A dynamic error correction system for scanning image reconstruction

Abstract

The invention belongs to the technical field of image error correction, and discloses a dynamic error correction system for scanning image reconstruction, comprising a data acquisition module, which is used for acquiring imaging device parameter data, scanning image parameter data and environmental parameter data; a data processing module, which is used for preprocessing the acquired imaging device parameter data, scanning image parameter data and environmental parameter data to obtain a device feature data set, an image feature data set and an environmental feature data set; analyzing the influence of the environmental feature data set on the device feature data set and the image feature data set to obtain an influence feature data set; an error prediction module, which is used for constructing an error prediction model, inputting the influence feature data set into the error prediction model, and predicting a scanning reconstructed image error; an adjustment evaluation module, which is used for comparing the predicted scanning reconstructed image error with a preset scanning reconstructed image error threshold to determine whether correction is required; and ensuring the stability of image reconstruction.

Description

Technical Field

[0001] The present invention relates to the technical field of image error correction, and more specifically, to an error dynamic correction system for scanning image reconstruction. Background Art

[0002] The patent publication number CN103186883A discloses a method for correcting bone sclerosis artifacts in CT image reconstruction, including the following steps: reconstructing an actual image from raw projection data using clinically selected reconstruction parameters; then reconstructing a target image from raw projection data using the maximum scanning field of view supported by CT, the rotation center of the frame, and the standard reconstruction convolution kernel as reconstruction parameters; extracting a bone image from the target image; performing forward projection on the bone image to obtain bone projection data, and performing polynomial fitting on the bone projection data to obtain bone sclerosis error projection data; reconstructing a bone sclerosis artifact image from the bone sclerosis error projection data using clinically selected parameters as reconstruction parameters; and removing the bone sclerosis artifact image from the reconstructed actual image to obtain a corrected image. The target image reconstructed by the present invention contains all scanned bone information, so that the obtained bone sclerosis error projection data is consistent with the real projection data passing through the bone during the actual scanning process, thereby achieving a good bone sclerosis artifact correction effect.

[0003] The existing error correction system for scanning image reconstruction has the following main problems:

[0004] If the Kendall rank correlation coefficient is not used to analyze and remove features that are too strongly or too weakly correlated, a large number of redundant features may remain in the data set. This will increase the dimension of the data, resulting in increased model complexity, longer training time, and increased consumption of computing resources; if the scaling factor is not restricted, the model adjustment process may be affected by extreme scaling factors. When the coefficient is too large, the model may be too sensitive to changes in the training data, resulting in large fluctuations in the results; when the coefficient is too small, the model is not adjusted enough and it is difficult to adapt to changes in data features.

[0005] Images that lack Gaussian filtering and smoothing will contain a large amount of random noise, which will mask important image features and make subsequent processing and analysis more difficult; the model may be disturbed by noise, resulting in reduced accuracy; the filter window size is not dynamically adjusted to meet the needs of different images, and fixed filter parameters may not be able to adapt to the diversity of images; some images may require a stronger smoothing effect, while others may need to retain more detail information; input data that has not undergone reasonable image preprocessing may contain a large amount of noise and redundant information, which will increase the complexity of the model and reduce computational efficiency; at the same time, the system's resource consumption increases and the processing speed slows down.

[0006] In view of this, the present invention proposes an error dynamic correction system for scanning image reconstruction to solve the above problem. Summary of the invention

[0007] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: a dynamic error correction system for scanning image reconstruction, comprising:

[0008] A data acquisition module, used to acquire imaging device parameter data, scanned image parameter data and environmental parameter data;

[0009] The data processing module is used to pre-process the acquired imaging device parameter data, scanning image parameter data and environmental parameter data to obtain a device feature data set, an image feature data set and an environmental feature data set; analyze the influence of the environmental feature data set on the device feature data set and the image feature data set to obtain an influence feature data set;

[0010] For imaging device parameter data and environmental parameter data, feature extraction is performed by calculating the Kendall rank correlation coefficient, thereby obtaining a device feature data set and an environmental feature data set; by presetting observation samples, each observation sample corresponds to an imaging device parameter data and an environmental parameter data, and each data in the imaging device parameter data and the environmental parameter data is integrated and represented as a matrix; the number of consistent pairs and the number of inconsistent pairs between any two sample pairs are calculated; based on the number of consistent pairs and the number of inconsistent pairs, the Kendall rank correlation coefficient between two features in any observation sample is obtained, and feature extraction is performed according to the Kendall rank correlation coefficient value between two features in any observation sample, thereby obtaining a device feature data set and an environmental feature data set;

[0011] The error prediction module is used to construct an error prediction model, input the influencing feature data set into the error prediction model, and predict the error of the scanned and reconstructed image;

[0012] An adjustment evaluation module is used to compare the predicted scan reconstructed image error with a preset scan reconstructed image error threshold to determine whether correction is required;

[0013] The error diagnosis module directly outputs the scanned and reconstructed image if no correction is required; if correction is required, the error data of the scanned and reconstructed image is collected, an error type diagnosis model is constructed, and the error data of the scanned and reconstructed image is input into the error type diagnosis model to predict the error type of the reconstructed image;

[0014] The dynamic correction module is used to perform corresponding correction adjustment strategies according to the type of reconstructed image error, obtain the corrected scanned reconstructed image, and reconstruct the corrected image model through the image reconstruction algorithm. The various modules are connected by wired and / or wireless means.

[0015] Furthermore, the characteristic categories of the imaging device parameter data include the scanning angle of the device, the scanning speed of the device, the intensity of the radiation emitted by the device, the exposure duration of the device scanning and the temperature of the scanning device; the characteristic categories of the scanning image parameter data include the brightness of the image, the spatial resolution of the image, the contrast of the image, the signal-to-noise ratio of the image and the uniformity of the image; the characteristic categories of the environmental parameter data include the temperature of the imaging environment, the humidity of the imaging environment and the light intensity of the scanning environment.

[0016] Furthermore, the method for acquiring the device feature data set and the environment feature data set includes:

[0017] S31. m observation samples are preset. For each observation sample, there is an imaging device parameter data and an environmental parameter data. In the imaging device parameter data, the scanning angle of the device is recorded as X1, the scanning speed of the device is recorded as X2, the intensity of the radiation emitted by the device is recorded as X3, the exposure duration of the device scanning is recorded as X4, and the temperature of the scanning device is recorded as X5. In the environmental parameter data, the temperature of the imaging environment is recorded as X6, the humidity of the imaging environment is recorded as X7, and the light intensity of the scanning environment is recorded as X8.

[0018] S32, integrating each data in the imaging device parameter data and the environmental parameter data into an m×8 matrix D, where each row of the matrix represents a sample, and there are m rows in total; each column of the matrix represents a feature in the imaging device parameter data and the environmental parameter data, and there are 8 columns in total;

[0019] The matrix D is: Among them, X e′c′ is the value of the e′th sample on the c′th feature, e′=1,...,m, c′=1,...,8;

[0020] S33, calculating the number of consistent pairs between any two sample pairs; for any two sample pairs, judging whether the two sample pairs are consistent pairs by using a consistency judgment formula and an inconsistency judgment formula, the consistency judgment formula is:

[0021] Among them, X i′ is the i′th sample value of variable X in the imaging device parameter data and the environmental parameter data; X j′ is the j′th sample value of variable X in the imaging device parameter data and environmental parameter data; Y i′ is the i′th sample value of variable Y in the imaging device parameter data and the environmental parameter data; Y j′is the j′th sample value of variable Y in the imaging device parameter data and the environmental parameter data; I[] is an indicator function, when the condition in the brackets is met, the result is 1, otherwise it is 0; I[(X i′ -X j′ )(Y i′ -Y j′ )>0] is an indicator function used to judge two sample pairs (X i′ ,Y i′ ) and (X j′ ,Y j′ ) is a consistent pair; i′ and j′ are the indexes of the sample values;

[0022] If (X i′ -X j′ )>0 and (Y i′ -Y j′ )>0, or (X i′ -X j′ )<0 and (Y i′ -Y j′ )<0, then this pair is a consistent pair and returns 1; otherwise, returns 0; by summing all possible sample pairs, calculate the number C of all consistent pairs;

[0023] The inconsistency judgment formula is:

[0024]

[0025] ; Among them, I[(X i′ -X j′ )(Y i′ -Y j′ )<0] is an indicator function used to judge the two sample pairs (x i ,y i ) and (x j ,y j ) is an inconsistent pair;

[0026] If (X i′ -X j′ )>0 and (Y i′ -Y j′ )<0, or (X i′ -X j′ )<0 and (Y i′ -Y j′ )>0, then this pair is an inconsistent pair and returns 1; otherwise, returns 0; by summing all possible sample pairs, calculate the number D of all inconsistent pairs;

[0027] S34, given imaging device parameter data and environmental parameter data, two features X in any observation sample Z and X f , calculate XZ and X f The Kendall rank correlation coefficient between them is calculated as: Where τ is X Z and X f The Kendall rank correlation coefficient between them; C is the number of consistent pairs between any two sample pairs; D is the number of inconsistent pairs between any two sample pairs; n is the number of samples; δ is the scaling factor for adjusting the magnitude of the Kendall rank correlation coefficient τ; z and f are the indexes of the sample values;

[0028] The scaling factor δ for adjusting the magnitude of the Kendall rank correlation coefficient τ is limited by a scaling factor limiting formula, and the scaling factor limiting formula is: Among them, α is a positive constant, representing the maximum value of the scaling factor; β is a positive constant used to control the minimum value of the scaling factor; l is the number of features; k is the type of feature;

[0029] S35. The value range of τ is between [-1,1]; a positive τ value indicates positive correlation, a negative τ value indicates negative correlation, and 0 indicates no correlation; when the |τ| value is close to 1, it indicates that there is a strong correlation between the two features; when it is close to 0, it indicates that there is a weak correlation between the two features; according to the calculated τ value, select and remove the features of |τ|∈[h′,1]∪[0,λ′] to obtain the device feature data set and the environment feature data set; where h′ is the minimum strong correlation threshold; λ′ is the maximum weak correlation threshold.

[0030] Furthermore, the method for acquiring the image feature data set includes:

[0031] Smoothing the scanned image parameter data, enhancing the contrast and extracting the image edge, and then obtaining the image feature data set;

[0032] S41, the smoothing method is: smoothing the image by using a Gaussian filter kernel function; the Gaussian filter kernel function is: Among them, I′(a,b) is the pixel value of the original image; (a,b) is the horizontal and vertical coordinates of the image; I″(a,b) is the pixel value of the image after Gaussian filtering; G σ (a, b) is the Gaussian kernel function; σ is the standard deviation of the Gaussian function, which controls the smoothness of the filter; g is the size of the filter window; p is the horizontal coordinate of the center point of the filter; q is the vertical coordinate of the center point of the filter;

[0033] The standard deviation σ of the Gaussian function is adjusted by a smoothness adjustment formula, and the smoothness adjustment formula is: Where H is the height of the image; W is the width of the image;

[0034] The size g of the filter window is limited by the filter window radius limiting formula, and the filter window radius limiting formula is: Among them, s is the limiting factor that controls the speed of window size change;

[0035] S42, the method for enhancing contrast is: enhancing the contrast in the image by using a histogram equalization formula; the histogram equalization formula is: Among them, I * (a, b) are the image pixel values ​​after histogram equalization; h(d) is the histogram of gray level d in the image; L′ is the maximum gray level; is the floor rounding function;

[0036] S43, the method for extracting the edge of the image is: calculating the second-order derivative of the image by using the Laplace operator to enhance the edge in the image, using the bicubic interpolation method to improve the resolution of the image, and using the Sobel algorithm to calculate the gradient of the image, and detecting and extracting the edge of the image by calculating the modulus of the gradient to obtain an image feature data set;

[0037] Calculate the gradient of an image in the horizontal direction: Among them, G a is the horizontal gradient of the image;

[0038] Calculate the gradient of an image in the vertical direction:

[0039] Among them, G b is the gradient of the image in the vertical direction;

[0040] Calculate the magnitude of the gradient: Among them, G is the modulus of the gradient;

[0041] The acquired device feature dataset, image feature dataset and environment feature dataset are processed by outlier elimination and standard deviation normalization, and converted into standard normal distribution to obtain normalized device feature dataset, image feature dataset and environment feature dataset.

[0042] Furthermore, the method of analyzing the influence of the environment feature data set on the device feature data set and the image feature data set to obtain the influenced feature data set includes:

[0043] Construct a multivariate regression model to quantify the comprehensive impact of the environmental feature data set on the accuracy of the device feature data set and the image feature data set, and obtain the influencing feature data; respectively establish a device feature regression model and an image feature regression model;

[0044] The equipment feature regression model is:

[0045] Y de=ε0+ε1X en1 +ε2X en2 +…+ε v X env +∈ de ; Among them, Y de is the device feature dataset; X env is the vth environmental feature; ∈ de is the error term of the model; ε0 is the constant term of the regression model; ε1 is the regression coefficient of the first environmental characteristic; ε m is the regression coefficient of the mth environmental characteristic; ε v is the regression coefficient of the vth environmental characteristic;

[0046] The image feature regression model is:

[0047] Y ige =ε0+ε1X en1 +ε2X en2 +…+ε v X env +∈ ige ; Among them, Y ige is the image feature dataset; ∈ ige is the error term of the model.

[0048] Furthermore, the method for constructing the error prediction model includes:

[0049] The historical impact feature dataset and the corresponding scanned reconstructed image errors are used as sample sets; the sample set is used as a dataset and the dataset is divided into a training set, a validation set and a test set; an error prediction model is constructed, and the error prediction model includes input data and output labels; the training set is used to train the parameters of the machine learning model, the validation set is used to monitor the model performance during the training process and adjust the model hyperparameters, and the test set is used to finally evaluate the performance of the trained model;

[0050] The input data of the model is a historical impact feature data set, and the output label of the model is a scanned reconstructed image error; RBF is selected as the kernel function of the model, and the hyperparameters of the model are initialized. The hyperparameters of the model include a regularization parameter C and a γ value of the RBF kernel function; the error prediction model is a support vector machine model;

[0051] Define the loss function, and use the L2 regularized mean square error loss function to measure the difference between the output scanned reconstructed image error and the actual scanned reconstructed image error; the L2 regularized mean square error loss function is:

[0052] Where n' is the sample size, is the predicted value of the b′th sample; y b′is the actual value of the b′th sample, λ is the regularization coefficient; N is the number of training samples; ∑ w w 2 is the sum of squares of all trainable weight parameters; w is the model weight parameter;

[0053] Use the training set data to train the model and minimize the loss function using the Adam optimizer. Use the validation set to evaluate the performance of the model and tune the model's hyperparameters until the model performance no longer improves significantly or reaches the preset stopping condition.

[0054] The test set is used to evaluate the performance of the model in the prediction task, and the current influencing feature data set is input into the trained error prediction model to obtain the scanned reconstructed image error.

[0055] Furthermore, the method of comparing the predicted scan reconstructed image error with a preset scan reconstructed image error threshold to determine whether correction is required includes:

[0056] If the predicted scan reconstructed image error is greater than or equal to a preset scan reconstructed image error threshold, it is determined that correction is required;

[0057] If the predicted scan reconstructed image error is less than a preset scan reconstructed image error threshold, it is determined that no correction is required.

[0058] Furthermore, the method of constructing the error type diagnosis model, inputting the scanned reconstructed image error data into the error type diagnosis model, and predicting the reconstructed image error type includes:

[0059] The scanned reconstructed image error data includes the pixel value of the image, the structural similarity measurement of the image, the signal-to-noise ratio of the image and the spatial resolution of the image; the structural similarity measurement of the image includes the brightness of the image, the contrast of the image and the structural similarity of the image;

[0060] The data set is divided into a training set, a validation set, and a test set; the data set includes historical scanned reconstructed image error data and the corresponding reconstructed image error types; an error type diagnosis model is constructed, and the error type diagnosis model includes an input layer, a convolution layer, a pooling layer, a batch normalization layer, a fully connected layer, and an output layer;

[0061] The input layer of the model is used to input historical scan reconstruction image error data, and the output layer of the model is used to output the reconstruction image error type; softmax is used as the activation function; the error type diagnosis model is a 3D convolutional neural network model;

[0062] Multi-classification cross entropy is used as the loss function of the model to measure the difference between the predicted value and the actual value of the model; the multi-classification cross entropy loss function is:

[0063] Among them, L mro is the average loss of the data set; N′ is the total number of samples in the data set; A is the number of reconstructed image error types; y b′a is the true label of the b′th sample for the ath type; p b′a The probability that the b′th sample predicted by the model belongs to the ath type;

[0064] Use the training set data to train the model and minimize the loss function by selecting the SGD optimizer; use the validation set to evaluate the performance of the error type diagnosis model, calculate the accuracy index to measure the performance of the model, and tune the model's hyperparameters to improve the model performance;

[0065] The performance of the model in the prediction task is evaluated through the test set. When the performance of the model in the prediction task reaches the preset performance threshold, the test is stopped to obtain the error type diagnosis model. The trained error type diagnosis model is used to predict the error data of the current scanned reconstructed image to obtain the reconstructed image error type.

[0066] Furthermore, the method of performing a corresponding correction adjustment strategy according to the reconstructed image error type to obtain a corrected scanned reconstructed image includes:

[0067] According to different types of reconstructed image errors, corresponding correction and adjustment strategies are adopted, specifically:

[0068] If the type of reconstructed image error is geometric distortion, use the control point method or image registration technology to correct it; select the same control points in the original image and the target image; use the correspondence between the control points and use the least squares method to calculate the transformation matrix; use the obtained transformation matrix to perform a geometric transformation on the original image to obtain a corrected image;

[0069] If the type of reconstructed image error is uneven brightness, gamma correction is used to improve brightness contrast and adjust image intensity; the best gamma value is selected through cross-validation according to the specific uneven brightness of the image; the gamma correction formula is applied to each pixel value in the image to calculate a new intensity value; the corrected intensity value is used to replace the corresponding pixel in the original image to generate a corrected image;

[0070] If the type of reconstructed image error is noise disturbance, wavelet transform is used to suppress random noise in the image; the noisy image is subjected to wavelet transform to decompose the image into sub-bands of different scales and directions; soft threshold is used to process the coefficients of each sub-band after wavelet decomposition. The specific formula is: Among them, w C′ is the wavelet coefficient; T is the threshold; sgn(w C′ ) is the wavelet coefficient w C′The sign function of ; C′ is the index of the wavelet coefficient;

[0071] Perform inverse wavelet transform on the wavelet coefficients after threshold processing to reconstruct the denoised image;

[0072] If the type of reconstructed image error is an artifact problem, a de-artifacting algorithm is used to correct the artifacts caused by the reconstruction algorithm and equipment limitations;

[0073] If the type of reconstructed image error is insufficient resolution, super-resolution reconstruction technology is used to improve the quality and details of the image; the low-resolution image is Fourier transformed, and then a high-resolution image is reconstructed in the spatial domain through an optimization algorithm;

[0074] If the type of reconstructed image error is reconstruction algorithm error, an optimized interpolation method is used to adaptively adjust the reconstruction algorithm.

[0075] Furthermore, the image reconstruction algorithm is used to reconstruct the corrected image model.

[0076] Types of methods include:

[0077] Reconstructing the rectified image by a bilinear interpolation algorithm, generating an image model from the rectified image; calculating a new value at each pixel position of the image based on the values ​​of four pixels around the image by a bilinear interpolation formula;

[0078] The bilinear interpolation formula is:

[0079] O(p′,q′)=(1-A′)(1-B′)O 11 +A′(1-B′)O 21 +(1-A′)B′O 12 +A′B′O 22 ; Wherein, O(p′,q′) is the pixel value of the interpolated image at the position (p′,q′); O 11 , O 21 , O 12 and O 22 are the values ​​of the four neighboring pixels surrounding the position point (p′, q′); A′ and B′ are the relative positions of the position point (p′, q′) in the neighboring pixels.

[0080] The technical effects and advantages of the error dynamic correction system for scanning image reconstruction of the present invention are as follows:

[0081] By calculating the Kendall rank correlation coefficient between features, the present invention can selectively remove features with too strong or too weak correlation; effectively reduce the dimension of data, reduce the risk of overfitting, and improve the generalization ability and computational efficiency of machine learning models; by limiting the scaling factor, it can avoid the problem of excessive fluctuation of results due to too large a coefficient, or insufficient model adjustment due to too small a coefficient; appropriate restrictions can make the model more stable in data preprocessing and reduce unnecessary deviations and noise;

[0082] By smoothing the image with the Gaussian kernel function, the noise in the image can be effectively removed and important image information can be retained; the random noise in the image can be reduced, and the global information structure of the image can be retained while smoothing the image; the degree of image smoothing can be controlled by adjusting the standard deviation of the Gaussian function. The larger the standard deviation, the stronger the smoothing effect of the filter; by dynamically adjusting the window size of the filter, it can adapt to the needs of different images; through reasonable image preprocessing, the noise and redundant information in the image are reduced, and more important features are extracted, thereby reducing the complexity of model calculation, improving calculation efficiency, and reducing system resource consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 A schematic diagram of the structure of a dynamic error correction system for scanning image reconstruction according to the present invention;

[0084] Figure 2 The figure is a flow chart of a method for dynamic error correction for scanning image reconstruction according to the present invention. DETAILED DESCRIPTION

[0085] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0086] Embodiment 1

[0087] See also Figure 1 As shown, the error dynamic correction system for scanning image reconstruction described in this embodiment includes:

[0088] A data acquisition module, used to acquire imaging device parameter data, scanned image parameter data and environmental parameter data;

[0089] The data processing module is used to pre-process the acquired imaging device parameter data, scanning image parameter data and environmental parameter data to obtain a device feature data set, an image feature data set and an environmental feature data set; analyze the influence of the environmental feature data set on the device feature data set and the image feature data set to obtain an influence feature data set;

[0090] For imaging device parameter data and environmental parameter data, feature extraction is performed by calculating the Kendall rank correlation coefficient, thereby obtaining a device feature data set and an environmental feature data set; by presetting observation samples, each observation sample corresponds to an imaging device parameter data and an environmental parameter data, and each data in the imaging device parameter data and the environmental parameter data is integrated and represented as a matrix; the number of consistent pairs and the number of inconsistent pairs between any two sample pairs are calculated; based on the number of consistent pairs and the number of inconsistent pairs, the Kendall rank correlation coefficient between two features in any observation sample is obtained, and feature extraction is performed according to the Kendall rank correlation coefficient value between two features in any observation sample, thereby obtaining a device feature data set and an environmental feature data set;

[0091] The error prediction module is used to construct an error prediction model, input the influencing feature data set into the error prediction model, and predict the error of the scanned and reconstructed image;

[0092] An adjustment evaluation module is used to compare the predicted scan reconstructed image error with a preset scan reconstructed image error threshold to determine whether correction is required;

[0093] The error diagnosis module directly outputs the scanned and reconstructed image if no correction is required; if correction is required, the error data of the scanned and reconstructed image is collected, an error type diagnosis model is constructed, and the error data of the scanned and reconstructed image is input into the error type diagnosis model to predict the error type of the reconstructed image;

[0094] The dynamic correction module is used to perform corresponding correction adjustment strategies according to the type of reconstructed image error, obtain the corrected scanned reconstructed image, and reconstruct the corrected image model through the image reconstruction algorithm. The various modules are connected by wired and / or wireless means.

[0095] The characteristic categories of the imaging device parameter data include the scanning angle of the device, the scanning speed of the device, the intensity of the radiation emitted by the device, the exposure duration of the device scanning and the temperature of the scanning device; the characteristic categories of the scanning image parameter data include the brightness of the image, the spatial resolution of the image, the contrast of the image, the signal-to-noise ratio of the image and the uniformity of the image; the characteristic categories of the environmental parameter data include the temperature of the imaging environment, the humidity of the imaging environment and the light intensity of the scanning environment.

[0096] The scanning angle of the device is obtained by a scanning angle sensor, the scanning speed of the device is obtained by a scanning speed sensor, the intensity of the radiation emitted by the device is obtained by a radiation intensity sensor; the exposure duration of the device scanning is obtained by an exposure duration sensor; the temperature of the scanning device is obtained by a device temperature sensor;

[0097] The brightness of the image is obtained by a brightness sensor, the spatial resolution of the image is obtained by a spatial resolution sensor, the contrast of the image and the signal-to-noise ratio of the image are obtained by a contrast and signal-to-noise ratio analyzer; the uniformity of the image is obtained by an optical uniformity measuring instrument;

[0098] The temperature of the imaging environment is acquired through a temperature sensor, the humidity of the imaging environment is acquired through a humidity sensor, and the light intensity of the scanning environment is acquired through a photoelectric sensor.

[0099] The method for acquiring the device feature data set and the environment feature data set includes:

[0100] S31. m observation samples are preset. For each observation sample, there is an imaging device parameter data and an environmental parameter data. In the imaging device parameter data, the scanning angle of the device is recorded as X1, the scanning speed of the device is recorded as X2, the intensity of the radiation emitted by the device is recorded as X3, the exposure duration of the device scanning is recorded as X4, and the temperature of the scanning device is recorded as X5. In the environmental parameter data, the temperature of the imaging environment is recorded as X6, the humidity of the imaging environment is recorded as X7, and the light intensity of the scanning environment is recorded as X8.

[0101] S32, integrating each data in the imaging device parameter data and the environmental parameter data into an m×8 matrix D, where each row of the matrix represents a sample, and there are m rows in total; each column of the matrix represents a feature in the imaging device parameter data and the environmental parameter data, and there are 8 columns in total;

[0102] The matrix D is: Among them, X e′c′ is the value of the e′th sample on the c′th feature, e′=1,...,m, c′=1,...,8;

[0103] S33, calculating the number of consistent pairs between any two sample pairs; for any two sample pairs, judging whether the two sample pairs are consistent pairs by using a consistency judgment formula and an inconsistency judgment formula, the consistency judgment formula is:

[0104] Among them, X i′ is the i′th sample value of variable X in the imaging device parameter data and the environmental parameter data; X j′ is the j′th sample value of variable X in the imaging device parameter data and environmental parameter data; Yi′ is the i′th sample value of variable Y in the imaging device parameter data and the environmental parameter data; Y j′ is the j′th sample value of variable Y in the imaging device parameter data and the environmental parameter data; I[] is an indicator function, when the condition in the brackets is met, the result is 1, otherwise it is 0; I[(X i′ -X j′ )(Y i′ -Y j′ )>0] is an indicator function used to judge two sample pairs (X i′ ,Y i′ ) and (X j′ ,Y j′ ) is a consistent pair; i′ and j′ are the indexes of the sample values;

[0105] If (X i′ -X j′ )>0 and (Y i′ -Y j′ )>0, or (X i′ -X j′ )<0 and (Y i′ -Y j′ )<0, then this pair is a consistent pair and returns 1; otherwise, returns 0; by summing all possible sample pairs, calculate the number C of all consistent pairs;

[0106] The inconsistency judgment formula is:

[0107]

[0108] Among them, I[(X i′ -X j′ )(Y i′ -Y j′ )<0] is an indicator function used to judge the two sample pairs (x i ,y i ) and (x j ,y j ) is an inconsistent pair;

[0109] If (X i′ -X j′ )>0 and (Y i′ -Y j′ )<0, or (X i′ -X j′ )<0 and (Y i′ -Y j′ )>0, then this pair is an inconsistent pair and returns 1; otherwise, returns 0; by summing all possible sample pairs, calculate the number D of all inconsistent pairs;

[0110] S34, given imaging device parameter data and environmental parameter data, two features X in any observation sample Z and X f , calculate X Z and X f The Kendall rank correlation coefficient between them is calculated as: Where τ is X Z and X f The Kendall rank correlation coefficient between them; C is the number of consistent pairs between any two sample pairs; D is the number of inconsistent pairs between any two sample pairs; n is the number of samples; δ is the scaling factor for adjusting the magnitude of the Kendall rank correlation coefficient τ; z and f are the indexes of the sample values;

[0111] The scaling factor δ for adjusting the magnitude of the Kendall rank correlation coefficient τ is limited by a scaling factor limiting formula, and the scaling factor limiting formula is: Among them, α is a positive constant, representing the maximum value of the scaling factor; β is a positive constant used to control the minimum value of the scaling factor; l is the number of features; k is the type of feature;

[0112] S35, the value range of τ is between [-1,1]; a positive τ value indicates positive correlation, a negative τ value indicates negative correlation, and 0 indicates no correlation; when the |τ| value is close to 1, it indicates that there is a strong correlation between the two features; when it is close to 0, it indicates that there is a weak correlation between the two features; according to the calculated τ value, select and remove the features of |τ|∈[h′,1]∪[0,λ′] to obtain the device feature data set and the environment feature data set; where h′ is the minimum strong correlation threshold; λ′ is the maximum weak correlation threshold; for example, select and remove the features of |τ|∈[0.8,1]∪[0,0.2] to obtain the device feature data set and the environment feature data set;

[0113] For example, suppose there are the following 4 samples of data, two features X i and X j :

[0114] Sample No. <![CDATA[X i (Scanning speed)]]> <![CDATA[X j (Ray Intensity)]]> 1 5 100 2 6 110 3 7 120 4 8 90

[0115] Compute all possible pairs of samples and check whether they are consistent pairs:

[0116] Sample pair (1,2): X i1 =5,X i2 =6;X j1 =100,X j2 =110;X i1 <X i2 And X j1 <X j2 , is consistent with each other.

[0117] Sample pair (1,3): X i1 =5,X i3 =7;X j1 =100,X j3 =120;X i1 <X i3 And X j1 <X j3 , is consistent with each other.

[0118] Sample pair (1,4): X i1 =5,X i4 =8;X j1 =100,X j4 =90;X i1 <X i4 And X j1 >X j4 , is an inconsistent pair.

[0119] Sample pair (2,3): X i2 =6,X i3 =7;X j2 =110,X j3 =120;X i2 <X i3 And X j2 <X j3 , is consistent with each other.

[0120] Sample pair (2,4): X i2 =6,X i4 =8;X j2 =110,X j4 =90;X i2 <X i4 And X j2 >X j4 , is an inconsistent pair.

[0121] Sample pair (3,4): X i3 =7,X i4 =8;X j3 =120,X j4 =90;X i3 <X i4 And X j3 >X j4 , is an inconsistent pair.

[0122] The method for acquiring the image feature data set includes:

[0123] Smoothing the scanned image parameter data, enhancing the contrast and extracting the image edge, and then obtaining the image feature data set;

[0124] S41, the smoothing method is: smoothing the image by using a Gaussian filter kernel function; the Gaussian filter kernel function is: Among them, I′(a,b) is the pixel value of the original image; (a,b) is the horizontal and vertical coordinates of the image; I"(a,b) is the pixel value of the image after Gaussian filtering; G σ (a, b) is the Gaussian kernel function; σ is the standard deviation of the Gaussian function, which controls the smoothness of the filter; g is the size of the filter window; p is the horizontal coordinate of the center point of the filter; q is the vertical coordinate of the center point of the filter;

[0125] The standard deviation σ of the Gaussian function is adjusted by a smoothness adjustment formula, and the smoothness adjustment formula is: Where H is the height of the image; W is the width of the image;

[0126] The size g of the filter window is limited by the filter window radius limiting formula, and the filter window radius limiting formula is: Among them, s is the limiting factor that controls the speed of window size change;

[0127] S42, the method for enhancing contrast is: enhancing the contrast in the image by using a histogram equalization formula; the histogram equalization formula is: Among them, I * (a, b) are the image pixel values ​​after histogram equalization; h(d) is the histogram of gray level d in the image; L′ is the maximum gray level; is the floor rounding function;

[0128] S43, the method for extracting the edge of the image is: calculating the second-order derivative of the image by using the Laplace operator to enhance the edge in the image, using the bicubic interpolation method to improve the resolution of the image, and using the Sobel algorithm to calculate the gradient of the image, and detecting and extracting the edge of the image by calculating the modulus of the gradient to obtain an image feature data set;

[0129] Calculate the gradient of an image in the horizontal direction: Among them, G a is the horizontal gradient of the image;

[0130] Calculate the gradient of an image in the vertical direction: Among them, G b is the gradient of the image in the vertical direction;

[0131] Calculate the magnitude of the gradient: Among them, G is the modulus of the gradient;

[0132] The acquired device feature dataset, image feature dataset and environment feature dataset are processed by outlier elimination and standard deviation normalization, and converted into standard normal distribution to obtain normalized device feature dataset, image feature dataset and environment feature dataset.

[0133] The method of analyzing the influence of the environment feature data set on the device feature data set and the image feature data set to obtain the influence feature data set includes:

[0134] Construct a multivariate regression model to quantify the comprehensive impact of the environmental feature data set on the accuracy of the device feature data set and the image feature data set, and obtain the influencing feature data; respectively establish a device feature regression model and an image feature regression model;

[0135] The equipment feature regression model is:

[0136] Y de =ε0+ε1X en1 +ε2X en2 +…+ε v X env +∈ de ; Among them, Y de is the device feature dataset; X env is the vth environmental feature; ∈ de is the error term of the model; ε0 is the constant term of the regression model; ε1 is the regression coefficient of the first environmental characteristic; ε m is the regression coefficient of the mth environmental characteristic; ε v is the regression coefficient of the vth environmental characteristic;

[0137] The image feature regression model is:

[0138] Y ige =ε0+ε1X en1 +ε2X en2 +…+ε v X env +∈ ige ; Among them, Y ige is the image feature dataset; ∈ ige is the error term of the model.

[0139] The construction method of the error prediction model includes:

[0140] The historical impact feature dataset and the corresponding scanned reconstructed image errors are used as sample sets; the sample set is used as a dataset and the dataset is divided into a training set, a validation set and a test set; an error prediction model is constructed, and the error prediction model includes input data and output labels; the training set is used to train the parameters of the machine learning model, the validation set is used to monitor the model performance during the training process and adjust the model hyperparameters, and the test set is used to finally evaluate the performance of the trained model;

[0141] The input data of the model is a historical impact feature data set, and the output label of the model is a scanned reconstructed image error; RBF is selected as the kernel function of the model, and the hyperparameters of the model are initialized. The hyperparameters of the model include a regularization parameter C and a γ value of the RBF kernel function; the error prediction model is a support vector machine model;

[0142] Define the loss function, and use the L2 regularized mean square error loss function to measure the difference between the output scanned reconstructed image error and the actual scanned reconstructed image error; the L2 regularized mean square error loss function is:

[0143] Where n' is the sample size, is the predicted value of the b′th sample; y b′ is the actual value of the b′th sample, λ is the regularization coefficient, which controls the regularization strength; N is the number of training samples; ∑ w w 2 is the sum of squares of all trainable weight parameters; w is the model weight parameter;

[0144] Use the training set data to train the model and minimize the loss function using the Adam optimizer. Use the validation set to evaluate the performance of the model and tune the model's hyperparameters until the model performance no longer improves significantly or reaches the preset stopping condition.

[0145] The test set is used to evaluate the performance of the model in the prediction task, and the current influencing feature data set is input into the trained error prediction model to obtain the scanned reconstructed image error.

[0146] The method of comparing the predicted scan reconstructed image error with a preset scan reconstructed image error threshold to determine whether correction is required includes:

[0147] If the predicted scan reconstructed image error is greater than or equal to a preset scan reconstructed image error threshold, it is determined that correction is required;

[0148] If the predicted scan reconstructed image error is less than a preset scan reconstructed image error threshold, it is determined that no correction is required.

[0149] The method of constructing the error type diagnosis model, inputting the scanned reconstructed image error data into the error type diagnosis model, and predicting the reconstructed image error type includes:

[0150] The scanned reconstructed image error data includes the pixel value of the image, the structural similarity measurement of the image, the signal-to-noise ratio of the image and the spatial resolution of the image; the structural similarity measurement of the image includes the brightness of the image, the contrast of the image and the structural similarity of the image;

[0151] The data set is divided into a training set, a validation set, and a test set; the data set includes historical scanned reconstructed image error data and the corresponding reconstructed image error types; an error type diagnosis model is constructed, and the error type diagnosis model includes an input layer, a convolution layer, a pooling layer, a batch normalization layer, a fully connected layer, and an output layer;

[0152] The input layer of the model is used to input historical scan reconstruction image error data, and the output layer of the model is used to output the reconstruction image error type; softmax is used as the activation function; the error type diagnosis model is a 3D convolutional neural network model;

[0153] Multi-classification cross entropy is used as the loss function of the model to measure the difference between the predicted value and the actual value of the model; the multi-classification cross entropy loss function is:

[0154] Among them, L mro is the average loss of the data set; N′ is the total number of samples in the data set; A is the number of reconstructed image error types; y b′a is the true label of the b′th sample for the ath type; p b′a The probability that the b′th sample predicted by the model belongs to the ath type;

[0155] Use the training set data to train the model and minimize the loss function by selecting the SGD optimizer; use the validation set to evaluate the performance of the error type diagnosis model, calculate the accuracy index to measure the performance of the model, and tune the model's hyperparameters to improve the model performance;

[0156] The performance of the model in the prediction task is evaluated through the test set. When the performance of the model in the prediction task reaches the preset performance threshold, the test is stopped to obtain the error type diagnosis model. The trained error type diagnosis model is used to predict the error data of the current scanned reconstructed image to obtain the reconstructed image error type.

[0157] According to the type of reconstructed image error, a corresponding correction adjustment strategy is performed to obtain a corrected scanned reconstructed image, including:

[0158] According to different types of reconstructed image errors, corresponding correction and adjustment strategies are adopted, specifically:

[0159] If the type of reconstructed image error is geometric distortion, use the control point method or image registration technology to correct it; select the same control points in the original image and the target image; use the correspondence between the control points and use the least squares method to calculate the transformation matrix; use the obtained transformation matrix to perform a geometric transformation on the original image to obtain a corrected image;

[0160] If the type of error in the reconstructed image is uneven brightness, gamma correction is used to improve the brightness contrast and adjust the image intensity; the optimal gamma value is selected through cross-validation according to the specific uneven brightness of the image; the gamma correction formula is applied to each pixel value in the image to calculate a new intensity value; the output value is limited to the range of 0 to 1 (or 0 to 255) to ensure that the image is displayed normally; the corrected intensity value replaces the corresponding pixel in the original image to generate a corrected image;

[0161] If the type of reconstructed image error is noise disturbance, wavelet transform is used to suppress random noise in the image; the noisy image is subjected to wavelet transform to decompose the image into sub-bands of different scales and directions; soft threshold is used to process the coefficients of each sub-band after wavelet decomposition. The specific formula is:

[0162] Among them, w C′ is the wavelet coefficient; T is the threshold; sgn(w C′ ) is the wavelet coefficient w C′ The sign function of ; C′ is the index of the wavelet coefficient;

[0163] Perform inverse wavelet transform on the wavelet coefficients after threshold processing to reconstruct the denoised image;

[0164] If the type of reconstructed image error is an artifact problem, a de-artifacting algorithm is used to correct the artifacts caused by the reconstruction algorithm and equipment limitations;

[0165] If the type of reconstructed image error is insufficient resolution, super-resolution reconstruction technology is used to improve the quality and details of the image; the low-resolution image is Fourier transformed, and then a high-resolution image is reconstructed in the spatial domain through an optimization algorithm;

[0166] If the type of reconstructed image error is reconstruction algorithm error, an optimized interpolation method is used to adaptively adjust the reconstruction algorithm.

[0167] Methods for reconstructing a corrected image model through an image reconstruction algorithm include:

[0168] Reconstructing the rectified image by a bilinear interpolation algorithm, generating an image model from the rectified image; calculating a new value at each pixel position of the image based on the values ​​of four pixels around the image by a bilinear interpolation formula;

[0169] The bilinear interpolation formula is:

[0170] O(p′,q′)=(1-A′)(1-B′)O 11 +A′(1-B′)O 21 +(1-A′)B′O 12 +A′B′O22 ; Wherein, O(p′,q′) is the pixel value of the interpolated image at the position (p′,q′); O 11 , O 21 , O 12 and O 22 are the values ​​of the four neighboring pixels surrounding the position point (p′, q′); A′ and B′ are the relative positions of the position point (p′, q′) in the neighboring pixels.

[0171] The preset scanning and reconstructing image error threshold is set by a staff member, and different scanning and reconstructing image errors are collected through an image scanning terminal, and an average value of multiple scanning and reconstructing image errors is taken as the preset scanning and reconstructing image error threshold.

[0172] In this embodiment, by calculating the Kendall rank correlation coefficient between features, features with too strong or too weak correlation can be selectively removed; the dimension of the data is effectively reduced, the risk of overfitting is reduced, and the generalization ability and computational efficiency of the machine learning model are improved; by limiting the scaling factor, the problem of excessive fluctuation of the result due to too large a coefficient or insufficient model adjustment due to too small a coefficient can be avoided; appropriate restrictions can make the model more stable in data preprocessing and reduce unnecessary deviations and noise;

[0173] By smoothing the image with the Gaussian kernel function, the noise in the image can be effectively removed and important image information can be retained; the random noise in the image can be reduced, and the global information structure of the image can be retained while smoothing the image; the degree of image smoothing can be controlled by adjusting the standard deviation of the Gaussian function. The larger the standard deviation, the stronger the smoothing effect of the filter; by dynamically adjusting the window size of the filter, it can adapt to the needs of different images; through reasonable image preprocessing, the noise and redundant information in the image are reduced, and more important features are extracted, thereby reducing the complexity of model calculation, improving calculation efficiency, and reducing system resource consumption.

[0174] Embodiment 2

[0175] See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, and a method for dynamic error correction for scanning image reconstruction is provided, comprising:

[0176] S1, collecting imaging equipment parameter data, scanning image parameter data and environmental parameter data;

[0177] S2, preprocessing the acquired imaging device parameter data, scanning image parameter data and environmental parameter data to obtain a device feature data set, an image feature data set and an environmental feature data set; analyzing the influence of the environmental feature data set on the device feature data set and the image feature data set to obtain an influence feature data set;

[0178] S3, constructing an error prediction model, inputting the influencing feature data set into the error prediction model, and predicting the scanning and reconstructed image error;

[0179] S4, comparing the predicted scan reconstructed image error with a preset scan reconstructed image error threshold to determine whether correction is required;

[0180] S5. If no correction is required, the scanned reconstructed image is directly output; if correction is required, the scanned reconstructed image error data is collected, an error type diagnosis model is constructed, and the scanned reconstructed image error data is input into the error type diagnosis model to predict the reconstructed image error type;

[0181] S6, performing corresponding correction adjustment strategies according to the error type of the reconstructed image to obtain a corrected scanned reconstructed image, and reconstructing a corrected image model through an image reconstruction algorithm;

[0182] Since the electronic device introduced in this embodiment is an electronic device used to implement the error dynamic correction system for scanning image reconstruction in the embodiment of this application, based on the error dynamic correction system for scanning image reconstruction introduced in the embodiment of this application, the technical personnel of this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not described in detail here. As long as the technical personnel of this field implement the electronic device used in the error dynamic correction system for scanning image reconstruction in the embodiment of this application, it belongs to the scope of protection of this application.

[0183] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0184] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A dynamic error correction system for scanning image reconstruction, characterized in that: include: A data acquisition module, used to acquire imaging device parameter data, scanned image parameter data and environmental parameter data; The data processing module is used to pre-process the acquired imaging device parameter data, scanning image parameter data and environmental parameter data to obtain a device feature data set, an image feature data set and an environmental feature data set; analyze the influence of the environmental feature data set on the device feature data set and the image feature data set to obtain an influence feature data set; For imaging device parameter data and environmental parameter data, feature extraction is performed by calculating the Kendall rank correlation coefficient, thereby obtaining a device feature data set and an environmental feature data set; by presetting observation samples, each observation sample corresponds to an imaging device parameter data and an environmental parameter data, and each data in the imaging device parameter data and the environmental parameter data is integrated and represented as a matrix; the number of consistent pairs and the number of inconsistent pairs between any two sample pairs are calculated; based on the number of consistent pairs and the number of inconsistent pairs, the Kendall rank correlation coefficient between two features in any observation sample is obtained, and feature extraction is performed according to the Kendall rank correlation coefficient value between two features in any observation sample, thereby obtaining a device feature data set and an environmental feature data set; The error prediction module is used to construct an error prediction model, input the influencing feature data set into the error prediction model, and predict the error of the scanned and reconstructed image; An adjustment evaluation module is used to compare the predicted scan reconstructed image error with a preset scan reconstructed image error threshold to determine whether correction is required; The error diagnosis module directly outputs the scanned reconstructed image if no correction is required; if correction is required, the error data of the scanned reconstructed image is collected, an error type diagnosis model is constructed, and the error data of the scanned reconstructed image is input into the error type diagnosis model to predict the error type of the reconstructed image; A dynamic correction module is used to perform a corresponding correction adjustment strategy according to the type of reconstructed image error, obtain a corrected scanned reconstructed image, and reconstruct a corrected image model through an image reconstruction algorithm. Each module is connected to another through a wired and / or wireless method; The characteristic categories of the imaging device parameter data include the scanning angle of the device, the scanning speed of the device, the intensity of the radiation emitted by the device, the exposure duration of the device scanning and the temperature of the scanning device; the characteristic categories of the scanning image parameter data include the brightness of the image, the spatial resolution of the image, the contrast of the image, the signal-to-noise ratio of the image and the uniformity of the image; the characteristic categories of the environmental parameter data include the temperature of the imaging environment, the humidity of the imaging environment and the light intensity of the scanning environment.

2. The error dynamic correction system for scanning image reconstruction according to claim 1, characterized in that: The method for acquiring the device feature data set and the environment feature data set includes: S31. m observation samples are preset. For each observation sample, there is an imaging device parameter data and an environmental parameter data. In the imaging device parameter data, the scanning angle of the device is recorded as X1, the scanning speed of the device is recorded as X2, the intensity of the radiation emitted by the device is recorded as X3, the exposure duration of the device scanning is recorded as X4, and the temperature of the scanning device is recorded as X5. In the environmental parameter data, the temperature of the imaging environment is recorded as X6, the humidity of the imaging environment is recorded as X7, and the light intensity of the scanning environment is recorded as X8. S32, integrating each data in the imaging device parameter data and the environmental parameter data into an m×8 matrix D, where each row of the matrix represents a sample, and there are m rows in total; each column of the matrix represents a feature in the imaging device parameter data and the environmental parameter data, and there are 8 columns in total; The matrix D is: Among them, X e′c′ is the value of the e′th sample on the c′th feature, e′=1,...,m, c′=1,...,8; S33, calculating the number of consistent pairs between any two sample pairs; for any two sample pairs, judging whether the two sample pairs are consistent pairs by using a consistency judgment formula and an inconsistency judgment formula, the consistency judgment formula is: Among them, X i′ is the i′th sample value of variable X in the imaging device parameter data and the environmental parameter data; X j′ is the j′th sample value of variable X in the imaging device parameter data and environmental parameter data; Y i′ is the i′th sample value of variable Y in the imaging device parameter data and the environmental parameter data; Y j′ is the j′th sample value of variable Y in the imaging device parameter data and the environmental parameter data; I[] is an indicator function, when the condition in the brackets is met, the result is 1, otherwise it is 0; I[(X i′ -X j′ )(Y i′ -Y j′ )>0] is an indicator function used to judge two sample pairs (X i′ ,Y i′ ) and (X j′ ,Y j′ ) is a consistent pair; i′ and j′ are the indexes of the sample values; If (X i′ -X j′ )>0 and (Y i′ -Y j′ )>0, or (X i′ -X j′ )<0 and (Y i′ -Y j′ )<0, then this pair is a consistent pair and returns 1; otherwise, returns 0; by summing all possible sample pairs, calculate the number C of all consistent pairs; The inconsistency judgment formula is: Among them, I[(X i′ -X j′ )(Y i′ -Y j′ )<0] is an indicator function used to judge the two sample pairs (x i ,y i ) and (x j ,y j ) is an inconsistent pair; If (X i′ -X j′ )>0 and (Y i′ -Y j′ )<0, or (X i′ -X j′ )<0 and (Y i′ -Y j′ )>0, then this pair is an inconsistent pair and returns 1; otherwise, returns 0; by summing all possible sample pairs, calculate the number D of all inconsistent pairs; S34, given imaging device parameter data and environmental parameter data, two features X in any observation sample Z and X f , calculate X Z and X f The Kendall rank correlation coefficient between them is calculated as: Where τ is X Z and X f The Kendall rank correlation coefficient between them; C is the number of consistent pairs between any two sample pairs; D is the number of inconsistent pairs between any two sample pairs; n is the number of samples; δ is the scaling factor for adjusting the magnitude of the Kendall rank correlation coefficient τ; z and f are the indexes of the sample values; The scaling factor δ for adjusting the magnitude of the Kendall rank correlation coefficient τ is limited by a scaling factor limiting formula, and the scaling factor limiting formula is: Among them, α is a positive constant, representing the maximum value of the scaling factor; β is a positive constant used to control the minimum value of the scaling factor; l is the number of features; k is the type of feature; S35. The value range of τ is between [-1,1]; a positive τ value indicates positive correlation, a negative τ value indicates negative correlation, and 0 indicates no correlation; when the |τ| value is close to 1, it indicates that there is a strong correlation between the two features; when it is close to 0, it indicates that there is a weak correlation between the two features; according to the calculated τ value, select and remove the features of |τ|∈[h′,1]∪[0,λ′] to obtain the device feature data set and the environment feature data set; where h′ is the minimum strong correlation threshold; λ′ is the maximum weak correlation threshold.

3. The error dynamic correction system for scanning image reconstruction according to claim 2, characterized in that: The method for acquiring the image feature data set includes: Smoothing the scanned image parameter data, enhancing the contrast and extracting the image edge, and then obtaining the image feature data set; S41, the smoothing method is: smoothing the image by using a Gaussian filter kernel function; the Gaussian filter kernel function is: Among them, I′(a,b) is the pixel value of the original image; (a,b) is the horizontal and vertical coordinates of the image; I″(a,b) is the pixel value of the image after Gaussian filtering; G σ (a, b) is the Gaussian kernel function; σ is the standard deviation of the Gaussian function, which controls the smoothness of the filter; g is the size of the filter window; p is the horizontal coordinate of the center point of the filter; q is the vertical coordinate of the center point of the filter; The standard deviation σ of the Gaussian function is adjusted by a smoothness adjustment formula, and the smoothness adjustment formula is: Where H is the height of the image; W is the width of the image; The size g of the filter window is limited by the filter window radius limiting formula, and the filter window radius limiting formula is: Among them, s is the limiting factor that controls the speed of window size change; S42, the method for enhancing contrast is: enhancing the contrast in the image by using a histogram equalization formula; the histogram equalization formula is: Among them, I * (a, b) are the image pixel values ​​after histogram equalization; h(d) is the histogram of gray level d in the image; L′ is the maximum gray level; is the floor rounding function; S43, the method for extracting the edge of the image is: calculating the second-order derivative of the image by using the Laplace operator to enhance the edge in the image, using the bicubic interpolation method to improve the resolution of the image, and using the Sobel algorithm to calculate the gradient of the image, and detecting and extracting the edge of the image by calculating the modulus of the gradient to obtain an image feature data set; Calculate the gradient of an image in the horizontal direction: Among them, G a is the horizontal gradient of the image; Calculate the gradient of an image in the vertical direction: Among them, G b is the gradient of the image in the vertical direction; Calculate the magnitude of the gradient: Among them, G is the modulus of the gradient; The acquired device feature dataset, image feature dataset and environment feature dataset are processed by outlier elimination and standard deviation normalization, and converted into standard normal distribution to obtain normalized device feature dataset, image feature dataset and environment feature dataset.

4. The error dynamic correction system for scanning image reconstruction according to claim 3, characterized in that: The method of analyzing the influence of the environment feature data set on the device feature data set and the image feature data set to obtain the influenced feature data set includes: Construct a multivariate regression model to quantify the comprehensive impact of the environmental feature data set on the accuracy of the device feature data set and the image feature data set, and obtain the influencing feature data; respectively establish a device feature regression model and an image feature regression model; The equipment feature regression model is: Y de =ε0+ε1X en1 +…+ε m X enm +…+ε v X env +∈ de ; Among them, Y de is the device feature dataset; X env is the vth environmental feature; ∈ de is the error term of the model; ε0 is the constant term of the regression model; ε1 is the regression coefficient of the first environmental characteristic; ε m is the regression coefficient of the mth environmental characteristic; ε v is the regression coefficient of the vth environmental characteristic; The image feature regression model is: Y ige =ε0+ε1X en1 +ε2X en2 +…+e v X env +∈ ige Among them, Y ige is the image feature dataset; ∈ ige is the error term of the model.

5. The error dynamic correction system for scanning image reconstruction according to claim 4, characterized in that: The method for constructing the error prediction model includes: The historical impact feature dataset and the corresponding scanned reconstructed image errors are used as sample sets; the sample sets are used as datasets, and the datasets are divided into training sets, validation sets, and test sets; an error prediction model is constructed, and the error prediction model includes input data and output labels; The training set is used to train the parameters of the machine learning model, the validation set is used to monitor the model performance during the training process and adjust the model hyperparameters, and the test set is used to finally evaluate the performance of the trained model; The input data of the model is a historical impact feature data set, and the output label of the model is a scanned reconstructed image error; RBF is selected as the kernel function of the model, and the hyperparameters of the model are initialized. The hyperparameters of the model include a regularization parameter C and a γ value of the RBF kernel function; the error prediction model is a support vector machine model; Define the loss function, and use the L2 regularized mean square error loss function to measure the difference between the output scanned reconstructed image error and the actual scanned reconstructed image error; the L2 regularized mean square error loss function is: Where n' is the sample size, is the predicted value of the b′th sample; y b′ is the actual value of the b′th sample, λ is the regularization coefficient; N is the number of training samples; ∑ w w 2 is the sum of squares of all trainable weight parameters; w is the model weight parameter; Use the training set data to train the model and minimize the loss function using the Adam optimizer. Use the validation set to evaluate the performance of the model and tune the model's hyperparameters until the model performance no longer improves significantly or reaches the preset stopping condition. The test set is used to evaluate the performance of the model in the prediction task, and the current influencing feature data set is input into the trained error prediction model to obtain the scanned reconstructed image error.

6. The error dynamic correction system for scanning image reconstruction according to claim 5, characterized in that: The method of comparing the predicted scan reconstructed image error with a preset scan reconstructed image error threshold to determine whether correction is required includes: If the predicted scan reconstructed image error is greater than or equal to a preset scan reconstructed image error threshold, it is determined that correction is required; If the predicted scan reconstructed image error is less than a preset scan reconstructed image error threshold, it is determined that no correction is required.

7. The error dynamic correction system for scanning image reconstruction according to claim 6, characterized in that: The method of constructing the error type diagnosis model, inputting the scanned reconstructed image error data into the error type diagnosis model, and predicting the reconstructed image error type includes: The scanned reconstructed image error data includes the pixel value of the image, the structural similarity measurement of the image, the signal-to-noise ratio of the image and the spatial resolution of the image; the structural similarity measurement of the image includes the brightness of the image, the contrast of the image and the structural similarity of the image; The data set is divided into a training set, a validation set, and a test set; the data set includes historical scanned reconstructed image error data and the corresponding reconstructed image error types; an error type diagnosis model is constructed, and the error type diagnosis model includes an input layer, a convolution layer, a pooling layer, a batch normalization layer, a fully connected layer, and an output layer; The input layer of the model is used to input historical scan reconstruction image error data, and the output layer of the model is used to output the reconstruction image error type; softmax is used as the activation function; the error type diagnosis model is a 3D convolutional neural network model; Use multi-classification cross entropy as the loss function of the model to measure the difference between the model's predicted value and the actual value; Use the training set data to train the model and minimize the loss function by selecting the SGD optimizer; use the validation set to evaluate the performance of the error type diagnosis model, calculate the accuracy index to measure the performance of the model, and tune the model's hyperparameters to improve the model performance; The performance of the model in the prediction task is evaluated through the test set. When the performance of the model in the prediction task reaches the preset performance threshold, the test is stopped to obtain the error type diagnosis model. The trained error type diagnosis model is used to predict the error data of the current scanned reconstructed image to obtain the reconstructed image error type.

8. The error dynamic correction system for scanning image reconstruction according to claim 7, characterized in that: The method of performing a corresponding correction adjustment strategy according to the reconstructed image error type to obtain a corrected scanned reconstructed image includes: According to different types of reconstructed image errors, corresponding correction and adjustment strategies are adopted, specifically: If the type of reconstructed image error is geometric distortion, use the control point method or image registration technology to correct it; select the same control points in the original image and the target image; use the correspondence between the control points and use the least squares method to calculate the transformation matrix; use the obtained transformation matrix to perform a geometric transformation on the original image to obtain a corrected image; If the type of reconstructed image error is uneven brightness, gamma correction is used to improve brightness contrast and adjust image intensity; the best gamma value is selected through cross-validation according to the specific uneven brightness of the image; the gamma correction formula is applied to each pixel value in the image to calculate a new intensity value; the corrected intensity value is used to replace the corresponding pixel in the original image to generate a corrected image; If the type of reconstructed image error is noise disturbance, wavelet transform is used to suppress random noise in the image; the noisy image is subjected to wavelet transform to decompose the image into sub-bands of different scales and directions; soft threshold is used to process the coefficients of each sub-band after wavelet decomposition. The specific formula is: Among them, w C′ is the wavelet coefficient; T is the threshold; sgn(w C′ ) is the wavelet coefficient w C′ The sign function of ; C′ is the index of the wavelet coefficient; Perform inverse wavelet transform on the wavelet coefficients after threshold processing to reconstruct the denoised image; If the type of reconstructed image error is an artifact problem, a de-artifacting algorithm is used to correct the artifacts caused by the reconstruction algorithm and equipment limitations; If the type of reconstructed image error is insufficient resolution, super-resolution reconstruction technology is used to improve the quality and details of the image; the low-resolution image is Fourier transformed, and then a high-resolution image is reconstructed in the spatial domain through an optimization algorithm; If the type of reconstructed image error is reconstruction algorithm error, an optimized interpolation method is used to adaptively adjust the reconstruction algorithm.

9. The error dynamic correction system for scanning image reconstruction according to claim 8, characterized in that: The method for reconstructing the corrected image model through an image reconstruction algorithm comprises: Reconstructing the rectified image by a bilinear interpolation algorithm, generating an image model from the rectified image; calculating a new value at each pixel position of the image based on the values ​​of four pixels around the image by a bilinear interpolation formula; The bilinear interpolation formula is: O(p′,q′)=(1-A′)(1-B′(O 11 +A′(1-B′(O 21 +(1-A′)B′O 12 +A′B′O 22 ; Wherein, O(p′,q′) is the pixel value of the interpolated image at the position (p′,q′); O 11 , O 21 , O 12 and O 22 are the values ​​of the four neighboring pixels surrounding the position point (p′, q′); A′ and B′ are the relative positions of the position point (p′, q′) in the neighboring pixels.

Images