t)(a≤t≤b) in the function space H M The eigenvector X′ in is as shown in formula (21), Where X′=[X′(1),X′(2),...,X′(M)], and X′(m) is the mth element of X′.

[0024] (2) The eigenvector X′ in the set A lj After nonlinear mapping on Project again on to get the projection value y: Where X q Yes A lj The qth vector of the set.

[0025] (3) Based on the projection value y of the feature vector X′, the unknown sample x′(t) is classified into two categories, that is, x′(t) is judged as ω l Class or ω j Class, as follows: If y>w 0 , then x′(t) belongs to ω l class, and correct NumClass(l)=NumClass(l)+1; otherwise, x′(t) belongs to ω j class, and correct NumClass(j)=NumClass(j)+1. If j<l-1, then modify j=j+1 and return to step 3S 1 Continue execution; otherwise, go to step ④.

[0026] Step ④ If l<L, reset l=l+1 and j=1, and return to step ③ to continue the loop execution; otherwise, the sequence number corresponding to the largest element in the array NumClass=[NumClass(1),NumClass(2),...,NumClass(L)] is used as the final class of x′(t). For example, if NumClass(k) is the maximum value, the pulse waveform x′(t) is classified as ω k kind.

[0027] In summary, through steps ① to ④, a dynamically changing basis function is used to form a function space, the pulse waveform x output by the detection system is projected into the function space, and the similar function (or similar waveform) of the pulse waveform x in the function space is searched based on the projection vector. Reconstruct the pulse waveform x in series form within the error tolerance range; finally, based on the similarity function (or similar waveform) The shaping algorithm of the pulse waveform x is obtained by using the series components of the pulse waveform.

[0028] As described above, based on a multi-layer optimized mapping pulse waveform classification method, the complex and diverse pulse waveforms output by the front-end analog system of the nuclear radiation detector are classified, so that subsequent digital waveform shaping and waveform parameter identification methods can be carried out based on categories, overcoming the high sensitivity of waveform shaping and parameter identification methods to the complexity and diversity of pulse orders and waveforms, and ultimately providing a guarantee for the accuracy of radioactivity measurement (for example, energy spectrum measurement) under complex conditions. Ultimately, it provides a guarantee for the accurate measurement of energy spectrum.

[0029] In the above-mentioned embodiment of the present invention, a multi-layer optimized mapping pulse waveform classification method is described in detail, but it should be noted that the above is only one embodiment of the present invention. When other forms of classification and recognition involve the use of the method mentioned in this article, the present invention is still valid. Any modifications, equivalent substitutions, improvements, etc. 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 pulse waveform classification method with multi-layer optimization mapping, characterized in that: The classification of pulse signal waveform is achieved through the following steps ①~④: Step ① Design M basis functions and obtain a normalized orthogonal function set, which can be implemented as follows: (1) Taking the pulse waveform classification and recognition whose s-domain expression is as shown in formula (1) and whose poles are real numbers as an example, the designed basis functions are f1(t), f2(t), f3(t), ..., f M (t); the number of basis functions is M, and the expression of each basis function is shown in formula (2): The parameter τ m and M are set according to the pulse waveform characteristics; (2) Obtain a set of normalized orthogonal functions according to the following formulas (3) to (6): Right now: In formula (3), φ k (t) is calculated according to the following formula (5): <f m (t),φ k (t)> is calculated according to the following formula (6); From φ1(t),φ2(t),φ3(t),...,φ M (t) constitutes a set of normalized orthogonal functions, and the function space spanned by them is denoted as H M ; Step ② Find all samples x in the pulse waveform training set i (t)(a≤t≤b) in the function space H M The eigenvectors in X i , implemented as follows: where x i (t)(a≤t≤b) is the i-th pulse waveform sample in the training set, and the total number of samples is N; the feature vector X i The elements of are shown in formula (8); Where X i =[X i (1),X i (2),...,X i (M)],X i (m) is X i The mth element of Assume that there are L types of samples in the pulse waveform training set, and the category of the lth (l=1,...,L) type of sample is ω l Indicates that ω l The number of class samples is N l ,but: ω l The characteristic vector of the pth sample in the class sample is recorded as: The set of feature vectors of samples of the lth and jth categories is denoted as A lj ; In order to execute the subsequent steps, initialize l=2 and j=1; define the L-dimensional array as NumClass=[NumClass(1), NumClass(2), ..., NumClass(L)] and initialize it to 0; Step ③ uses the ω in the training set l With ω j Samples in the function space H M The feature vector in , and based on the nonlinear projection method, the unknown sample x′(t)(a≤t≤b) is roughly classified and identified (that is, roughly identified as ω l Class or ω j Class), implement it according to the following steps 3S1~3S4: Step 3S1 Find ω l The feature vector of the class sample is in the set A lj The average value of the nonlinear mapping in Where N′=N l +N j , is an N′×1 matrix, Represents ω l All feature vectors of class samples in set A lj The average value after nonlinear mapping on the qth vector; the nonlinear mapping here As shown in formula (12); Step 3S2 find ω j The feature vector of the class sample is in the set A lj The average value of the nonlinear mapping in Where N′=N l +N j , is an N′×1 matrix, Represents ω j All feature vectors of class samples in set A lj The average value after nonlinear mapping on the qth vector; the nonlinear mapping here As shown in formula (14); Step 3S3 Find ω l Class and ω j The two types of critical decision points w0 of the class are as follows: In formula (15) Calculate according to the following formulas (16) to (17): in, ω l The average vector of class samples after nonlinear mapping In the best direction projection on; ω j The average vector of class samples after nonlinear mapping In the best direction projection on; The optimal direction in formula (17) It is given by the following formula (18): In formula (18), I is the unit matrix, and υ is the matrix that makes B υ is the additional quantity of the positive definite matrix, T represents the matrix transpose, D l N l ×N l The matrix, whose elements are D j N j ×N j The matrix, whose elements are C l and C j As shown in formulas (19) and (20); Step 3S4 roughly classifies the unknown sample x′(t)(a≤t≤b) to be classified (i.e. roughly identifies it as ω l Class or ω j Class), follow the steps (1) to (3): (1) Find x′(t)(a≤t≤b) in the function space H M The eigenvector X′ in is as shown in formula (21), Where X′=[X′(1),X′(2),...,X′(M)], X′(m) is the mth element of X′; (2) The eigenvector X′ in the set A lj After nonlinear mapping on Project again on to get the projection value y: Where X q Yes A lj The qth vector of the set; (3) Based on the projection value y of the feature vector X′, the unknown sample x′(t) is classified into two categories, that is, x′(t) is judged as ω l Class or ω j Class, as follows: If y>w0, then x′(t) belongs to ω l class, and correct NumClass(l)=NumClass(l)+1; otherwise, x′(t) belongs to ω j Class, and correct NumClass(j)=NumClass(j)+1; If j<l-1, then correct j=j+1 and return to step 3S1 to continue execution; otherwise, go to step ④; In step ④, if l<L, reset l=l+1 and j=1, and return to step ③ to continue loop execution; otherwise, go to step ④ Then, the sequence number corresponding to the largest element in the array NumClass = [NumClass(1), NumClass(2), ..., NumClass(L)] is taken as the final class to which x′(t) belongs. For example, if NumClass(k) is the maximum value, the pulse waveform x′(t) belongs to ω k kind.

Images