CT image hybrid filtering optimization method
By combining projection domain filtering and image domain wavelet transform in a hybrid filtering optimization method, the problems of artifact suppression and detail enhancement in turbomachinery inspection were solved, and high-precision CT image reconstruction was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-12
- Publication Date
- 2026-03-17
AI Technical Summary
Existing CT image processing technology cannot effectively suppress stripe artifacts in the inspection of turbomachinery, resulting in image blurring and loss of details, making it difficult to meet the requirements of high-precision non-destructive testing.
A hybrid filtering optimization method is adopted, which combines adaptive filtering in the projection domain and wavelet transform in the image domain to suppress artifacts and enhance details for the specific structure of turbomachinery. This method includes techniques such as Hough circle detection, polar coordinate transformation, ramp filter, and wavelet enhancement.
It significantly improves artifact removal and enhances image detail resolution, especially the contrast of the blade tip-casing gap, meeting the requirements of high-precision non-destructive testing.
Smart Images
Figure CN121685271A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of nondestructive testing technology and relates to a method for optimizing hybrid filtering of CT images. Background Technology
[0002] In aerospace, energy, and shipbuilding industries, turbomachinery serves as a core component of energy conversion and fluid transport systems. The structural integrity and accurate characterization of internal defects directly determine the overall performance, efficiency, and service life of the machine. High-energy X-ray computed tomography (CT) technology, as an important means of non-destructive testing of dense materials and large-sized components, can be used for non-destructive testing of high-end turbomachinery equipment such as aero-engines and gas turbines, which is crucial for ensuring the safety and reliability of such equipment in service.
[0003] High-precision inspection of the internal microstructures of equipment requires high-quality CT reconstructed images. However, due to the complex mechanical structure of impellers, their core components (such as the impeller disk, blades, and casing) are made of high-density metal alloys with large radial dimensions (ranging from 350 mm to 1300 mm). According to Beer-Lambert's law, the photon flux of X-rays decays exponentially after penetrating an object. When the penetration path length (i.e., along the impeller diameter) is extremely long and the material's linear attenuation coefficient is extremely high, even with megavolt-level X-ray sources, the number of photons after penetration is still very small, resulting in a much lower number of photons received by the detector than normal. Simultaneously, scattering causes severe striped artifacts in the images reconstructed using conventional filtered back projection (FBP) algorithms. These artifacts lead to blurred edges and loss of detail, making it difficult to meet the requirements for high-precision inspection of the internal microstructures of samples.
[0004] To address this issue, existing image processing techniques can be divided into two categories: those based on the projection domain and those based on the image domain.
[0005] Projection domain-based techniques involve processing the acquired projection data before image reconstruction. For example, adaptive filtering can be applied to the projection data, and outdated projection data can be processed before backprojection. Alternatively, the preprocessed projection data can be combined with an iterative reconstruction algorithm to correct the statistical model in the original algorithm. The underlying idea of these methods is to suppress noise and artifacts from the source of the original data.
[0006] Image domain-based techniques: After obtaining the initial reconstructed image through algorithms such as Filtered Back Projection (FBP), the image itself is post-processed. For example, various spatial or frequency domain filters (such as Gaussian filtering, median filtering, and wavelet denoising) are used to smooth the image.
[0007] The disadvantages of existing technologies are as follows.
[0008] Incomplete artifact removal: Existing methods lack specificity for artifacts caused by the unique structure of impeller machinery (such as the thick central disc), and cannot fundamentally and effectively suppress the stripe artifacts caused by this area. Furthermore, as the size of the component increases, the improvement effect of existing methods is further limited.
[0009] Impaired detail resolution: In the process of filtering or iteration, strong smoothing constraints are often used to suppress artifacts and noise, which can lead to the loss or blurring of high-frequency detail information in the image. Summary of the Invention
[0010] To address the problems existing in the prior art, the present invention aims to provide a CT image hybrid filtering optimization method. This invention targets the specific structure of turbomachinery, aiming to effectively suppress artifacts while enhancing the detailed features of key areas such as the blade tip and casing of turbomachinery samples, improving their contrast, and thus enhancing their resolvability, thereby improving image quality to meet the requirements of high-precision non-destructive testing.
[0011] The technical solution of this invention is as follows: A method for optimizing CT image hybrid filtering, comprising the following steps: 1) The initial CT image of the sample is obtained by reconstructing the projection data of the sample using a filtered back projection algorithm; 2) The initial CT image is inspected to locate the position of the target structure in the sample within the initial CT image. The projection data corresponding to this position is binarized and then orthographically projected to determine the position of the ray passing through the target structure in the projection data. Then, the rays passing through the target structure are filtered using filter A, while other rays not passing through the target structure are filtered using filter B, resulting in a CT reconstructed image after artifact removal. The form of filter A is as follows: The filter B is in the form of Among them, the forms of filter A and filter B are derived from the iterative reconstruction algorithm. For a ramp filter, k is the number of iterations. The iteration step size, The coefficient of the regularization term; 3) Perform coordinate transformation on the CT reconstructed image, transforming it from rectangular coordinates to polar coordinates to obtain the image in polar coordinates; 4) Perform wavelet transform on the image in polar coordinates to decompose it into low-frequency components, vertical high-frequency components, horizontal high-frequency components, and diagonal high-frequency components at multiple scales; then multiply the vertical high-frequency components of each layer after wavelet decomposition by an enhancement coefficient to enhance the edge features in the image in polar coordinates, and then perform inverse wavelet transform. 5) Transform the image processed in step 4) back to rectangular coordinates to obtain the final optimized sample CT image.
[0012] Preferably, the iterative reconstruction algorithm is an iterative algorithm that incorporates prior knowledge, and its objective function is in the form of: The objective function is solved using the Landweber iterative algorithm, which incorporates a regularization term. , representing the ray projection through the target structure, through which parameters are used to further suppress artifacts. Used to control and weaken the projected signal strength of the target structural region. It is the image vector to be reconstructed. It is the projection vector. It is a system matrix. The system matrix A represents the image vector to be reconstructed. Orthographic projection operation, It is the projection difference; It is a submatrix of the system matrix A, containing only the rows corresponding to the rays passing through the target structure; It is an orthographic projection operation performed on the vector of the target structural region of the image to be reconstructed.
[0013] Preferably, filters A and B employ window functions with added noise weights for ray-by-ray filtering, and then back-project the ray-by-ray filtered projection data to obtain the artifact-free CT reconstruction image; wherein, the noise weight of the ray... , where n is the index of the ray in the projection data. This represents the maximum value of the projection after logarithmic transformation.
[0014] Preferably, the final form of filter A is: The final form of filter B is .
[0015] Preferably, the index n corresponds to the projection value of each ray after logarithmic processing.
[0016] Preferably, the filter in the filtering back projection algorithm is a ramp filter.
[0017] Preferably, the vertical high-frequency components of each layer after wavelet decomposition are multiplied by an enhancement coefficient greater than 1.
[0018] Preferably, the location of the target structure in the initial CT image is found using Hough circle detection or least squares circle fitting.
[0019] Preferably, the projection data of the sample is obtained by using a linear array scanning method with a high-energy X-ray CT system.
[0020] The present invention has the following characteristics.
[0021] 1. Classification mechanism of projection data. Based on the Hough circle detection results of the initial reconstructed image, a technical solution is used to distinguish the original projection data into two categories of rays: "passing through the turntable" and "not passing through the turntable" through orthographic projection.
[0022] 2. Differentiated filter design for the two types of rays. In particular, the filter form derived from an iterative algorithm that introduces a specific regularization term, and its combination with noise weights, is used for the "through-the-rotor" ray.
[0023] 3. Strategy combining polar coordinate transformation and wavelet enhancement. To achieve directional enhancement of the blade tip-casing clearance, a complete technical path is adopted: "rectangular coordinates → polar coordinates → wavelet transform → enhancement of specific components → inverse wavelet transform → inverse polar coordinate transform".
[0024] 4. Selective enhancement of vertical high-frequency components. In polar coordinate space, the vertical high-frequency components (HL) of wavelet decomposition are specifically enhanced to achieve targeted enhancement of the tip clearance characteristics, rather than global enhancement.
[0025] 5. It can be applied to CT non-destructive testing of turbomachinery (such as aero engines and gas turbines).
[0026] The advantages of this invention are as follows: This invention offers superior artifact removal performance. The method improves upon existing noise-weighted filtering back-projection algorithms, fully considering the difficulty of X-rays penetrating the central rotating disk region of turbomachinery samples, and designs a more targeted filtering strategy, thereby achieving more effective artifact suppression.
[0027] Meanwhile, this invention significantly improves the resolution of minute features in images. Targeting the geometric distribution characteristics of a key minute feature—the gap between the impeller blade tip and the casing—this technology employs a series of processes, including Hough circle detection, coordinate transformation, and wavelet transform, to specifically enhance this feature, effectively improving the contrast of the gap region. Furthermore, the invention has a clear process and, in addition to being applicable to typical impeller machinery such as aero-engines, can also be extended to other industrial components with similar structures, possessing broad practical application prospects. Attached Figure Description
[0028] Figure 1 This is a flowchart of the method of the present invention.
[0029] Figure 2(a) shows the result of reconstructing the simulation data of the turbomachinery components using the filtered back projection (FBP) algorithm.
[0030] Figure 2(b) is an enlarged image of the feature of interest (one of the impeller blades and the casing).
[0031] Figure 3(a) shows the CT image of the final optimized turbomachinery simulation sample.
[0032] Figure 3(b) is a magnified image of the corresponding feature of interest (one of the impeller blades and the casing). Detailed Implementation
[0033] The present invention will now be described in further detail with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.
[0034] The core of this invention lies in combining adaptive filtering in the projection domain with wavelet transform enhancement in the image domain. The first stage performs targeted artifact removal in the projection domain, while the second stage enhances details in the image domain. The flowchart of the overall technical solution is as follows: Figure 1 As shown, the detailed steps are as follows.
[0035] Step 1: Data Acquisition and Initial Reconstruction. Projection data of the sample was obtained by scanning it with a high-energy X-ray CT system, using a linear array scan to reduce the influence of scattered radiation. The projection data was reconstructed using a filtered back projection (FBP) algorithm. The filter used in the FBP algorithm was a Ram-Lak filter (i.e., a finite-bandwidth ramp filter), obtaining the initial CT image of the sample, which served as the basis for subsequent processing. Simulation data was used in this example to verify the effectiveness of the invention. Figure 2(a) shows the result of reconstructing the impeller mechanical component using the filtered back projection (FBP) algorithm. The casing material is iron, and the impeller blade material is nickel alloy. Scattering and noise were added to the projection. Imaging was performed using a 6MeV wide-spectrum simulated X-ray source. Figure 2(b) is an enlarged image of the feature of interest (one of the impeller blades and the casing).
[0036] Step 2: Projection Data Classification. Taking the central rotating disk of the impeller as the target structure, Hough circle detection is performed on the initial CT image in Figure 2(a) to find the position of the central rotating disk structure in the initial CT image. The result of the circle detection is binarized and then orthographically projected to determine the position of the rays passing through the disk in the projection data. The projection data is classified according to whether the rays pass through the disk. Rays that pass through the disk are filtered by filter A, and other rays that do not pass through the disk are filtered by filter B. Filter A is used to reduce the weight of the projection data of the rotating disk structure area in the reconstruction process, because rays are difficult to penetrate in this area, the noise is high, and the proportion of scattered photons is high. Reducing its weight in the reconstruction process helps to reduce image artifacts. Filter B is equivalent to the Ram-Lak filter, which helps to preserve image edge and detail information. The forms of filter A and filter B are used to derive the self-iterative reconstruction algorithm. The objective function form of a common iterative algorithm that incorporates prior knowledge (MAP) is: ;in It is the image vector to be reconstructed. It is the projection vector. It is a system matrix. The system matrix A represents the image vector to be reconstructed. Perform orthographic projection. It is the projection difference.
[0037] This method changes the form of its objective function to: Regularization terms were introduced. , It is a submatrix of the system matrix A, containing only the rows corresponding to the rays that pass through the target structure (impeller disk). This involves orthographically projecting the vector of the target structure (impeller disk) region in the image to be reconstructed. This term represents the projection of the impeller disk region. Artifacts are further suppressed by weakening the projection signal intensity of the impeller disk region. This is used to control the intensity of the weakening. The objective function can be solved using the Landweber iterative algorithm, and has the following form: Expanding it into a non-recursive form is as follows: Given the projection and backprojection operations on the image (i.e., multiplication by) ), equivalent to using in the frequency domain Filtering is then performed. Therefore, the forms of the two types of filters in step 2 can be derived as follows: For the ray passing through the impeller disk, the form of filter A is: For other rays, the filter type B is as follows: in, For ramp filters, the analytical forms of filters A and B mentioned above are derived from the iterative reconstruction algorithm, and their parameters also originate from iterative reconstruction. In the embodiment described, k is the number of iterations, which is set to 8000. The iteration step size is set to 0.00153. This is the regularization coefficient, which can be selected based on the sample size and the density of the constituent materials, or based on the degree of artifacts appearing in the initial reconstructed image. In this example, the parameter... A value of 0.03 is needed to achieve a better artifact removal effect.
[0038] Step 3: Based on Step 2, the statistical model of the noise is added as a weighting function to the filtering back-projection algorithm. A ray-by-ray filtering is performed on both types of filters A and B using a window function with noise weights. The noise weight for each ray is defined as: The index n of each ray can be 0, 1, 2, ..., N, depending on the magnitude of the projection value of each ray after logarithmic processing. This represents the maximum value of the projection after logarithmic transformation. This represents the projection value of each ray after logarithmic transformation, which involves taking the negative logarithm of the original projection data. In this implementation example, N is set to 6. The logarithmic transformation steps are as follows: The original projection value, This represents the original projection value of the background area.
[0039] Combining step 2, the final filter A used for the ray passing through the turntable is of the following form: The final filter form B used for other rays: After filtering the projection data ray by ray, backprojection yields the artifact-free CT reconstructed image. At this point, the stripe artifacts in the image have been significantly suppressed.
[0040] Step 4: Perform coordinate transformation on the artifact-removed CT reconstructed image, converting it from rectangular coordinates to polar coordinates. The center of the impeller disk is used as the center of the coordinate transformation circle, and the coordinates of this center are obtained by performing Hough circle detection on the initial reconstructed image. Cubic polynomial interpolation is used for the rectangular-to-polar coordinate transformation. After this transformation, each blade of the impeller and the blade tip-casing clearance change from their original curved shape to an approximately horizontal straight line, greatly facilitating subsequent targeted enhancement.
[0041] Step 5: Perform wavelet transform on the image in polar coordinates, decomposing it into low-frequency components (LL), vertical high-frequency components (LH), horizontal high-frequency components (HL), and diagonal high-frequency components (HH) at multiple scales. The Symlet wavelet basis function is used for the wavelet transform, and the decomposition level is 4. After the polar coordinate transform, the feature of the impeller tip-casing gap becomes a horizontal feature. Therefore, by multiplying the vertical high-frequency components of each wavelet decomposition level by an enhancement coefficient greater than 1, the feature of the impeller tip-casing gap (i.e., the edge feature of the sample) in the image can be enhanced. In this example, the enhancement coefficient is 1.25. Then, perform an inverse wavelet transform.
[0042] Step 6: Transform the image processed in Step 5 back to Cartesian coordinates. The center of the circle during coordinate transformation is the same as the center of the circle in Step 4, and the interpolation method used is cubic polynomial interpolation. The final optimized CT image of the impeller machinery simulation sample is shown in Figure 3(a), and Figure 3(b) is the corresponding magnified image of the feature of interest (one of the impeller blades and the casing).
[0043] By quantitatively calculating the contrast at this gap, the contrast ratio was found to be 45.91% before optimization and 77.46% after optimization. The contrast ratio of the gap was significantly improved after optimization. Furthermore, this method is applicable to various types of turbomachinery samples and has broad application prospects in practical applications.
[0044] In turntable detection, in addition to Hough circle detection, other image segmentation algorithms such as least squares circle fitting can also be used to locate the center turntable.
[0045] In wavelet transform, the Symlet wavelet can be replaced by wavelet bases with similar properties, such as the Coiflets wavelet.
[0046] During coordinate transformation, cubic polynomial interpolation can be replaced by linear interpolation or spline interpolation.
[0047] Although specific embodiments of the invention have been disclosed for illustrative purposes to aid in understanding and implementing the invention, those skilled in the art will understand that various substitutions, variations, and modifications are possible without departing from the spirit and scope of the invention and the appended claims. Therefore, the invention should not be limited to the content disclosed in the preferred embodiments, and the scope of protection claimed by the invention is defined by the claims.
Claims
1. A CT image blending filter optimization method comprising the steps of: 1) reconstructing the projection data of a sample using a filtered back-projection algorithm to obtain an initial CT image of the sample; 2) detecting the initial CT image to find the location of the target structure in the sample in the initial CT image, binarizing the projection data corresponding to the location and then orthographically projecting to determine the location of the rays passing through the target structure in the projection data, then filtering the rays passing through the target structure with filter A and filtering the rays not passing through the target structure with filter B to obtain a CT reconstruction image after removing artifacts; the form of the filter A is , the form of the filter B is ; wherein the form of the filter A and the filter B is derived from an iterative reconstruction algorithm, is a ramp filter, k is the number of iterations, is an iteration step, is a regularization term coefficient; 3) performing a coordinate transformation on the CT reconstructed image to transform it from Cartesian coordinates to polar coordinates to obtain an image in polar coordinates; 4) performing a wavelet transform on the image in polar coordinates to decompose it into low frequency components, vertical high frequency components, horizontal high frequency components and diagonal high frequency components at multiple scales; then multiplying the vertical high frequency components at each level of the wavelet decomposition by an enhancement factor to enhance the edge features in the image in polar coordinates, followed by an inverse wavelet transform; 5) transforming the image processed in step 4) back to Cartesian coordinates to obtain a final optimized CT image of the sample.
2. The method of claim 1, wherein, The iterative reconstruction algorithm is an iterative algorithm combined with prior knowledge, and the objective function thereof is in the form of The objective function is solved by a Landweber iterative algorithm; wherein a regularization term is introduced, representing the ray projection passing through the target structure, and by further suppressing artifacts, a parameter is used to control the weakening of the projection signal strength of the target structure region, is a to-be-reconstructed image vector, is a projection vector, is a system matrix, represents the forward projection operation of the to-be-reconstructed image vector on the system matrix A, is a projection difference value; is a sub-matrix of the system matrix A, only containing rows corresponding to rays passing through the target structure; is a forward projection operation on the to-be-reconstructed image target structure region.
3. The method of claim 2, wherein, The filter A and the filter B perform ray-by-ray filtering by using a window function with added noise weight, and then perform back-projection on the ray-by-ray filtered projection data to obtain a CT reconstructed image with removed artifacts; wherein the noise weight of a ray is n is an index of a ray in the projection data, is a maximum value of the projection after logarithmic processing.
4. The method of claim 3, wherein, The final form of filter A is The final form of filter B is .
5. The method of claim 3, wherein, The index n corresponds to the magnitude of the projection value of each ray after logarithmic processing.
6. The method of claim 1, wherein, The filter in the filtered back-projection algorithm is a Ram-Lak filter.
7. The method of claim 1, wherein, The vertical high frequency components at each level of the wavelet decomposition are multiplied by an enhancement factor greater than 1.
8. The method of claim 1, wherein, The location of the target structure in the initial CT image is found using Hough circle detection or least squares circle fitting.
9. The method of claim 1, wherein, The projection data of the sample is obtained by scanning the sample using a high-energy X-ray CT system with a line array. The projection data of the sample is obtained by scanning the sample using a high-energy X-ray CT system with a line array.