An image iterative reconstruction method suitable for a large number of non-penetrating area scenes

Abstract

The present application relates to a kind of image iterative reconstruction methods suitable for a large number of not-through area scene, belong to computer tomography imaging technical field, solve the poor stability of the reconstruction effect in prior art when reconstructing for different to-be-reconstructed workpieces.The iterative reconstruction method includes: segmentation is determined to initial reconstruction image for artifact region, using preset attenuation coefficient replaces the attenuation coefficient value of other pixel points except artifact region in initial reconstruction image to obtain compensation reconstruction image;Determine penetration threshold according to original sinogram;Compensation weight sinogram is generated using penetration threshold, original sinogram and initial reconstruction image, and iteration termination condition is determined based on penetration threshold and original sinogram;Obtain final reconstruction image by several iterations.Improve the stability of the reconstruction image effect and reconstruction process for different to-be-reconstructed workpieces.

Description

Technical Field

[0001] This invention relates to the field of computed tomography imaging technology, and in particular to an iterative image reconstruction method applicable to scenes with a large number of impermeable regions. Background Technology

[0002] Computed tomography (CT) imaging technology has wide applications in industrial inspection, quality assessment, and failure analysis because it can non-destructively visualize the internal three-dimensional structure of a workpiece.

[0003] However, when the workpiece to be inspected contains one or more high-attenuation areas, such as the battery pack which typically contains a variety of high-density components, such as metal connectors, busbars, housings, and tabs, these components have a strong attenuation effect on X-rays, forming a large area of ​​"impenetrable regions" in CT scans.

[0004] On the sine wave composed of projected data, the ray signals passing through these regions are extremely weak, even reaching the noise or saturation limit of the detector. This causes severe distortion or loss of the projected data in these areas. Traditional analytical reconstruction algorithms, such as filtered back-projection algorithms, introduce strong radial stripes, shadows, and "star-shaped" artifacts when directly processing such incomplete sine waves with severe nonlinear distortion. These artifacts not only obscure the true structure of adjacent low-density materials, such as electrode coatings, membranes, or electrolytes, but also lead to serious inaccuracies in the quantitative analysis of attenuation coefficients in critical areas, such as solder joints and interfaces, greatly reducing the interpretability of the image and the reliability of the detection.

[0005] Existing methods typically rely on a globally "preset" penetration threshold to distinguish between impermeable and penetrable areas in the projected data. However, existing battery packs vary widely in model, and the materials, arrangement density, and scanning conditions, such as voltage and current, of their internal components differ greatly. A fixed preset penetration threshold cannot adapt to different scenarios, which can easily lead to "undercorrection" (the threshold is too high, and some bad data is not identified) or "overcorrection" (the threshold is too low, and some valid data is mistakenly deleted), directly affecting the accuracy of all subsequent steps.

[0006] Meanwhile, the existing reconstruction methods are single forward-backward projection processes, lacking a closed-loop iterative optimization mechanism. They cannot adaptively determine the number of iterations and iteration conditions based on the sine curve. A single compensation and reconstruction is insufficient to fully estimate and correct the nonlinear errors caused by complex ray hardening, scattering, and photon starvation effects. For dense and complex impermeable regions inside the battery pack, the correction effect is often incomplete.

[0007] Therefore, there is an urgent need for a technical solution that can adaptively determine iterative reconstruction based on different workpieces to be reconstructed. Summary of the Invention

[0008] Based on the above analysis, the embodiments of the present invention aim to provide an image iterative reconstruction method applicable to scenes with a large number of impermeable areas, in order to solve the problem of poor reconstruction effect stability when reconstructing different workpieces in the prior art.

[0009] This invention provides an iterative image reconstruction method suitable for scenes with a large number of impermeable regions. The iterative reconstruction method includes: The original sinusoidal image of the workpiece to be reconstructed is reconstructed to obtain an initial reconstructed image. The initial reconstructed image is segmented to determine the artifact region. The attenuation coefficient values ​​of other pixels in the initial reconstructed image, excluding the artifact region, are replaced with a preset attenuation coefficient to obtain a compensated reconstructed image. The penetration threshold is determined based on the original sinusoidal image; a compensation weight sinusoidal image is generated using the penetration threshold, the original sinusoidal image, and the initial reconstructed image; and the iteration termination condition is determined based on the penetration threshold and the original sinusoidal image. The final reconstructed image is obtained through several iterations, with the following steps performed in each iteration: The compensated and reconstructed image is orthographically projected to obtain a compensated sinusoidal image. The original sinusoidal image and the compensated sinusoidal image are then fused according to the compensated weight sinusoidal image to obtain a fused sinusoidal image. The fused sinusoidal image is then reconstructed to obtain a fused reconstructed image. Determine whether the iteration termination condition is met; if it is met, use the fused and reconstructed image of the current iteration as the final reconstructed image; otherwise, use the fused and reconstructed image of the current iteration as the compensated reconstructed image of the next iteration and start the next iteration.

[0010] Based on a further improvement to the above iterative reconstruction method, the step of determining the iteration termination condition based on the penetration threshold and the original sinusoidal image includes: The number of pixels in the original sinusoidal image that belong to the impenetrable scene is determined based on the penetration threshold. Determine whether the ratio of the number of pixels that cannot penetrate the scene to the number of pixels in the original sine image exceeds a preset iteration ratio threshold; if it does not exceed the threshold, the iteration termination condition is to perform only one iteration.

[0011] Based on the further improvement of the above iterative reconstruction method, if the ratio of the number of pixels that cannot penetrate the scene to the number of pixels in the original sinusoidal image exceeds a preset iteration ratio threshold, then the iteration termination condition shall be any one of the following termination conditions: The number of iterations exceeds a preset threshold. The difference between the fused reconstructed image and the compensated reconstructed image in the current iteration is less than a preset difference threshold.

[0012] Based on a further improvement to the above iterative reconstruction method, the step of determining the penetration threshold based on the original sinusoidal image includes: The frequency corresponding to each attenuation value in the original sine image is counted. The attenuation value is used as the horizontal axis and the corresponding frequency is used as the vertical axis to obtain the distribution histogram of the original sine image. Find the attenuation value point corresponding to the maximum frequency in the distribution histogram as the global attenuation value point; Starting from the global attenuation value point, the attenuation value at the position corresponding to the preset frequency difference is searched along the direction of increasing attenuation value as the penetration threshold.

[0013] Based on a further improvement to the above iterative reconstruction method, the step of generating a compensated weighted sinusoidal image using a penetration threshold, the original sinusoidal image, and the initial reconstructed image includes: A confidence-based reconstructed image is generated from the initial reconstructed image, and a confidence-based sinusoidal image is obtained by row orthographic projection of the confidence-based reconstructed image. The penetration mask value of the pixel in the original sinusoidal image is determined based on the penetration threshold, and the penetration mask sinusoidal image is obtained. The compensation weight sine image is obtained by multiplying the pixel values ​​of corresponding pixels in the confidence sine image and the penetration mask sine image.

[0014] Based on a further improvement to the above iterative reconstruction method, the step of generating a confidence-based reconstructed image from the initial reconstructed image includes: The gradient magnitude image is obtained by calculating the gradient magnitude of the pixels in the initial reconstructed image. The gradient magnitude of the pixels in the gradient magnitude image is then normalized to obtain the normalized gradient magnitude image. Inverting the pixel values ​​of pixels in the normalized gradient magnitude image yields the intermediate confidence image; Gaussian smoothing and preset range mapping are performed sequentially on the intermediate confidence images to obtain the confidence reconstructed images.

[0015] Based on the further improvement of the above iterative reconstruction method, in each iteration, the pixel value of the pixel in the fused sinusoidal image is calculated using the following formula: ; in, Indicates the detector angle. Indicates the detector channel. This represents the pixel value of a pixel in a fused sine wave image. This represents the pixel value of a pixel in the compensated weight sine wave image. This represents the pixel value of the compensated sine wave image. This represents the pixel value of a pixel in the original sine wave image.

[0016] Based on the further improvement of the above iterative reconstruction method, the step of segmenting the initial reconstructed image to determine the artifact region includes: The initial reconstructed image is divided into boundary region, background region, and interior region; Artifact detection is performed on the internal region to obtain the artifact region and the non-artifact region; Based on multiple pre-set attenuation coefficient features, multiple attenuation coefficient intensity constraints, and fusion conditions, the non-artifact region is segmented to obtain multiple types of regions.

[0017] Based on a further improvement to the above iterative reconstruction method, the non-artifact region is segmented based on multiple pre-set attenuation coefficient features, multiple attenuation coefficient intensity constraints, and fusion conditions to obtain multiple types of regions, including: Multiple attenuation coefficient features are used to classify the pixels in the non-artifact region, resulting in multiple sets of attenuation coefficient features. Multiple attenuation coefficient intensity constraints are used to classify the intensity of pixels in the non-artifact region, resulting in multiple sets of attenuation coefficient intensity constraints. Based on the fusion conditions, multiple sets of attenuation coefficient features and multiple sets of attenuation coefficient intensity constraints are fused to obtain multiple sets of pixels in different types of regions. Morphological correction operations are performed on pixel sets of multiple types of regions to obtain multiple types of regions.

[0018] Based on further improvements to the above iterative reconstruction method, the plurality of attenuation coefficient features include one or more of the following: Gradient and edge features; Projection and boundary characteristics; Geometric positional features; Structural prior features; Confidence characteristics.

[0019] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects: 1. By segmenting the initial reconstructed image of the workpiece to be reconstructed, a compensated reconstructed image is obtained by replacing the attenuation coefficient values ​​of other pixels in the initial reconstructed image except for artifact regions with a preset attenuation coefficient. Furthermore, a penetration threshold is determined based on the original sinusoidal image, which improves the segmentation effect of handling the non-penetrating regions of different workpieces to be reconstructed. A compensated weight sinusoidal image is generated using the penetration threshold, the original sinusoidal image, and the initial reconstructed image. The iteration termination condition is determined based on the penetration threshold and the original sinusoidal image, which adapts to the reconstruction process of different penetration regions, improving processing efficiency and reconstruction effect. In each iteration, the compensated reconstructed image is orthographically projected to obtain a compensated sinusoidal image. The original sinusoidal image and the compensated sinusoidal image are fused based on the compensated weight sinusoidal image to obtain a fused sinusoidal image. The fused sinusoidal image is then reconstructed to obtain the final reconstructed image. Missing data is directly repaired in the sinusoidal domain, avoiding artifact diffusion caused by image domain correction, thus improving the reconstruction effect of the reconstructed image. 2. Based on the prior information of the workpiece to be reconstructed, the initial reconstructed image of the workpiece is divided into boundary region, background region and interior region. Artifact detection is performed on the interior region to obtain artifact region. Pixels in non-artifact region are classified using multiple attenuation coefficient features and multiple attenuation coefficient intensity constraints. Multiple attenuation coefficient feature sets and multiple attenuation coefficient intensity constraint sets are fused based on fusion conditions. Finally, morphological correction operation is used to obtain multiple types of regions, making the initial reconstructed image of the workpiece to be reconstructed more robust and the segmentation effect better.

[0020] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0021] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0022] Figure 1 A flowchart illustrating an image iterative reconstruction method applicable to scenes with a large number of impermeable regions, provided by an embodiment of the present invention; Figure 2 This is a schematic diagram of the process for segmenting and determining artifact regions in an initial reconstructed image according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an exemplary battery pack reconstruction image in an embodiment of the present invention; Figure 4This is a schematic diagram of the structure of an exemplary battery pack model in an embodiment of the present invention. Detailed Implementation

[0023] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0024] A specific embodiment of the present invention discloses an iterative image reconstruction method applicable to scenes with a large number of impermeable regions, such as... Figure 1 As shown, the iterative reconstruction method includes: Step S1: Reconstruct the original sinusoidal image of the workpiece to be reconstructed to obtain an initial reconstructed image. Segment the initial reconstructed image to determine the artifact region. Replace the attenuation coefficient values ​​of other pixels in the initial reconstructed image except for the artifact region with a preset attenuation coefficient to obtain a compensated reconstructed image. Step S2: Determine the penetration threshold based on the original sinusoidal image; generate a compensation weight sinusoidal image using the penetration threshold, the original sinusoidal image, and the initial reconstructed image; and determine the iteration termination condition based on the penetration threshold and the original sinusoidal image. Step S3: Obtain the final reconstructed image through several iterations, performing the following steps in each iteration: The compensated and reconstructed image is orthographically projected to obtain a compensated sinusoidal image. The original sinusoidal image and the compensated sinusoidal image are then fused according to the compensated weight sinusoidal image to obtain a fused sinusoidal image. The fused sinusoidal image is then reconstructed to obtain a fused reconstructed image. Determine whether the iteration termination condition is met; if it is met, use the fused and reconstructed image of the current iteration as the final reconstructed image; otherwise, use the fused and reconstructed image of the current iteration as the compensated reconstructed image of the next iteration and start the next iteration.

[0025] Specifically, such as Figure 1 As shown, when using a detector to collect projection data of a workpiece to be reconstructed, for example, when the workpiece to be reconstructed is a battery pack, there are high-density metal parts inside the battery pack, meaning that there are scenes that cannot be penetrated during the reconstruction of the battery pack. Insufficient X-ray penetration will lead to missing or distorted sine wave data. In this case, the detector is used to collect projection data of the battery pack as the original sine wave image.

[0026] Specifically, such as Figure 1 As shown, in step S1, the original sinusoidal image of the workpiece to be reconstructed is acquired by the detector, and the original sinusoidal image is reconstructed to obtain the initial reconstructed image.

[0027] It is understandable that existing reconstruction algorithms, such as filtered back projection algorithms and iterative reconstruction algorithms, are used when reconstructing the original sinusoidal image, which will not be elaborated here.

[0028] Specifically, such as Figure 1 As shown, in step S1, the initial reconstructed image of the workpiece to be reconstructed is segmented to obtain a background region, a boundary region, an artifact region, and multiple types of regions. For example, the initial reconstructed image of a battery pack is segmented to obtain a casing region, an air region, an artifact region, a cell region, and an adhesive layer region.

[0029] Preferably, such as Figure 2 As shown, the step of segmenting the initial reconstructed image to determine the artifact region includes: Step S81: Divide the initial reconstructed image into boundary region, background region and interior region; Step S82: Perform artifact detection on the internal region to obtain the artifact region and the non-artifact region; Step S83: Based on multiple pre-set attenuation coefficient features, multiple attenuation coefficient intensity constraints and fusion conditions, the non-artifact region is segmented to obtain multiple types of regions.

[0030] Specifically, such as Figure 2 As shown, in step S81, the initial reconstructed image can be divided according to the foreground distribution, outer contour, background contrast information, boundary information, or prior geometric information of the initial reconstructed image to determine the boundary region, background region, and interior region.

[0031] For example, when the workpiece to be reconstructed is a battery pack, a preset outer shell size is determined based on prior information about the battery pack, such as... Figure 4 As shown, the outer shell size is set to 4mm according to the battery pack model, and the initial reconstructed image is divided into boundary and background regions according to the preset outer shell size of the battery pack.

[0032] Specifically, such as Figure 3 As shown, in the initial reconstructed image of the battery pack, the reconstructed image of the black background area (air area of ​​the battery pack) is relatively clear. At the same time, combined with the size of the battery pack shell, the area with a fixed size outward from the background area is divided into the boundary area (shell area of ​​the battery pack). The boundary area and background area obtained at this time are relatively accurate, and no further segmentation is required.

[0033] Specifically, such as Figure 3 As shown, the area within the inner boundary of the background area is considered the internal area.

[0034] It is worth noting that, such as Figure 2As shown, in step S82, artifact detection is performed on the internal region to identify artifact regions caused by the inability of rays to penetrate, and these regions are excluded from subsequent segmentation to avoid artifacts interfering with the segmentation results, thus obtaining artifact regions and non-artifact regions.

[0035] Preferably, the step of performing artifact detection on the internal region to obtain artifact-free and non-artifact-free regions includes: Anomalies in the attenuation coefficient values ​​of pixels in the internal region are detected to obtain a set of pixels with abnormal intensity artifacts. Structural anomaly detection is performed on the pixels in the internal region to obtain a set of structural anomaly artifact pixels; The set of intensity anomalous artifact pixels and the set of structure anomalous artifact pixels are fused together, and the artifact region is obtained by combining morphological closure operation. At the same time, the other regions in the internal region other than the artifact region are regarded as non-artifact regions.

[0036] It is understandable that, such as Figure 3 As shown, the pixel value of a pixel in the initial reconstructed image is a grayscale value, which represents the attenuation coefficient value of the pixel. In this invention, the pixel value in the reconstructed image is uniformly described using the attenuation coefficient value, which will not be elaborated further.

[0037] Specifically, intensity anomaly detection is performed on all pixels within the internal region to identify anomalous pixels that do not conform to physical laws or significantly deviate from the local statistical distribution. For example, anomalous candidate regions can be constructed based on global mean, global standard deviation, local mean, local standard deviation, and adaptive low-value thresholds.

[0038] Specifically, based on the physical characteristics, statistical distribution features, or empirical thresholds of the workpiece to be reconstructed, abnormally low pixel values ​​can be detected in the internal region. For example, such as... Figure 3 As shown, when the workpiece to be reconstructed is a battery pack, physically speaking, the attenuation coefficient value of the pixels in the reconstructed image should generally not be negative. Therefore, a negative value is set as the threshold for detecting abnormal attenuation coefficient values. For example, the negative attenuation coefficient threshold is -0.0003 mm. - ¹.

[0039] Specifically, all pixels in the internal region are traversed, and the attenuation coefficient value of each pixel is compared with the negative attenuation coefficient threshold. If the attenuation coefficient value of a pixel is less than the negative attenuation coefficient threshold, then that pixel is identified as an intensity anomalous artifact pixel. After comparing all pixels in the internal region, the set of all intensity anomalous artifact pixels is taken as the intensity anomalous artifact pixel set.

[0040] Preferably, the method for detecting abnormal attenuation coefficient values ​​includes one or more of the following: Detection of abnormally low values; Detection of abnormally high values; Local statistical outlier detection; Detection of anomalies in neighborhood mean.

[0041] Specifically, based on the performance of the initial reconstructed image of the workpiece to be reconstructed, the detectable anomalies include: abnormally low value regions, abnormally high value regions, extreme deviation regions in local statistical significance, and pixels with significantly inconsistent neighborhood means and standard deviations.

[0042] Specifically, structural anomaly detection is performed on pixels in the internal region to identify areas with obvious directionality, morphological abnormalities, or inconsistencies with normal structures.

[0043] Specifically, detectable structural anomalies include: striped anomaly regions, radial anomaly regions, banded anomaly regions, wedge-shaped or block-shaped anomaly regions, low-value regions clustered along a specific direction, and abnormal connected regions that do not conform to the normal structural topology distribution. The detection forms are preset based on the initial reconstructed image of the workpiece to be reconstructed.

[0044] Specifically, the set of intensity anomalous artifact pixels and the set of structure anomalous artifact pixels are fused together. A morphological closure operation is performed on the fused pixels to obtain the artifact region in the internal region. Finally, the other regions in the internal region except for the artifact region are taken as the non-artifact region.

[0045] For example, when the workpiece to be reconstructed is a battery pack, due to the presence of impenetrable areas in the battery pack, the structural anomalies caused by the impenetrable areas can manifest as butterfly-shaped low-value artifacts distributed along the diagonal direction.

[0046] Specifically, a diagonal distance threshold is pre-set based on prior information about the battery pack. At the same time, the average value and standard deviation of all pixels in the internal region are calculated as the average value and standard deviation of the attenuation coefficient. The pixels in the internal region are detected using the diagonal distance threshold, the average value of the attenuation coefficient, and the standard deviation of the attenuation coefficient to obtain a butterfly-shaped artifact pixel set, i.e., a set of structural anomaly artifact pixels.

[0047] Specifically, the step of detecting pixels in the internal region using a diagonal distance threshold and the average and standard deviation of the attenuation coefficient values ​​of pixels in the internal region includes: The butterfly artifact attenuation coefficient threshold is constructed based on the average value and standard deviation of the attenuation coefficient values ​​of the pixels in the internal region. The pixel in the internal region is detected using the butterfly artifact attenuation coefficient threshold to obtain the first candidate butterfly artifact pixel set. Calculate the distance between the pixels in the internal region and the two diagonals of the internal region, and select the smaller distance as the diagonal distance of the pixel; select the set of pixels in the internal region whose diagonal distance is less than the diagonal distance threshold as the second candidate butterfly-shaped artifact pixel set; The intersection of the first candidate butterfly-shaped artifact pixel set and the second candidate butterfly-shaped artifact pixel set is taken as the butterfly-shaped artifact pixel set.

[0048] Specifically, the butterfly artifact attenuation coefficient threshold is constructed based on the average value and standard deviation of the attenuation coefficient value. For example, the butterfly artifact attenuation coefficient threshold is set as: average value of attenuation coefficient value - standard deviation of attenuation coefficient value.

[0049] Specifically, all pixels in the internal region are traversed, and the attenuation coefficient value of each pixel is compared with the butterfly artifact attenuation coefficient threshold. If the attenuation coefficient value of a certain pixel is less than the butterfly artifact attenuation coefficient threshold, the pixel is selected as the first candidate butterfly artifact pixel.

[0050] Specifically, after comparing all pixels in the internal region with the butterfly artifact attenuation coefficient threshold, the set of first candidate butterfly artifact pixels is taken as the first candidate butterfly artifact pixel set.

[0051] Specifically, the expressions for the two diagonals of the internal region are calculated, all pixels in the internal region are traversed, and the distances of each pixel to the two diagonals are calculated by combining the two-dimensional coordinates of all pixels.

[0052] Specifically, for a given pixel, the distance between the pixel and the two diagonals is calculated, the two distances are compared, and the smaller distance is selected as the diagonal distance of the pixel.

[0053] Specifically, the diagonal distance of all pixels in the internal region is compared with a pre-set diagonal distance threshold. If the diagonal distance of a pixel is less than the diagonal distance threshold, the pixel is selected as the second candidate butterfly artifact pixel. Finally, the set of all second candidate butterfly artifact pixels in the internal region is selected as the second candidate butterfly artifact pixel set.

[0054] Specifically, after obtaining the first candidate butterfly-shaped artifact pixel set and the second candidate butterfly-shaped artifact pixel set, the intersection of the first candidate butterfly-shaped artifact pixel set and the second candidate butterfly-shaped artifact pixel set is taken as the butterfly-shaped artifact pixel set.

[0055] Specifically, artifact detection is performed on the internal region to obtain a set of negative artifact pixels (a set of intensity anomalous artifact pixels) and a set of butterfly-shaped artifact pixels (a set of structural anomalous artifact pixels). The union of the set of negative artifact pixels and the set of butterfly-shaped artifact pixels is taken as the artifact pixel set. Based on all the pixels in the artifact pixel set, morphological closure operations are used to fill the holes to obtain the artifact region.

[0056] It is worth noting that after filling the holes with morphological closure operations based on all pixels in the artifact pixel set, the isolated areas with an area of ​​less than 50 pixels are removed, and the resulting area is taken as the artifact region.

[0057] Specifically, such as Figure 2 As shown, in step S82, the artifact region is first obtained, and all pixels in the inner region are removed from the artifact region. The set of remaining pixels is taken as the non-artifact region.

[0058] It is worth noting that, based on the spatial distribution characteristics of X-ray impermeable artifacts and the butterfly-shaped low-value region along the diagonal direction, a special artifact detection criterion was designed to avoid artifact regions being incorrectly segmented as adhesive layers or air, thereby improving the accuracy of subsequent segmentation.

[0059] Specifically, such as Figure 2 As shown, in step S83, one or more representative attenuation parameters can be preset based on the prior material parameters, material library parameters, adaptive estimation parameters, or statistical reference parameters of the workpiece to be reconstructed. For example, when the workpiece to be reconstructed is a battery pack, the adhesive layer attenuation coefficient and the cell attenuation coefficient are preset based on the prior information of the battery pack.

[0060] Specifically, such as Figure 4 As shown, based on the structural composition of the battery pack model, the material degradation coefficient library of the battery pack (for 9MeV X-rays) is queried in advance to obtain the degradation coefficient of the adhesive layer and the degradation coefficient of the battery cell: Aluminum alloy casing ; lithium battery cells ; Epoxy resin adhesive layer ; Atmosphere .

[0061] Specifically, such as Figure 2 As shown, in step S83, the non-artifact region is segmented using multiple pre-set attenuation coefficient features, multiple attenuation coefficient intensity constraints and fusion conditions, and the non-artifact region is segmented into multiple types of regions; for example, when the workpiece to be reconstructed is a battery pack, all pixels in the non-artifact region can be further classified into cell region and adhesive layer region.

[0062] Preferably, the non-artifact region is segmented based on multiple pre-set attenuation coefficient features, multiple attenuation coefficient intensity constraints, and fusion conditions to obtain multiple types of regions, including: Multiple attenuation coefficient features are used to classify the pixels in the non-artifact region, resulting in multiple sets of attenuation coefficient features. Multiple attenuation coefficient intensity constraints are used to classify the intensity of pixels in the non-artifact region, resulting in multiple sets of attenuation coefficient intensity constraints. Based on the fusion conditions, multiple sets of attenuation coefficient features and multiple sets of attenuation coefficient intensity constraints are fused to obtain multiple sets of pixels in different types of regions. Morphological correction operations are performed on pixel sets of multiple types of regions to obtain multiple types of regions.

[0063] Specifically, different attenuation coefficient features are used to classify pixels in non-artifact regions, resulting in different sets of attenuation coefficient features corresponding to different attenuation coefficient features.

[0064] Preferably, the plurality of attenuation coefficient characteristics include one or more of the following: Gradient and edge features; Projection and boundary characteristics; Geometric positional features; Structural prior features; Confidence characteristics.

[0065] Specifically, gradient and edge features are calculated in the initial reconstructed image or non-artifact regions, including: gradient magnitude, gradient direction, horizontal edge response, vertical edge response, Sobel, Prewitt, Scharr and other edge responses, and multi-directional edge saliency.

[0066] Specifically, regarding projection and boundary features, to enhance the identification ability of layered, banded, periodic, or repeating structure boundaries, statistical analysis can be performed on non-artifact regions in the row and column directions to construct one-dimensional variation curves, including: row mean curve, column mean curve, row gradient mean curve, and column gradient mean curve. These curves can detect intensity trough locations, gradient peak locations, candidate boundary centers, and potential material transition zone locations.

[0067] Specifically, geometric position features calculate the geometric relationship between pixels and several reference regions to describe the structural distribution, including distance to the outer boundary, distance to the center line / skeleton, distance to the high-confidence core region, connected region area, thinness, aspect ratio, roundness, and orientation.

[0068] Specifically, regarding structural prior features, to improve the robustness of identifying low-contrast targets, high-confidence core regions can be identified first, and then surrounding candidate regions can be inferred based on these core regions. Extractable structural priors include layered structure candidates, strip structure candidates, ring structure candidates, block structure candidates, multi-connected main structure candidates, and one or more high-confidence core regions.

[0069] Specifically, the confidence feature establishes one or more intensity constraint intervals for the pixel attenuation coefficient value based on the prior material parameters or statistical parameters of the workpiece to be reconstructed. These intervals can be derived from: whether the region where the pixel is located is far from artifacts, the confidence of pixel classification, the score of consistency with local structure, and the degree of confidence in matching the subsequent compensation model.

[0070] For example, the workpiece to be reconstructed is a battery pack, and preferably, the plurality of attenuation coefficient feature sets include one or more of the following sets: A collection of pixels in a strip-shaped region of the adhesive layer; A set of pixels with low local attenuation coefficients; Pixel set constrained by distance between adhesive layer and battery cell; A set of pixels with high gradient magnitude and attenuation coefficient; The set of pixels with high attenuation coefficients in the horizontal direction edge values; The set of pixels with high attenuation coefficients at vertical edge values; Non-cell core area pixel set.

[0071] Specifically, each attenuation coefficient feature set represents the set of pixels in the non-artifact region that satisfy the corresponding feature conditions.

[0072] Preferably, the pixel set of the adhesive strip region is obtained through the following steps: A horizontal attenuation coefficient variation curve is constructed based on the average attenuation coefficient of each row of pixels in the internal region, and a vertical attenuation coefficient variation curve is constructed based on the average attenuation coefficient of each column of pixels in the internal region. Construct a horizontal gradient magnitude change curve based on the average gradient magnitude of each row of pixels in the internal region, and construct a vertical gradient magnitude change curve based on the average gradient magnitude of each column of pixels in the internal region. The position of the center pixel of the adhesive layer in the horizontal direction is determined by the change curve of the attenuation coefficient in the horizontal direction and the change curve of the gradient amplitude in the horizontal direction; the position of the center pixel of the adhesive layer in the vertical direction is determined by the change curve of the attenuation coefficient in the vertical direction and the change curve of the gradient amplitude in the vertical direction. The set of pixels in the horizontal adhesive layer strip region is determined by using the center pixel position of the horizontal adhesive layer and the preset horizontal adhesive layer pixel width; the set of pixels in the vertical adhesive layer strip region is determined by using the center pixel position of the vertical adhesive layer and the preset vertical adhesive layer pixel width. The intersection of the union of the pixel set of the horizontal adhesive strip region and the pixel set of the vertical adhesive strip region with the pixel points in the non-artifact region is taken as the pixel set of the adhesive strip region.

[0073] Specifically, such as Figure 3 As shown, the number of pixels in each row of the internal region and the attenuation coefficient value of each row of pixels are counted. Then, the average attenuation coefficient corresponding to each row of pixels is calculated. The average attenuation coefficient corresponding to all rows of pixels is curve-fitted to obtain the attenuation coefficient change curve in the horizontal direction.

[0074] Specifically, such as Figure 3 As shown, the number of pixels in each column and the attenuation coefficient value of each column of pixels in the internal region are counted. Then, the average attenuation coefficient of each column of pixels is calculated. The average attenuation coefficient of all columns of pixels is curve-fitted to obtain the attenuation coefficient change curve in the vertical direction.

[0075] Specifically, such as Figure 3 As shown, the gradient magnitude of the attenuation coefficient values ​​of all pixels in the internal region is calculated to obtain the gradient magnitude map of the internal region.

[0076] Specifically, based on the gradient magnitude map of the internal region, the number of pixels in each row of the internal region and the gradient magnitude of each row of pixels are counted. Then, the average gradient magnitude of each row of pixels is calculated, and the average gradient magnitude of all rows of pixels is curve-fitted to obtain the gradient magnitude change curve in the horizontal direction.

[0077] Specifically, based on the gradient magnitude map of the internal region, the number of pixels in each column and the gradient magnitude of each column of pixels in the internal region are counted, and then the average gradient magnitude of each column of pixels is calculated. The average gradient magnitude of all columns of pixels is then curve-fitted to obtain the gradient magnitude change curve in the vertical direction.

[0078] Preferably, determining the position of the center pixel of the adhesive layer in the horizontal direction based on the horizontal attenuation coefficient variation curve and the horizontal gradient amplitude variation curve, and determining the position of the center pixel of the adhesive layer in the vertical direction based on the vertical attenuation coefficient variation curve and the vertical gradient amplitude variation curve, includes: The low point in the horizontal attenuation coefficient variation curve is taken as the initial horizontal adhesive layer center pixel position, and the peak point in the horizontal gradient amplitude variation curve is taken as the horizontal boundary position of the adhesive layer cell. The initial horizontal adhesive layer center pixel position is filtered using the horizontal boundary position of the adhesive layer cell to obtain the horizontal adhesive layer center pixel position. The trough position in the vertical attenuation coefficient variation curve is taken as the initial vertical adhesive layer center pixel position, and the peak position in the vertical gradient amplitude variation curve is taken as the vertical boundary position of the adhesive layer cell. The initial vertical adhesive layer center pixel position is merged and deduplicated using the vertical boundary position of the adhesive layer cell to obtain the vertical adhesive layer center pixel position.

[0079] Specifically, the trough position in the horizontal attenuation coefficient change curve is detected as the initial horizontal adhesive layer center pixel position, and the peak position in the horizontal gradient amplitude change curve is detected as the horizontal boundary position of the adhesive layer cell. The horizontal boundary position of the adhesive layer cell and the initial horizontal adhesive layer center pixel position are merged and deduplicated to obtain the horizontal adhesive layer center pixel position.

[0080] Specifically, the trough position in the vertical attenuation coefficient change curve is detected as the initial vertical adhesive layer center pixel position, and the peak position in the vertical gradient amplitude change curve is detected as the vertical boundary position of the adhesive layer cell. The vertical boundary position of the adhesive layer cell and the initial vertical adhesive layer center pixel position are merged and deduplicated to obtain the vertical adhesive layer center pixel position.

[0081] Specifically, the horizontal adhesive layer pixel width and the vertical adhesive layer pixel width are preset. The set of pixels in the horizontal adhesive layer strip region is determined by the center pixel position of the horizontal adhesive layer and the preset horizontal adhesive layer pixel width. The set of pixels in the vertical adhesive layer strip region is determined by the center pixel position of the vertical adhesive layer and the preset vertical adhesive layer pixel width. The intersection of the union of the horizontal adhesive layer strip region pixel set and the vertical adhesive layer strip region pixel set with the pixels in the non-artifact region is taken as the set of pixels in the adhesive layer strip region.

[0082] It is worth noting that a dual detection mechanism of intensity troughs and gradient peaks is employed to improve the robustness of boundary detection. When artifacts make the intensity troughs insignificant, the gradient peaks can still provide boundary information; and vice versa. The two detection results are complementary, ensuring accurate location of the adhesive layer even in the presence of artifact interference.

[0083] Specifically, the set of pixels with low local attenuation coefficients is obtained through the following steps: A Gaussian weighted window is used to calculate the average and standard deviation of the local attenuation coefficients corresponding to pixels in the internal region. An exemplary Gaussian weighted window is a 15*15 window. The Gaussian weighted window is used to traverse all pixels in the internal region to obtain the average and standard deviation of the local attenuation coefficients corresponding to each pixel. The average and standard deviation of the local attenuation coefficients corresponding to all pixels in the internal region are combined to obtain the mean map and standard deviation map of the local attenuation coefficients of the internal region. Pixels whose attenuation coefficient values ​​are lower than the local attenuation coefficient low value threshold are regarded as local attenuation coefficient low value pixels. The intersection of the set of all local attenuation coefficient low value pixels and the pixels in the non-artifact region is regarded as the local attenuation coefficient low value pixel set.

[0084] For example, the local attenuation coefficient low threshold of a pixel is calculated as the average local attenuation coefficient of the pixel minus 0.3 * the standard deviation of the local attenuation coefficient of the pixel. The attenuation coefficient value of the pixel is compared with the local attenuation coefficient low threshold. If the attenuation coefficient value is less than the local attenuation coefficient low threshold, the pixel is considered a local attenuation coefficient low pixel. The set of local attenuation coefficient low pixels is obtained by scalaring the intersection of the set of all local attenuation coefficient low pixels in the internal region and the pixels in the non-artifact region.

[0085] Preferably, the pixel set of the non-cell core region is obtained through the following steps: Calculate the average of the preset adhesive layer attenuation coefficient and the preset cell attenuation coefficient, and use it as the preset classification threshold; By using a preset classification threshold, pixels in the non-artifact region are classified to obtain the core cell pixel set; Morphological erosion is performed on the pixels in the core cell pixel set to obtain multiple core cell regions; the remaining pixels after removing the pixels in the multiple core cell regions from the non-artifact region are taken as the non-core cell pixel set.

[0086] Specifically, the average of the preset adhesive layer attenuation coefficient and the preset battery cell attenuation coefficient is used as the preset classification threshold. The attenuation coefficient values ​​of pixels in the non-artifact area are compared with the preset classification threshold. The set of pixels with values ​​lower than the preset classification threshold is used as the core battery cell pixel set. Morphological erosion is performed on all pixels in the core battery cell pixel set to obtain multiple battery cell core regions.

[0087] Specifically, after obtaining multiple core regions of the battery cell, the set of pixels remaining after removing the pixels in the multiple core regions from all pixels in the non-artifact regions is taken as the set of pixels in the non-core regions.

[0088] Preferably, the set of pixels constrained by the distance between the adhesive layer and the battery cell is obtained through the following steps: The set of pixels remaining after removing pixels from multiple core areas of the battery cell in the non-artifact region is taken as the set of edge-uncertain pixels; Calculate the distances between pixels in the uncertain pixel set at the edge and multiple core regions of the battery cell, and take the shortest distance as the battery cell distance of the pixel; The set of pixels in the edge uncertain pixel set whose cell distance is less than the preset cell core distance threshold is used as the adhesive layer cell distance constraint pixel set.

[0089] Specifically, pixels in the non-core area pixel set are considered as edge-uncertain pixels, and the set of all edge-uncertain pixels is considered as the edge-uncertain pixel set.

[0090] Specifically, all pixels in the uncertain edge pixel set are traversed, the distance between each pixel and multiple core areas of the battery cell is calculated, the shortest distance is selected as the battery cell distance of that pixel, and pixels whose battery cell distance is less than the preset battery cell core distance threshold are selected as the set of adhesive layer battery cell distance constrained pixels.

[0091] Specifically, the set of pixels with high gradient magnitudes of attenuation coefficients is obtained through the following steps: Calculate the gradient magnitude of the attenuation coefficient of all pixels in the internal region, combine them to obtain the gradient magnitude map of the internal region, and count the pixels in the gradient magnitude map of the internal region whose gradient magnitude is greater than the preset quantile threshold as the high gradient pixel set. The intersection of the high gradient pixel set and the pixels in the non-artifact region is the high gradient magnitude pixel set of the attenuation coefficient.

[0092] For example, the set of pixels with gradient magnitudes greater than the 70th percentile in the gradient magnitude map of the statistical internal region, and the intersection of these pixels with pixels in the non-artifact region, are used as the set of pixels with high gradient magnitudes and attenuation coefficients.

[0093] Specifically, the set of pixels with high attenuation coefficients in the horizontal direction is obtained through the following steps: The horizontal Sobel operator is used to detect the horizontal attenuation coefficient edge values ​​in the internal region, resulting in a horizontal edge map of the internal region. The attenuation coefficient edge values ​​of all pixels in the horizontal edge map are statistically analyzed. The 80th percentile attenuation coefficient edge value is used as the horizontal edge threshold. The horizontal attenuation coefficient edge values ​​of all pixels in the non-artifact region are compared with the horizontal edge threshold. The set of pixels with attenuation coefficient edge values ​​greater than the horizontal edge threshold is used as the set of pixels with high horizontal attenuation coefficient edge values.

[0094] Specifically, the set of pixels with high attenuation coefficients in the vertical direction is obtained through the following steps: The vertical Sobel operator is used to detect the edge values of the attenuation coefficient in the vertical direction in the internal region, obtaining the vertical edge map of the internal region. The edge values of the attenuation coefficient of all pixel points in the vertical edge map are statistically calculated, and the 80th percentile attenuation coefficient edge value is used as the vertical edge threshold. The edge values of the attenuation coefficient in the vertical direction of all pixel points in the non-artifact region are respectively compared with the vertical edge threshold, and the combination of pixel points with the edge value of the attenuation coefficient in the vertical direction greater than the vertical edge threshold is used as the pixel set with high vertical edge value of the attenuation coefficient.

[0095] Specifically, multiple attenuation coefficient intensity constraints are respectively used to classify the pixel points in the non-artifact region, obtaining multiple attenuation coefficient intensity constraint sets.

[0096] Exemplarily, when the workpiece to be reconstructed is a battery pack, preferably, the multiple attenuation coefficient intensity constraint sets include one or more of the following sets: {0.0002 < f < 0.85 μcell}; {0.4 μglue < f < 2.0 μglue}; {0.4 μglue < f < 0.85 μcell}; Where, μcell represents the preset cell attenuation coefficient of the battery pack, and μglue represents the preset glue layer attenuation coefficient of the battery pack.

[0097] Specifically, multiple attenuation coefficient intensity constraints are set according to the preset cell attenuation coefficient and the preset glue layer attenuation coefficient: The set of pixel points in the non-artifact region with the attenuation coefficient value satisfying 0.0002 < f < 0.85 μcell is used as {0.0002 < f < 0.85 μcell}; The set of pixel points in the non-artifact region with the attenuation coefficient value satisfying 0.4 μglue < f < 2.0 μglue is used as {0.4 μglue < f < 2.0 μglue}; The set of pixel points in the non-artifact region with the attenuation coefficient value satisfying 0.4 μglue < f < 0.85 μcell is used as {0.4 μglue < f < 0.85 μcell}.

[0098] Specifically, multiple types of region pixel sets are determined according to the multiple attenuation coefficient feature sets and the multiple attenuation coefficient intensity constraint sets.

[0099] Exemplarily, when the workpiece to be reconstructed is a battery pack, the following fusion conditions are set: Condition A, Condition B, Condition C, Condition D, and Condition E, where: In Condition A, the intersection of the pixel set of the glue layer strip region and {0.0002 < f < 0.85 μcell} is used as the pixel point set obtained by Condition A; In condition B, the intersection of the set of pixels with low local attenuation coefficient, the set of pixels with glue layer cell distance constraint, and the set of pixels with high gradient magnitude of attenuation coefficient is used as the set of pixel points obtained under condition B; In condition C, the intersection of the set of pixels with high gradient magnitude of attenuation coefficient and {0.4μglue < f < 2.0μglue} is used as the set of pixel points obtained under condition C; In condition D, the intersection of the set of pixels with high horizontal edge value of attenuation coefficient, the set of pixels with high vertical edge value of attenuation coefficient, and {0.4μglue < f < 0.85μcell} is used as the set of pixel points obtained under condition D; In condition E, the pixel points in the set of pixels in the non-core area of the cell are used as the set of pixel points obtained under condition E.

[0100] Finally, the union of the sets of pixel points obtained under condition A, condition B, condition C, condition D, and condition E is used as the set of non-artifact glue layer pixels.

[0101] It should be noted that during fusion, multiple attenuation coefficient features, multiple attenuation coefficient intensity constraints, and fusion conditions need to be set according to different workpieces to be reconstructed, and a comprehensive judgment mechanism of multi-condition logic is adopted. Each condition judges pixel points of different types of regions from different perspectives, such as position, intensity, gradient, and geometric direction constraints. When artifacts cause certain conditions to fail, other conditions can still provide a basis for judgment, realizing robust recognition of low-contrast type regions.

[0102] It should be noted that a single feature may fail under conditions of artifacts, noise, or low contrast, but multiple features provide judgment bases from different perspectives, and the segmentation robustness can be improved through fusion.

[0103] Specifically, the fusion method may include but is not limited to logical union, logical intersection, rule combination, voting mechanism, weighted scoring, clustering or classifier, region growing, and graph model optimization. After fusion determination, multiple types of regions are obtained.

[0104] Exemplarily, when segmenting the initial reconstructed image of a battery pack, after obtaining the set of non-artifact glue layer pixels, the remaining pixel points after removing all pixel points of the non-artifact glue layer pixels in the non-artifact region are used as the set of non-artifact cell pixels.

[0105] Preferably, the morphological correction operation includes one or more of the following: Hole filling operation; Opening and closing operation; Connected component screening operation; Small area removal operation; Boundary smoothing operation; Region merging or splitting operations; Label consistency correction operation.

[0106] Specifically, morphological correction operations are performed on pixel sets of multiple types of regions. For example, when the workpiece to be reconstructed is a battery pack, a morphological closure operation is performed on the pixels in the non-artifact cell pixel set to obtain the cell region, and the remaining region in the non-artifact region after removing the cell region is taken as the adhesive layer region.

[0107] Specifically, the internal region is divided into artifact regions and multiple types of regions through hole filling operation, opening and closing operation, connected component filtering operation, small region removal operation, boundary smoothing operation, region merging or splitting operation, and label consistency correction operation.

[0108] Specifically, such as Figure 2 As shown, in step S82, the initial reconstructed image is segmented into a background region, a boundary region, an artifact region, and multiple types of regions. The regions of the initial reconstructed image other than the artifact region include the background region, the boundary region, and multiple types of regions. The preset attenuation coefficients include a preset background attenuation coefficient, a preset boundary attenuation coefficient, and different types of attenuation coefficients.

[0109] For example, when the workpiece to be reconstructed is a battery pack, the preset attenuation coefficients include preset shell attenuation coefficients, preset air attenuation coefficients, preset cell attenuation coefficients, and preset adhesive layer attenuation coefficients. It is worth noting that the preset attenuation coefficients are used to replace the attenuation coefficient values ​​of pixels other than those in the artifact regions in the initial reconstructed image. The attenuation coefficient values ​​of pixels in the artifact regions are retained, and the preset shell attenuation coefficient, preset air attenuation coefficient, preset cell attenuation coefficient, and preset adhesive layer attenuation coefficient are used to replace the attenuation coefficient values ​​of pixels in the shell region, air region, cell region, and adhesive layer region, respectively. The resulting image is used as the compensated reconstructed image.

[0110] Specifically, such as Figure 1 As shown, in step S2, the penetration threshold is determined based on the original sine image. The penetration threshold is set by statistically analyzing the attenuation values ​​corresponding to each pixel in the original sine image.

[0111] Preferably, determining the penetration threshold based on the original sine wave image includes: The frequency corresponding to each attenuation value in the original sine image is counted. The attenuation value is used as the horizontal axis and the corresponding frequency is used as the vertical axis to obtain the distribution histogram of the original sine image. Find the attenuation value point corresponding to the maximum frequency in the distribution histogram as the global attenuation value point; Starting from the global attenuation value point, the attenuation value at the position corresponding to the preset frequency difference is searched along the direction of increasing attenuation value as the penetration threshold.

[0112] Specifically, all attenuation values ​​in the original sine image are expanded into a one-dimensional array. The number of times each attenuation value appears in the one-dimensional array is counted, and the frequency of each attenuation value in the one-dimensional array is calculated. The distribution histogram of the one-dimensional array is plotted with the attenuation value as the x-axis and the corresponding frequency as the y-axis, which is the distribution histogram of the original sine image.

[0113] It is worth noting that, due to the impenetrable region and missing data, this distribution histogram typically presents a sharp left peak and a trailing right shoulder. The left peak corresponds to all rays that pass through the "impenetrable region," where attenuation values ​​are concentrated near the theoretical minimum due to photon starvation or detector saturation, forming a tall, narrow peak or plateau. The right shoulder corresponds to rays that pass through permeable materials inside the battery, such as electrodes, separators, electrolytes, and plastic casings; their attenuation values ​​are more widely distributed, forming a slowly descending tail.

[0114] Specifically, the distribution histogram of the original sinusoidal image is lightly Gaussian smoothed to suppress minor jitter caused by random noise.

[0115] Specifically, the attenuation value point corresponding to the maximum frequency in the distribution histogram is found as the global attenuation value point. In the distribution histogram of the original sine wave image of the battery pack, this represents the left peak, and the maximum frequency value corresponding to the global attenuation value point is determined.

[0116] Specifically, the attenuation value of a fixed frequency at the maximum frequency interval is determined as the preset frequency difference. For example, a frequency value 10% smaller than the maximum frequency value is set. Starting from the global attenuation value point, a frequency value 10% smaller than the maximum frequency value is found in the direction of increasing attenuation value. The corresponding attenuation value is then used as the penetration threshold.

[0117] It is worth noting that the penetration threshold is determined based on the original sine wave in this invention, and the penetration threshold can be adaptively adjusted according to the signal of the workpiece to be reconstructed and the current and voltage during CT scanning, so that this invention can produce a good reconstruction effect when processing workpieces with different signals.

[0118] Specifically, such as Figure 1 As shown, in step S2, a compensated weighted sinusoidal image is generated using the penetration threshold, the original sinusoidal image, and the initial reconstructed image.

[0119] Preferably, the step of generating a compensated weighted sinusoidal image using a penetration threshold, the original sinusoidal image, and the initial reconstructed image includes: A confidence-based reconstructed image is generated from the initial reconstructed image, and a confidence-based sinusoidal image is obtained by row orthographic projection of the confidence-based reconstructed image. The penetration mask value of the pixel in the original sinusoidal image is determined based on the penetration threshold, and the penetration mask sinusoidal image is obtained. The compensation weight sine image is obtained by multiplying the pixel values ​​of corresponding pixels in the confidence sine image and the penetration mask sine image.

[0120] Specifically, a confidence-based reconstructed image is generated based on the initial reconstructed image.

[0121] It's worth noting that segmentation results with small gradients and clear material boundaries in the initial reconstructed image are more reliable, such as in uniform regions inside the battery cell. Accurate segmentation implies reliable orthographic projection results, and in these cases, the compensation data is more reliable. Conversely, segmentation results with large gradients and ambiguous material attributions in the initial reconstructed image are uncertain, such as in material boundaries or artifact regions. In these cases, the original data is more reliable to avoid the propagation of segmentation errors to the final result.

[0122] Preferably, generating a confidence-based reconstructed image from the initial reconstructed image includes: The gradient magnitude image is obtained by calculating the gradient magnitude of the pixels in the initial reconstructed image. The gradient magnitude of the pixels in the gradient magnitude image is then normalized to obtain the normalized gradient magnitude image. Inverting the pixel values ​​of pixels in the normalized gradient magnitude image yields the intermediate confidence image; Gaussian smoothing and preset range mapping are performed sequentially on the intermediate confidence images to obtain the confidence reconstructed images.

[0123] Specifically, the gradient magnitude of each pixel in the initial reconstructed image is calculated based on its attenuation coefficient, resulting in a gradient magnitude map. Locations with large gradients are typically situated at material boundaries or in artifact regions, leading to uncertain segmentation results; locations with small gradients are located within homogeneous material, resulting in reliable segmentation results.

[0124] Specifically, the gradient magnitude of the pixels in the gradient magnitude image is normalized to the range of [0, 1], resulting in a normalized gradient magnitude image.

[0125] Specifically, the confidence level is negatively correlated with the normalized gradient magnitude. The intermediate confidence level image is obtained by inverting the pixel values ​​of the pixels in the normalized gradient magnitude image. ; in, This represents the pixel value of a pixel in the intermediate confidence image. This represents the pixel value of a pixel in the normalized gradient magnitude.

[0126] Specifically, the intermediate confidence images are sequentially subjected to Gaussian smoothing and preset range mapping to obtain the confidence reconstructed images: Gaussian smoothing is applied to the intermediate confidence image to eliminate confidence fluctuations caused by local noise. For example, a 5×5 Gaussian kernel with a standard deviation σ = 1.5 pixels is used. The Gaussian-smoothed intermediate confidence image is mapped to a preset confidence range, for example, [0.3, 0.9], to avoid extreme weights.

[0127] It is worth noting that range mapping ensures that even in the highest confidence region, the original data still retains 10% of the contribution, while in the lowest confidence region, the compensated data still retains 30% of the contribution. This avoids complete reliance on a single data source and improves the robustness of the fusion results of the original and compensated sine images.

[0128] Specifically, the confidence reconstructed image is orthogonally projected to obtain a confidence sine image. For example, the orthogonal projection algorithm used here is the ray-driven orthogonal projection algorithm.

[0129] Specifically, the penetration threshold is used to determine which pixels in the original sinusoidal image belong to the impenetrable scene and which belong to the penetrable scene. For pixels in the impenetrable scene, the penetration mask value is set to 1, and the compensated data and the original data are fused together. For pixels in the penetrable scene, the penetration mask value is set to 0, and the original data is directly retained without any modification.

[0130] It is understandable that after setting the penetration mask values ​​of all pixels in the original sinusoidal image, the image composed of the penetration mask values ​​of all pixels is used as the penetration mask sinusoidal image, and the pixel value of each pixel in the penetration mask sinusoidal image is the penetration mask value.

[0131] Specifically, the compensation weight sinusoidal image is obtained by multiplying the pixel values ​​of corresponding pixels in the confidence sinusoidal image and the penetration mask sinusoidal image. In other words, the result of multiplying the pixel values ​​of pixels in the confidence sinusoidal image and the pixel values ​​of corresponding pixels in the penetration mask sinusoidal image is used as the pixel value of the corresponding pixel in the compensation weight sinusoidal image.

[0132] Specifically, such as Figure 1 As shown, in step S2, the iteration termination condition is determined based on the penetration threshold and the original sine image.

[0133] Preferably, determining the iteration termination condition based on the penetration threshold and the original sinusoidal image includes: The number of pixels in the original sinusoidal image that belong to the impenetrable scene is determined based on the penetration threshold. Determine whether the ratio of the number of pixels that cannot penetrate the scene to the number of pixels in the original sine image exceeds a preset iteration ratio threshold; if it does not exceed the threshold, the iteration termination condition is to perform only one iteration.

[0134] Specifically, by using the penetration threshold, we can determine which pixels in the original sine image belong to the impenetrable scene and which belong to the penetrable scene, and then count the number of pixels in the impenetrable scene and the number of pixels in the original sine image.

[0135] Specifically, it is determined whether the ratio of the number of pixels in the impenetrable scene to the number of pixels in the original sinusoidal image exceeds a preset iteration ratio threshold. If the ratio does not exceed the preset iteration ratio threshold, it indicates that the proportion of the impenetrable area in the original sinusoidal image is small. In this case, a good reconstruction effect can be achieved with one iteration, without the need for multiple iterations.

[0136] Preferably, if the ratio of the number of pixels that cannot penetrate the scene to the number of pixels in the original sinusoidal image exceeds a preset iteration ratio threshold, then the iteration termination condition shall be any of the following termination conditions: The number of iterations exceeds a preset threshold. The difference between the fused reconstructed image and the compensated reconstructed image in the current iteration is less than a preset difference threshold.

[0137] Specifically, if the ratio of the number of pixels in the impenetrable scene to the number of pixels in the original sinusoidal image exceeds the preset iteration ratio threshold, it indicates that the proportion of the impenetrable area in the original sinusoidal image is large. In this case, the iteration can be terminated when the number of iterations exceeds the preset iteration number threshold or when the difference between the fused reconstructed image and the compensated reconstructed image in the current iteration is less than the preset difference threshold, so as to improve the reconstruction effect as much as possible.

[0138] Specifically, the difference between the fused reconstructed image and the compensated reconstructed image in the current iteration can be calculated using the root mean square error of the fused reconstructed image and the compensated reconstructed image. For example, the preset difference threshold is set to 0.01. When the root mean square error of the fused reconstructed image and the compensated reconstructed image in the iteration is less than 0.01, it is determined to be converged, and the iteration terminates at this time.

[0139] It is worth noting that, such as Figure 1 As shown, steps S1 and S2 are for providing the initial compensated reconstructed image for step S3, as well as the fixed compensation weight sine image and the original sine image. Steps S1 and S2 can be executed in parallel or sequentially, without specific limitations here.

[0140] It is understandable that in step S3, the compensated reconstructed image of the first iteration uses the compensated reconstructed image obtained in step S1, and the compensated reconstructed image of subsequent iterations uses the fused reconstructed image obtained in the previous iteration.

[0141] Specifically, such as Figure 1 As shown, in step S3, the final reconstructed image is obtained through several iterations.

[0142] Specifically, in each iteration, the compensated reconstructed image is orthographically projected to obtain a compensated sinusoidal image. Then, based on the compensated weighted sinusoidal image, the original sinusoidal image and the compensated sinusoidal image are fused. The fused image is used as the fused sinusoidal image. The original sinusoidal image and the compensated sinusoidal image are adaptively fused. In areas where the rays penetrate well, the authenticity of the original data is preserved. In areas where the rays cannot penetrate, the compensated data is used to repair the missing information, so as to achieve the complementary advantages of the two.

[0143] Preferably, the pixel value of each pixel in the fused sinusoidal image is calculated using the following formula: ; in, Indicates the detector angle. Indicates the detector channel. This represents the pixel value of a pixel in a fused sine wave image. This represents the pixel value of a pixel in the compensated weight sine wave image. This represents the pixel value of the compensated sine wave image. This represents the pixel value of a pixel in the original sine wave image.

[0144] Specifically, through This means that any pixel in the sinusoidal image corresponds one-to-one with the pixels in the compensated weighted sinusoidal image, the original sinusoidal image, and the compensated sinusoidal image. The pixel value of the pixel in the fused sinusoidal image is calculated using the above formula.

[0145] Specifically, such as Figure 1 As shown, in each iteration of step S3, the obtained fused sinusoidal image is used for reconstruction. Existing reconstruction algorithms, such as filtered back projection algorithm and iterative reconstruction algorithm, are used during reconstruction, which will not be elaborated here.

[0146] Specifically, such as Figure 1 As shown, in step S3, after obtaining the fused reconstructed image of each iteration, it is necessary to determine whether the iteration termination condition is met. For example, it is determined whether the number of iterations exceeds a preset iteration number threshold, or whether the difference between the fused reconstructed image and the compensated reconstructed image of the current iteration is less than a preset difference threshold.

[0147] If the number of iterations exceeds a preset iteration threshold, or if the difference between the fused reconstructed image and the compensated reconstructed image in the current iteration is less than a preset difference threshold, then the iteration termination condition is met; otherwise, it is not met.

[0148] Specifically, in each iteration, if the iteration termination condition is met, the fused reconstructed image of the current iteration is used as the final reconstructed image to obtain the reconstructed image corresponding to the original sine image, that is, to obtain the final reconstructed image of the workpiece to be reconstructed.

[0149] Specifically, in each iteration, if the iteration termination condition is not met, the fused reconstructed image of the current iteration is used as the compensation reconstructed image for the next iteration, and the next iteration begins until the iteration termination condition is met, and the final reconstructed image is obtained.

[0150] It is worth noting that the image iterative reconstruction method provided by the embodiments of the present invention, applicable to scenarios with a large number of impermeable areas, is suitable for industrial CT online inspection scenarios, especially for inspection tasks containing high-density materials such as lithium battery packs and metal components. It can effectively suppress artifacts, improve the clarity of material boundaries and the visibility of internal structures while ensuring processing efficiency.

[0151] Compared with existing technologies, the image iterative reconstruction method for scenarios with a large number of impenetrable regions provided by this invention segments the initial reconstruction image of the workpiece to be reconstructed, and replaces the attenuation coefficient values ​​of other pixels in the initial reconstruction image (excluding artifact regions) with a preset attenuation coefficient to obtain a compensated reconstruction image. Furthermore, it determines a penetration threshold based on the original sine image, improving the segmentation effect for impenetrable regions of different workpieces to be reconstructed. It generates a compensated weighted sine image using the penetration threshold, the original sine image, and the initial reconstruction image, and determines the iteration termination condition based on the penetration threshold and the original sine image, adapting to the reconstruction process of different penetration regions and improving processing efficiency and reconstruction effect. In each iteration, the compensated reconstruction image is orthographically projected to obtain a compensated sine image, and the original sine image is then compared with the compensated weighted sine image. The image and the compensated sinusoidal image are fused to obtain a fused sinusoidal image. The fused sinusoidal image is then reconstructed to obtain the final reconstructed image. Missing data is directly repaired in the sinusoidal domain, avoiding artifact diffusion caused by image domain correction and improving the reconstruction effect. Furthermore, the initial reconstructed image of the workpiece to be reconstructed is divided into boundary region, background region, and interior region based on the prior information of the workpiece to be reconstructed. Artifact detection is performed on the interior region to obtain artifact regions. Pixels in the non-artifact regions are classified using multiple attenuation coefficient features and multiple attenuation coefficient intensity constraints. Multiple sets of attenuation coefficient features and multiple sets of attenuation coefficient intensity constraints are fused based on fusion conditions. Finally, morphological correction operations are used to obtain multiple types of regions, making the initial reconstructed image of the segmented workpiece more robust and the segmentation effect better.

[0152] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0153] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. An iterative image reconstruction method applicable to scenes with a large number of impermeable regions, characterized in that, The iterative reconstruction method includes: The original sinusoidal image of the workpiece to be reconstructed is reconstructed to obtain an initial reconstructed image. The initial reconstructed image is segmented to determine the artifact region. The attenuation coefficient values ​​of other pixels in the initial reconstructed image, excluding the artifact region, are replaced with a preset attenuation coefficient to obtain a compensated reconstructed image. The penetration threshold is determined based on the original sinusoidal image; a compensation weight sinusoidal image is generated using the penetration threshold, the original sinusoidal image, and the initial reconstructed image; and the iteration termination condition is determined based on the penetration threshold and the original sinusoidal image. The final reconstructed image is obtained through several iterations, with the following steps performed in each iteration: The compensated and reconstructed image is orthographically projected to obtain a compensated sinusoidal image. The original sinusoidal image and the compensated sinusoidal image are then fused according to the compensated weight sinusoidal image to obtain a fused sinusoidal image. The fused sinusoidal image is then reconstructed to obtain a fused reconstructed image. Determine whether the iteration termination condition is met; if it is met, use the fused and reconstructed image of the current iteration as the final reconstructed image; otherwise, use the fused and reconstructed image of the current iteration as the compensated reconstructed image of the next iteration and start the next iteration.

2. The iterative reconstruction method according to claim 1, characterized in that, The determination of the iteration termination condition based on the penetration threshold and the original sinusoidal image includes: The number of pixels in the original sinusoidal image that belong to the impenetrable scene is determined based on the penetration threshold. Determine whether the ratio of the number of pixels that cannot penetrate the scene to the number of pixels in the original sine image exceeds a preset iteration ratio threshold; if it does not exceed the threshold, the iteration termination condition is to perform only one iteration.

3. The iterative reconstruction method according to claim 2, characterized in that, If the ratio of the number of pixels that cannot penetrate the scene to the number of pixels in the original sinusoidal image exceeds a preset iteration ratio threshold, then the iteration termination condition shall be any of the following termination conditions: The number of iterations exceeds a preset threshold. The difference between the fused reconstructed image and the compensated reconstructed image in the current iteration is less than a preset difference threshold.

4. The iterative reconstruction method according to any one of claims 1-3, characterized in that, The step of determining the penetration threshold based on the original sinusoidal image includes: The frequency corresponding to each attenuation value in the original sine image is counted. The attenuation value is used as the horizontal axis and the corresponding frequency is used as the vertical axis to obtain the distribution histogram of the original sine image. Find the attenuation value point corresponding to the maximum frequency in the distribution histogram as the global attenuation value point; Starting from the global attenuation value point, the attenuation value at the position corresponding to the preset frequency difference is searched along the direction of increasing attenuation value as the penetration threshold.

5. The iterative reconstruction method according to claim 1, characterized in that, The process of generating a compensated weighted sinusoidal image using a penetration threshold, the original sinusoidal image, and the initial reconstructed image includes: A confidence-based reconstructed image is generated from the initial reconstructed image, and a confidence-based sinusoidal image is obtained by row orthographic projection of the confidence-based reconstructed image. The penetration mask value of the pixel in the original sinusoidal image is determined based on the penetration threshold, and the penetration mask sinusoidal image is obtained. The compensation weight sine image is obtained by multiplying the pixel values ​​of corresponding pixels in the confidence sine image and the penetration mask sine image.

6. The iterative reconstruction method according to claim 5, characterized in that, The step of generating a confidence-based reconstructed image from the initial reconstructed image includes: The gradient magnitude image is obtained by calculating the gradient magnitude of the pixels in the initial reconstructed image. The gradient magnitude of the pixels in the gradient magnitude image is then normalized to obtain the normalized gradient magnitude image. Inverting the pixel values ​​of pixels in the normalized gradient magnitude image yields the intermediate confidence image; Gaussian smoothing and preset range mapping are performed sequentially on the intermediate confidence images to obtain the confidence reconstructed images.

7. The iterative reconstruction method according to claim 1, characterized in that, In each iteration, the pixel value of each pixel in the fused sinusoidal image is calculated using the following formula: ； in, Indicates the detector angle. Indicates the detector channel. This represents the pixel value of a pixel in a fused sine wave image. This represents the pixel value of a pixel in the compensated weight sine wave image. This represents the pixel value of the compensated sine wave image. This represents the pixel value of a pixel in the original sine wave image.

8. The iterative reconstruction method according to claim 1, characterized in that, The step of segmenting the initial reconstructed image to determine artifact regions includes: The initial reconstructed image is divided into boundary region, background region, and interior region; Artifact detection is performed on the internal region to obtain the artifact region and the non-artifact region; Based on multiple pre-set attenuation coefficient features, multiple attenuation coefficient intensity constraints, and fusion conditions, the non-artifact region is segmented to obtain multiple types of regions.

9. The iterative reconstruction method according to claim 8, characterized in that, Based on multiple pre-set attenuation coefficient features, multiple attenuation coefficient intensity constraints, and fusion conditions, the non-artifact region is segmented to obtain multiple types of regions, including: Multiple attenuation coefficient features are used to classify the pixels in the non-artifact region, resulting in multiple sets of attenuation coefficient features. Multiple attenuation coefficient intensity constraints are used to classify the intensity of pixels in the non-artifact region, resulting in multiple sets of attenuation coefficient intensity constraints. Based on the fusion conditions, multiple sets of attenuation coefficient features and multiple sets of attenuation coefficient intensity constraints are fused to obtain multiple sets of pixels in different types of regions. Morphological correction operations are performed on pixel sets of multiple types of regions to obtain multiple types of regions.

10. The iterative reconstruction method according to claim 9, characterized in that, The plurality of attenuation coefficient characteristics include one or more of the following: Gradient and edge features; Projection and boundary characteristics; Geometric positional features; Structural prior features; Confidence characteristics.

Images