Ray CT dosimeter fault detection method

Abstract

The invention discloses a fault detection method for a ray CT dosimeter, and relates to the field of fault detection.According to the method, an improved continuous wavelet transform function and principal component analysis are combined to obtain a fault detection result of the ray CT dosimeter to be detected, parameters in the continuous wavelet transform function are optimized to obtain optimized parameters, and then the fault detection result of the ray CT dosimeter to be detected is obtained. The fault detection of the method is more accurate; according to the method, the accuracy of fault detection is improved by establishing the local kernel principal component analysis model corresponding to each local region.

Description

Technical Field

[0001] The present invention relates to the field of fault detection, and specifically, to a method for detecting faults in a ray CT dosimeter. Background Art

[0002] A ray CT dosimeter is an instrument specifically used to measure and monitor the radiation dose generated by a computed tomography (CT) device. Ray CT dosimeters are increasingly widely used in medical, industrial and other fields. However, most of the existing methods for detecting faults in ray CT dosimeters are based on traditional statistical analysis or single signal processing techniques, and have poor adaptability to the complex working environment and diverse fault modes of ray CT dosimeters. Although wavelet analysis can effectively extract the local features of signals, it has limitations in dealing with high-dimensional data and complex non-linear relationships; principal component analysis has certain advantages in reducing the dimensionality of data, but the mining of local features is not sufficient, resulting in insufficient accuracy in detecting faults in ray CT dosimeters. Therefore, the existing methods for detecting faults in ray CT dosimeters have problems of limitations or insufficient accuracy. Summary of the Invention

[0003] The object of the present invention is to solve the problems of limitations or insufficient accuracy in the existing methods for detecting faults in ray CT dosimeters.

[0004] In order to achieve the above object of the invention, the present invention provides a method for detecting faults in a ray CT dosimeter, the method comprising: Step 1: Under the normal working state of the ray CT dosimeter, collect the dose data of the ray CT dosimeter to obtain a number of training samples; Step 2: Associate all the training samples with each wavelet basis function in the wavelet basis function library to obtain the first wavelet basis function with the highest degree of association with the dose data of the ray CT dosimeter; Step 3: Optimize the parameters in the continuous wavelet transform function to obtain optimized parameters; Step 4: Obtain an improved continuous wavelet transform function based on the optimized parameters and the first wavelet basis function; Step 5: Use the improved continuous wavelet transform function to perform wavelet transform on each training sample, and each training sample correspondingly obtains several layers of wavelet coefficients. For each training sample, extract the mean, variance and energy features of each corresponding layer of wavelet coefficients, and obtain the feature vector of each training sample based on the mean, variance and energy of each layer of wavelet coefficients; Step 6: Divide each training sample into several local regions. For each local region, perform principal component analysis on the feature vector corresponding to the local region to obtain an analysis result, and establish a local kernel principal component analysis model corresponding to each local region based on the analysis result; Step 7: Collect the detection samples of the ray CT dosimeter to be detected; Step 8: Perform wavelet transform on the detection samples using the improved continuous wavelet transform function to obtain several layers of wavelet coefficients, extract the mean, variance, and energy characteristics of each layer of wavelet coefficients of the detection samples, and obtain the feature vector of the detection samples based on the mean, variance, and energy of each layer of wavelet coefficients of the detection samples; Step 9: Obtain the local region to which the feature vector of the detection sample belongs; Step 10: Obtain the corresponding first local kernel principal component analysis model based on the local region to which the feature vector of the detection sample belongs, and calculate the first projection score of the feature vector of the detection sample in the principal component space based on the first local kernel principal component analysis model and the feature vector of the detection sample; Step 11: Calculate the difference between the first projection score and the standard projection score, and obtain the fault detection result of the ray CT dosimeter to be detected based on the difference.

[0005] Among them, in this method, by optimizing the parameters in the continuous wavelet transform function to obtain the optimized parameters, the dynamic optimization of parameters is realized, the local characteristics of the signal are captured more accurately, the efficiency and adaptability of feature extraction are significantly enhanced, and then the improved continuous wavelet transform function can obtain more accurate features, and further the fault detection of this method is more accurate.

[0006] By establishing the local kernel principal component analysis model corresponding to each local region, this method realizes the flexible regulation of the kernel function parameters according to the local region data distribution, breaks the global rigid processing mode of traditional principal component analysis. This method can deeply explore the local non-linear characteristics of the data, make the fault detection focus on the subtle abnormal points, and improve the accuracy of fault detection through this refined processing.

[0007] This method starts from the initial data preprocessing, extracts features through the improved wavelet transform, then smoothly connects to the local kernel principal component analysis, and finally completes the fault detection. Each link is interrelated and affects each other, prompting the complex dosimeter data to be deeply and multi-dimensionally analyzed. Overall, the collaborative mode can efficiently detect faults.

[0008] Preferably, step 3 specifically includes: obtaining the initial scaling factor a 1 and the initial translation factor b 1, performing wavelet transform using the first wavelet basis function based on the initial scaling factor a 1 and the initial translation factor b 1 to obtain the first processing result, performing chaos and stability analysis on the data in the first processing result to obtain the analysis result, and optimizing the initial scaling factor a 1 and the initial translation factor b1. Obtain the optimized scale factor a 2 and the optimized translation factor b 2; Step 4 specifically includes: based on the first wavelet basis function, the optimized scale factor a 2 and the optimized translation factor b 2. Obtain the improved continuous wavelet transform function.

[0009] Preferably, the improved continuous wavelet transform function is: ; in, x ( t ) is the dose data of the t-th ray CT dosimeter, is the first wavelet basis function, W ( a2 , b 2) is the improved continuous wavelet transform function, t The sample number.

[0010] Preferably, the wavelet basis function library is obtained by constructing a wavelet basis function set, and obtaining the wavelet basis function library based on the wavelet basis function set.

[0011] Preferably, the correlation between the dose data of the X-ray CT dosimeter and the wavelet basis function is calculated as follows: ; in, C i is the i-th wavelet basis function Dose data from X-ray CT dosimeter x ( t ), t is the sample number, is the i-th wavelet basis function The complex conjugate of .

[0012] Preferably, the performing of promiscuity and stability analysis on the data in the first processing result specifically includes: Extracting and obtaining local data from the first processing result; Calculate the variance and entropy of local data; The clutter and stability analysis is performed through the variance and entropy of local data.

[0013] Preferably, the step 8 specifically includes: The improved continuous wavelet transform function is used to perform wavelet transform on the detection sample to obtain several layers of wavelet coefficients, where: d j ( k )for x (t the k-th wavelet coefficient of the j-th layer of ( Extract x ( t ) the mean value of the wavelet coefficients of the j-th layer μ tj , variance and energy characteristics E tj , and the calculation methods are as follows: ; ; ; where N j is the number of wavelet coefficients of the j-th layer, and k is the number of the wavelet coefficient; Obtain the feature vector of the detection sample based on the mean value, variance and energy of the wavelet coefficients of each layer of the detection sample F t ; where , m is the number of layers of wavelet transform, μ tm is x ( t ) the mean value of the wavelet coefficients of the m-th layer, is x ( t ) the variance of the wavelet coefficients of the m-th layer, E tm is x ( t ) the energy characteristic of the wavelet coefficients of the m-th layer.

[0014] Preferably, step 6 specifically includes: Perform clustering processing based on the feature vector F t of the detection sample to obtain a clustering result, and divide the training sample into p local regions based on the clustering result: R 1, R 2,..., R P ; In each local region R s , construct a kernel matrix R s based on the similarity between sample points within the local region K s , 1 ≤ s ≤ p; Center the kernel matrix K s to obtain a centered kernel matrix ; Solve the centered kernel matrix Eigenvalues λ lk and the corresponding eigenvectors v lk , k = 1, 2, ..., q, where q is the dimension of the eigenvector; According to the eigenvalues λ lk Sort the eigenvectors v lk in descending order, and select the eigenvectors of the first several positions before sorting as the principal components based on the sorting result; Establish a local kernel principal component analysis model corresponding to each local area based on the selected principal components.

[0015] Preferably, step 7 further includes: preprocessing the detection samples of the ray CT dosimeter to be detected.

[0016] Preferably, step 11 specifically includes: Calculate the absolute value of the difference between the first projection score and the standard projection score d ; Set a threshold T. When d > T, it is determined that the ray CT dosimeter to be detected is faulty; when d≤ ≤ T, it is determined that the ray CT dosimeter to be detected is normal; wherein, ; y new is the first projection score, y normal is the standard projection score.

[0017] One or more technical solutions provided by the present invention have at least the following technical effects or advantages: Compared with the traditional detection method, the detection accuracy of this method is correspondingly improved: the improved wavelet transform captures more subtle and key features, and the subsequent local kernel principal component analysis further explores the potential relationships in the data. The combination of the two can accurately locate the fault features, reduce misjudgment and missed judgment, and can also keenly detect early minor faults.

[0018] This method has wide adaptability: the adaptive wavelet transform and the adjustable local kernel principal component analysis make the whole scheme not limited to specific models of dosimeters and fixed working scenarios. Whether it is equipment with different precisions or changing industrial and medical use environments, it can be flexibly adapted to stably detect faults.

[0019] This method has high intelligence and automation characteristics: the whole scheme has a clear process, does not require manual adjustment of parameters based on experience, and the system autonomously completes data processing, analysis, and diagnosis, reducing the dependence on professional operators, improving efficiency, and reducing the uncertainty introduced by human errors. Brief Description of the Drawings

[0020] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, form a part of the present invention, and do not limit the embodiments of the present invention; Figure 1 It is a schematic flow chart of a method for detecting faults in a ray CT dosimeter. Detailed Description of the Embodiments

[0021] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention will be further described in detail below with reference to the drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other.

[0022] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described within the scope hereof. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0023] Embodiment 1; Please refer to Figure 1 , Figure 1 It is a schematic flow chart of a method for detecting faults in a ray CT dosimeter. The present invention provides a method for detecting faults in a ray CT dosimeter, and the method includes: Step 1: Under the normal working state of the ray CT dosimeter, collect the dose data of the ray CT dosimeter to obtain a number of training samples; for example, under the normal working state of the ray CT dosimeter, collect the corresponding dose data as training samples, denoted as , where x ( t ) is the t-th sample, and N is the number of samples.

[0024] Step 2: Associate all the training samples with each wavelet basis function in the wavelet basis function library to obtain the first wavelet basis function with the highest correlation degree with the dose data of the ray CT dosimeter; in the improved wavelet transform feature extraction stage, by quantifying the correlation degree between the original dose data and different wavelet basis functions (for example, calculating the matching degree using the integral form), select the first wavelet basis function with the highest correlation degree. The purpose is to adapt to the characteristics of the current data. Different wavelet basis functions have different time-frequency characteristics. Selecting the wavelet basis function that best fits the data can more accurately capture the local features of the signal (such as mutations, trends, etc.). This adaptability avoids the feature extraction deviation caused by the fixed wavelet basis in the traditional method, thereby significantly improving the accuracy of subsequent fault detection.

[0025] Among them, the method for obtaining the wavelet basis function library is as follows: construct a set of wavelet basis functions, and obtain the wavelet basis function library based on the set of wavelet basis functions. The specific implementation method can be: form a set of wavelet basis functions, incorporate Coiflet, Meyer wavelets, etc. together to form a wavelet basis function library, where is the i-th wavelet basis function, i i = 1, 2, …, n, n and n is the total number of wavelet basis functions in the wavelet basis function library. For the original dose data collected by the ray CT dosimeter x ( t ), use the integral form to quantify x ( t ) and the correlation degree of each wavelet basis function. Subsequently, pick out C i the largest one in , and the corresponding

[0026] is the first wavelet basis function that best fits the data at present. a Step 3: Optimize the parameters in the continuous wavelet transform function to obtain the optimized parameters; the core purpose of parameter optimization (dynamically adjusting the scale factor a and the translation factor b) is to enhance the flexibility of the wavelet transform so that it can adapt to the local characteristics of the data; specifically: assign empirical initial values to the initial scale factor b a1 and the initial translation factor

[0027] b1. During the wavelet transform of the training samples using the first wavelet basis function, continuously track the characteristics of the local data. In addition to calculating the variance of the local data, it is also necessary to calculate the entropy value of the local data. During the wavelet transform, it is necessary to track both the variance and the entropy value of the local data at the same time. The variance reflects the degree of data dispersion, and the entropy value characterizes the complexity of the data (such as randomness). The combination of the two can more comprehensively describe the characteristics of the local signal and provide a multi-dimensional basis for parameter optimization. Among them, assuming that the local data is L , the window length is μ Ju N, and the mean value of the data within the window is , then the calculation formula for the local variance is: Among them, ju ( i ) is the i k-th data point within the window.

[0028] Among them, the calculation method of the entropy value H H is: , where M is the local window length selected when calculating the entropy value of the local data, p ( m1) is the probability of the data value appearing within the local window. The larger the entropy value, the more chaotic the data indicates. j 1 is the starting number of the current window. m 1 is the value number. If the entropy value is high and the variance is large, it indicates that the data fluctuates and is chaotic. In this case, according to experience, reduce a 1 to make the wavelet more focused on high-frequency details, and at the same time adjust b 1; if the entropy value is low and the variance is small, it indicates that the data is stable and orderly. In this case, according to experience, enlarge a 1 to broaden the time horizon and correspondingly adjust b 1, so as to obtain the optimized scale factor a 2 and the translation factor b 2.

[0029] Step 4: Obtain the improved continuous wavelet transform function based on the optimized parameters and the first wavelet basis function; based on the optimized scale factor a 2 and the optimized translation factor b 2, substitute them into the improved continuous wavelet transform formula for calculation.

[0030] Step 5: Use the improved continuous wavelet transform function to perform wavelet transform on each training sample. Each training sample correspondingly obtains several layers of wavelet coefficients. For each training sample, extract the mean, variance, and energy characteristics of each corresponding layer of wavelet coefficients, and obtain the feature vector of each training sample based on the mean, variance, and energy of each layer of wavelet coefficients; specifically: With the wavelet transform after optimizing the parameters, decompose the original dose data into wavelet coefficients of different scales and frequencies, where d j ( k ) is the wavelet coefficient of the j-th layer. Each layer of wavelet coefficients has its own function. The high layer captures subtle changes, and the low layer shows the overall outline. Among them, the low layer corresponds to the decomposition layer: the lower decomposition layer (such as the 1st layer, the 2nd layer). The high layer corresponds to the decomposition layer: the higher decomposition layer (such as the 5th layer, the 6th layer). The low-layer wavelet coefficients extract low-frequency information through a wide time window to show the overall structure of the signal. The high-layer wavelet coefficients extract high-frequency information through a narrow time window to focus on the local details of the signal.

[0031] For the wavelet coefficient of the j-th layer, calculate the mean , which reveals the average level of the signal in this frequency band; calculate the variance , which characterizes the discrete situation of the data; calculate the energy , in order to characterize the total intensity or power of the signal on a specific frequency component. The energy reflects the cumulative intensity of this frequency component in the entire signal. The larger the energy value, the more significant the contribution of the signal component in this frequency band to the overall signal. For each training sample x (t ) Perform improved wavelet transform and extract the mean value of wavelet coefficients μ tj , variance and energy features E tj to form a feature vector F t ; where , m is the number of layers of wavelet transform, μ tm is x ( t ) the mean value of the m-th layer wavelet coefficients of is x ( t ) the variance of the m-th layer wavelet coefficients of E tm is x ( t ) the energy feature of the m-th layer wavelet coefficients of

[0032] Step 6: Divide each training sample into several local regions. For each local region, perform principal component analysis on the feature vector corresponding to the local region to obtain an analysis result, and establish a local kernel principal component analysis model corresponding to each local region; specifically: Input the extracted feature vector F t into the kmean clustering algorithm to divide the training samples into p local regions R 1, R 2,..., R P .

[0033] In each local region R s , construct a kernel matrix R s according to the similarity between sample points within the local region K s , 1 ≤ s ≤ p; such as the Gaussian kernel function. Then, in order to eliminate the mean shift of the data in the feature space and ensure that the principal component analysis can accurately capture the variance structure of the data distribution, perform centering processing on the kernel matrix K s to obtain a centered kernel matrix . Solve the eigenvalues of the centered kernel matrix λ lk and the corresponding eigenvectors v lk , k = 1, 2,..., q, q is the dimension of the eigenvector; according to the eigenvalues λ lkSort the feature vectors in descending order v lk and select the top r feature vectors and their corresponding eigenvectors as the principal components based on the sorting result, that is, select the feature vectors that have a greater impact on the result; establish a local kernel principal component analysis model corresponding to each local region based on the selected principal components Model p .

[0034] .

[0035] Among them, Model p is the p-th model, is the r-th eigenvector of the p-th model, is the r-th eigenvalue of the p-th model, μ p is the mean vector in the principal component space

[0036] Step 7: Collect the detection samples of the ray CT dosimeter to be detected; specifically: Collect the data of the ray CT dosimeter to be detected x ( new ), perform preprocessing on it, such as removing noise (eliminating irrelevant interference in the signal), normalization (adjusting the signal to a unified scale, eliminating the dimension difference, and improving the consistency and efficiency of processing), etc., to obtain the preprocessed sample x ( pre ).

[0037] Step 8: Perform wavelet transform on the detection samples using the improved continuous wavelet transform function to obtain several layers of wavelet coefficients, extract the mean, variance, and energy features of each layer of wavelet coefficients of the detection samples, and obtain the feature vectors of the detection samples based on the mean, variance, and energy of each layer of wavelet coefficients of the detection samples; for example: perform improved wavelet transform on x ( pre ), extract features such as the mean, variance, and energy of the wavelet coefficients, and form the feature vector F new .

[0038] Step 9: Obtain the local region to which the feature vector of the detection sample belongs; specifically: according to the clustering result in the training stage, use the Euclidean distance as the similarity measurement formula and determine the local region to which F new belongs R new .

[0039] Step 10: Obtain the corresponding first local kernel principal component analysis model based on the local region to which the feature vector of the detection sample belongs. Calculate the first projection score of the feature vector of the detection sample in the principal component space based on the first local kernel principal component analysis model and the feature vector of the detection sample. Specifically: Substitute F new into R new the corresponding local kernel principal component analysis model, and calculate its projection score in the principal component space y new .

[0040] Step 11: Calculate the difference between the first projection score and the standard projection score, and obtain the fault detection result of the ray CT dosimeter to be detected based on the difference. Specifically: Calculate the absolute value of the difference between the first projection score and the standard projection score d ; Set a threshold T. When d > T, it is determined that the ray CT dosimeter to be detected is faulty; when d≤ ≤ T, it is determined that the ray CT dosimeter to be detected is normal; where ; y new is the first projection score, y normal is the standard projection score, and the size of the threshold T can be adjusted according to actual needs.

[0041] The present invention intelligently matches on demand, dynamically optimizes parameters, captures the local features of signals more accurately, and significantly enhances the efficiency and adaptability of feature extraction.

[0042] The present invention flexibly regulates the kernel function parameters according to the local data distribution, breaking the global and rigid processing mode of traditional principal component analysis. It can deeply explore the local non-linear characteristics of data, enabling fault detection to focus on extremely subtle abnormal points, reflecting refined processing.

[0043] The present invention starts from the initial data preprocessing, extracts features through improved wavelet transform, then smoothly connects to local kernel principal component analysis, and finally completes fault detection. Each link is interrelated and affects each other, prompting in-depth and multi-dimensional analysis of complex dosimeter data. This collaborative mode is crucial for efficiently screening out faults.

[0044] The present invention has the following advantages: Improved detection accuracy: The improved wavelet transform captures more subtle and key features, and subsequent local kernel principal component analysis further explores the potential relationships in the data. The combination of the two can accurately locate fault features, reduce misjudgment and missed judgment situations, and can also keenly detect early minor faults.

[0045] Wide adaptability: Adaptive wavelet transform and adjustable local kernel principal component analysis make the entire solution not limited to specific models of dosimeters and fixed working scenarios. Whether it is equipment with different precisions or changing industrial and medical usage environments, it can be flexibly adapted to stably detect faults.

[0046] Intelligence and automation: The entire solution has a clear process and does not require manual adjustment of parameters based on experience. The system autonomously completes data processing, analysis, and diagnosis, reducing the dependence on professional operators, improving efficiency, and reducing uncertainties introduced by human errors.

[0047] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present invention.

[0048] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A method for detecting faults in a radiation CT dosimeter, characterized in that, The method includes: Step 1: Under the normal working state of the X-ray CT dosimeter, collect the dose data of the X-ray CT dosimeter to obtain a number of training samples; Step 2: Associate all the training samples with each wavelet basis function in the wavelet basis function library to obtain the first wavelet basis function with the highest correlation degree with the dose data of the X-ray CT dosimeter; Step 3: Optimize the parameters in the continuous wavelet transform function to obtain the optimized parameters; Step 4: Obtain an improved continuous wavelet transform function based on the optimized parameters and the first wavelet basis function; Step 5: Use the improved continuous wavelet transform function to perform wavelet transform on each training sample. Each training sample correspondingly obtains several layers of wavelet coefficients. For each training sample, extract the mean, variance, and energy features of each corresponding layer of wavelet coefficients, and obtain the feature vector of each training sample based on the mean, variance, and energy of each layer of wavelet coefficients; Step 6: Divide each training sample into several local regions. For each local region, perform principal component analysis on the feature vector corresponding to the local region to obtain an analysis result, and establish a local kernel principal component analysis model corresponding to each local region based on the analysis result; Step 7: Collect the detection samples of the X-ray CT dosimeter to be detected; Step 8: Use the improved continuous wavelet transform function to perform wavelet transform on the detection samples to obtain several layers of wavelet coefficients, extract the mean, variance, and energy features of each layer of wavelet coefficients of the detection samples, and obtain the feature vector of the detection samples based on the mean, variance, and energy of each layer of wavelet coefficients of the detection samples; Step 9: Obtain the local region to which the feature vector of the detection sample belongs; Step 10: Obtain the corresponding first local kernel principal component analysis model based on the local region to which the feature vector of the detection sample belongs. Based on the first local kernel principal component analysis model and the feature vector of the detection sample, calculate and obtain the first projection score of the feature vector of the detection sample in the principal component space; Step 11: Calculate the difference between the first projection score and the standard projection score, and obtain the fault detection result of the X-ray CT dosimeter to be detected based on the difference.

2. The fault detection method of a ray CT dosimeter according to claim 1, characterized in that, Step 3 specifically includes: obtaining an initial scaling factor a 1 and an initial translation factor b 1, and performing wavelet transform using a first wavelet basis function based on the initial scaling factor a 1 and the initial translation factor b 1 to obtain a first processing result, analyzing the randomness and stability of the data in the first processing result to obtain an analysis result, and optimizing the initial scaling factor a 1 and the initial translation factor b 1 to obtain an optimized scaling factor a 2 and an optimized translation factor b 2; Step 4 specifically includes: obtaining an improved continuous wavelet transform function based on the first wavelet basis function, the optimized scaling factor a 2 and the optimized translation factor b 2.

3. The method for detecting a fault of a radiation CT dosimeter according to claim 2, wherein, The improved continuous wavelet transform function is: ； Among them, x ( t ) is the dose data of the t-th radiographic CT dosimeter, is the first wavelet basis function, W ( a2 , b 2) is the improved continuous wavelet transform function, t is the sample number.

4. A method for detecting faults of a ray CT dosimeter according to claim 1, characterized in that, The obtaining method of the wavelet basis function library is: construct a wavelet basis function set, and obtain the wavelet basis function library based on the wavelet basis function set.

5. A method for detecting faults in a radiation CT dosimeter according to claim 1, characterized in that, The calculation method of the correlation degree between the dose data of the X-ray CT dosimeter and the wavelet basis function is: ； Among them, C i is the i-th wavelet basis function and the dose data of the ray CT dosimeter x ( t ) the correlation degree between them, t is the sample number, is the i-th wavelet basis function the complex conjugate of.

6. A method for detecting faults in a ray CT dosimeter according to claim 1, characterized in that, The specific analysis of the clutter and stability of the data in the first processing result includes: Extract local data from the first processing result; Calculate the variance and entropy value of the local data; Perform clutter and stability analysis through the variance and entropy value of the local data.

7. A method for detecting faults of a ray CT dosimeter according to claim 1, characterized in that, The specific content of Step 8 includes: Performing wavelet transform on the detection sample by using the improved continuous wavelet transform function to obtain several layers of wavelet coefficients, where d j ( k ) is x ( t )'s wavelet coefficient of the j-th layer and the k-th one; Extraction x ( t ) Mean of the j-th layer wavelet coefficients μ tj , Variance and Energy feature E tj , The calculation methods are as follows: ； ； ； wherein, N j is the number of wavelet coefficients of the j-th layer, and k is the number of the wavelet coefficient; Obtain the feature vector of the detection sample based on the mean, variance, and energy of the wavelet coefficients of each layer of the detection sample F t ; Among them, , where m is the number of layers of wavelet transform, μ tm is x ( t ) the mean value of the m-th layer wavelet coefficients of is x ( t ) the variance of the m-th layer wavelet coefficients of E tm is x ( t ) the energy feature of the m-th layer wavelet coefficients of 8. A method for detecting faults in a radiation CT dosimeter according to claim 1, characterized in that The specific content of Step 6 includes: Feature vector based on the detection sample F t Perform clustering processing to obtain a clustering result, and divide the training samples into p local regions based on the clustering result: R 1, R 2,..., R P ; In each local region R s within, according to the local region R s construct a kernel matrix based on the similarity between sample points within K s , 1 ≤ s ≤ p; Center the kernel matrix K s to obtain the centered kernel matrix ; Solve the centralized kernel matrix for eigenvalues λ lk and the corresponding eigenvectors v lk , where k = 1, 2, ..., q and q is the dimension of the eigenvectors; According to the eigenvalues λ lk sort the eigenvectors v lk in descending order, and select the eigenvectors of the first several positions before sorting as the principal components based on the sorting result; Establish a local kernel principal component analysis model corresponding to each local region based on the selected principal components.

9. The method for detecting the failure of a ray CT dosimeter according to claim 1, characterized in that Step 7 further includes: preprocessing the detection samples of the X-ray CT dosimeter to be detected.

10. A method for detecting faults of a ray CT dosimeter according to claim 1, characterized in that, The specific content of Step 11 includes: Calculate the absolute value of the difference between the first projection score and the standard projection score d ; Set a threshold value T. When d > T, it is determined that the ray CT dosimeter to be detected is faulty; when d≤ ≤ T, it is determined that the ray CT dosimeter to be detected is normal; Among them, ; y new is the first projection score, y normal is the standard projection score.

Images