Cylindrical multilayer structure neutron / X-ray heterogenous image three-dimensional registration method

Abstract

The invention provides a three-dimensional registration method for neutron / X-ray heterogeneous images of a cylindrical multilayer structure. The method comprises the following steps of: adjusting the perpendicularity of an image pair by an edge extraction operator and PCA (Principal Component Analysis); the edge contour extraction module determines the height mapping of the heterogenous cross-sectional image; calculating a circle extraction algorithm and a circle center positioning translation amount; an image super-resolution reconstruction algorithm is used for realizing scale change of the image; and realizing image rotation matching based on a characteristic fault and phase cross-correlation method. The method can effectively correct geometric and modal differences between different-source images, realizes accurate alignment of metal and nonmetal regions, has significant advantages in the aspects of registration precision, edge preservation and modal feature alignment, has high adaptability and stability for complex structures and multi-modal feature characterization, and can be widely applied to the field of image registration. The method is suitable for three-dimensional registration of neutron / X-ray heterogenous images of cylindrical multi-layer structures such as cylindrical lithium batteries and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing and nondestructive testing, and in particular to a three-dimensional registration method for neutron / X-ray heterogeneous images of a cylindrical multilayer structure, which is particularly suitable for nondestructive testing and performance analysis of the internal structure of a lithium battery. Background Art

[0002] Ensuring efficient energy storage and utilization is a key issue for achieving sustainable development in today's society. Lithium-ion batteries, with their high energy density, excellent cycle stability, and lack of memory effect, have become the energy solution of choice across multiple industries, widely used in electric vehicles, aerospace equipment, and defense applications. As their use expands, demands for lithium battery performance continue to rise, particularly in terms of safety. In actual operation, lithium batteries may encounter problems such as overcharging, heat generation, and lithium dendrite growth, which can lead to corrosion of the current collector and gas generation. These issues not only shorten the battery lifespan but can also cause serious accidents such as thermal runaway, fire, and even explosion. To mitigate these potential risks, researchers have explored various improvement methods. First, battery thermal management systems can effectively control battery temperature during operation to prevent overheating. Furthermore, flame-retardant electrolytes, redox shuttles, safer separator materials, and more stable cathode materials have been developed to mitigate potential safety hazards. However, even with these measures, completely eliminating all risks remains challenging. Therefore, the use of non-destructive testing technologies to monitor the quality of lithium batteries in real time and ensure their stability during use has become an important solution. Through non-destructive testing technology, not only can the structural changes inside lithium batteries be detected, but it can also provide valuable basis for future material improvements and technology optimization.

[0003] Simultaneously utilizing neutron and X-ray imaging techniques to inspect complex structural samples containing light elements (hydrogen, lithium, boron, etc.) and metals (iron, copper, etc.), such as multi-stream mixtures, lithium-ion batteries, and electronic packaging devices, can provide more comprehensive information on the internal structure and composition of the sample, facilitating an accurate, multi-angle interpretation of the sample's material, structure, and quality information. Aligning and fusing images from the two modes facilitates the analysis and assessment of the internal quality status of lithium batteries, providing stronger support for lithium battery quality testing. However, since the two images are often acquired using different imaging devices, differences in imaging coordinate systems and imaging parameters can lead to many differences between X-ray tomographic images and neutron tomographic images in terms of image size, grayscale distribution, position, and scale. Therefore, a targeted image registration scheme needs to be designed to achieve high-precision alignment of heterogeneous images. Summary of the Invention

[0004] The purpose of the present invention is to provide a three-dimensional registration method for neutron / X-ray heterogeneous images of cylindrical multilayer structures to solve the problem of insufficient registration accuracy in the prior art and achieve high-precision non-destructive testing of the internal structure of lithium batteries.

[0005] The present invention is achieved through the following technical solutions: A method for three-dimensional registration of neutron / X-ray heterogeneous images of a cylindrical multi-layer structure is provided, comprising: Edge extraction operator and PCA principal component analysis are used to adjust the verticality of the image pairs; The edge contour extraction module determines the height map of the heterogeneous tomographic image; Use circle detection algorithm to detect the circle center and determine the translation amount to achieve translation alignment of heterogeneous tomographic images; According to the scale ratio, an image super-resolution reconstruction algorithm is used to realize the scale transformation super-resolution reconstruction of the tomographic image; Image rotation matching is achieved based on feature tomography and phase cross-correlation methods.

[0006] Furthermore, the verticality adjustment is implemented based on the projection image; the image pair to be registered is a neutron projection image and an X-ray projection image; the verticality adjustment requires grayscale normalization of the input image pair, normalizing the grayscale value to the range of [0, 1]; for the normalized projection image, the OSTU algorithm is used to perform threshold segmentation on the projection image to extract the edge contour of the projection image; The PCA principal component analysis means: firstly counting the coordinates of the boundary pixels, and then using PCA to determine that the first principal component is the direction of the symmetry axis.

[0007] Furthermore, the key to the threshold segmentation of the OSTU algorithm is to determine the segmentation threshold. By selecting a segmentation threshold T, the projected image is divided into two parts, the foreground and the background, so that the inter-class variance between the foreground and the background is Maximum; for the acquired edge contour image, the Sobel operator is used to extract the vertical boundary of the foreground area.

[0008] Furthermore, the edge contour extraction module determines the purpose of the height mapping of heterogeneous tomographic images to map the neutron and X-ray tomographic images to the same height; the height mapping determines the height corresponding to the head and tail tomographic images of the cylindrical multi-layer structure by determining the horizontal boundary of the projected image.

[0009] Furthermore, the height mapping method is: using an adaptive edge extraction operator to detect the horizontal boundary of the projected image, sorting the pixels at the horizontal boundary, and taking the average of the row coordinates of the first 20% of the pixels as the height of the horizontal boundary.

[0010] Furthermore, before edge detection, the input tomographic image pair is Gaussian smoothed using a Gaussian smoothing template, and then edge detection is performed. After edge detection, the holes of the tomographic image are filled based on the region growing principle. After the hole filling is completed, the center of the circle is detected using a differentiable Lie group Hough circle detection algorithm. The coordinates of the center of the circle and the diameter of the circle can be calculated based on the regional connectivity. The translation amount is determined according to the position of the center of the circle. After the centers of the heterogeneous tomographic images are moved to the center of the circle, the translation alignment of the heterogeneous tomographic images is completed.

[0011] Furthermore, the image super-resolution reconstruction algorithm includes using sinusoidal image stacking for noise reduction, adaptive histogram equalization to improve image clarity, homomorphic filtering to sharpen details, bilinear interpolation algorithm for upsampling, and homomorphic filtering to enhance contrast, thereby achieving super-resolution reconstruction and scale alignment of tomographic images.

[0012] Furthermore, the tomographic image of the top shell area of ​​the cylindrical multi-layer structure is used as the registration object, and the rotation angle is calculated before rotational registration is performed on all the tomographic images; the gradient of the tomographic image is first calculated and the central area of ​​the tomographic image is selectively retained, and the phase correlation method is used to match the heterogeneous gradient images.

[0013] The present invention also provides an electronic device comprising: a processor, a communication interface and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; the processor calls the logic instructions in the memory to execute the various steps of the three-dimensional registration method of neutron / X-ray heterogeneous images of a cylindrical multi-layer structure.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, it is implemented to perform the various steps of the three-dimensional registration method of neutron / X-ray heterogeneous images of a cylindrical multi-layer structure.

[0015] Based on the complex internal structure and multi-material distribution of lithium batteries, this paper proposes a 3D registration framework for heterogeneous imaging systems. The framework consists of four complementary stages and introduces three new algorithms at key stages to enhance overall robustness and accuracy. A designed adaptive edge extraction operator precisely locates structural edges in a multi-scale fractional gradient domain, automatically estimates projection tilt, and achieves pixel-level pose correction. A bilinear interpolation algorithm performs super-resolution upsampling on low-resolution images. The algorithm adaptively selects Padé poles within a conformal domain in the complex plane and combines them with Chebyshev nodes to minimize boundary ringing, effectively suppressing high-frequency artifacts and reconstructing details. For the registration of tomographic images, a differentiable Lie group Hough circle detection algorithm is used to accurately estimate the center and radius of the workpiece tomographic layer, and its gradient information is embedded in the manifold optimization. This algorithm simultaneously solves for the Euclidean translation vector and rotation angle, achieving global optimal registration of circular tomographic images. This four-stage "edge-geometry-scale-pose" pipeline couples detection, analysis, and optimization: an adaptive edge extraction operator ensures basic pose accuracy, a bilinear interpolation algorithm provides high-fidelity detail, and a differentiable Lie Group Hough circle detection algorithm directly incorporates circle geometry into the pose solution in a differentiable manner. This process not only eliminates scale, grayscale, and spatial mismatches caused by hardware differences, but also significantly improves the clarity and interpretable geometric consistency of the registered images, establishing a robust and refined alignment benchmark for subsequent cross-source data fusion.

[0016] This method offers significant advantages in registration accuracy, edge preservation, and modal feature alignment, demonstrating high adaptability and stability for characterizing the complex structures and multimodal properties of cylindrical lithium batteries. It can also be used for 3D registration of heterogeneous neutron / X-ray images of cylindrical multilayer structures other than cylindrical lithium batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 This is a flow chart of the method for three-dimensional registration of neutron / X-ray heterogeneous images of cylindrical multilayer structures provided by the present invention; Figure 2 It is the overall flow chart of the present invention; Figure 3 This is an example diagram of the relative tilt of the projection image provided by the present invention; Figure 4 is a grayscale distribution diagram of a neutron-X-ray projection image of a cylindrical lithium battery in an embodiment of the present invention; Figure 5is an example diagram of vertical boundary extraction in an embodiment of the present invention; Figure 6 is a flowchart of neutron-X-ray tomographic image translation calculation and alignment in an embodiment of the present invention; Figure 7 is a flow chart of super-resolution reconstruction in an embodiment of the present invention; Figure 8 This is a tomographic image registration result diagram of a No. 1 lithium battery in an embodiment of the present invention; Figure 9 This is a rendering effect diagram of the superimposed tomographic image in an embodiment of the present invention; Figure 10 This is a three-dimensional registered solid cross-sectional view of a No. 1 lithium battery in an embodiment of the present invention; Figure 11 This is a tomographic image registration result diagram of a No. 2 lithium battery in an embodiment of the present invention; Figure 12 This is a cross-sectional view of a No. 2 lithium battery in an embodiment of the present invention; Figure 13 This is a diagram showing the registration effect of a micro DC motor tomographic image in an embodiment of the present invention; Figure 14 This is a diagram showing the effect of motor three-dimensional image registration in an embodiment of the present invention; Figure 15 It is a structural schematic diagram of the heterogeneous image three-dimensional registration device provided by the present invention; Figure 16 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0019] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0020] Figure 2The present invention provides a method for three-dimensional registration of neutron / X-ray heterogeneous images of cylindrical multi-layer structures. The cylindrical multi-layer structure of this embodiment takes cylindrical lithium batteries as the research object. Different information can be observed by performing CT imaging inspection of cylindrical lithium batteries using neutrons and X-rays. However, since neutron CT and X-ray CT are two independent imaging systems, when imaging an object, the differences in the posture of the object, the geometric relationship of the imaging system, and the imaging parameters (such as ray energy, photosensitive element size, imaging magnification ratio, etc.) will cause the collected projection image and the reconstructed tomographic image to differ in verticality, rotation angle, foreground position, tomographic height, and image resolution. In response to the existing problems, the present invention has designed a method for three-dimensional registration of neutron / X-ray heterogeneous images of cylindrical multi-layer structures of this embodiment. Figure 2 The three-dimensional registration process of heterogeneous images of cylindrical lithium batteries shown in the figure decouples image matching into five sub-problems: data loading, verticality adjustment, inter-layer matching, translation and scale transformation, and rotation angle matching, which are solved step by step.

[0021] Reference Figure 1 and Figure 2 The embodiment of the present invention provides a method for three-dimensional registration of neutron / X-ray heterogeneous images of a cylindrical multilayer structure, comprising: S1, edge extraction operator and PCA principal component analysis are used to adjust the verticality of the image pairs.

[0022] The present invention designs a method for adjusting the verticality of the projected image. Image fusion requires that the tomographic images to be fused reflect the information at the same position in the actual object, that is, the tomographic images must correspond to slices of the same plane of the cylindrical lithium battery. However, during actual imaging detection, differences in system structure or external experimental factors may cause the projected images of the cylindrical lithium battery on the two imaging systems to be relatively tilted, such as Figure 3 As shown, the slice positions of the two sets of tomographic images are not in the same plane. Under this condition, the two sets of tomographic images reconstructed cannot be mapped to the same plane, and the verticality of the acquired images needs to be adjusted. Due to the large difference in information reflected in neutron and X-ray projection images, satisfactory registration results cannot be achieved by similarity matching based on grayscale information or using feature matching. Considering that the edges of the projection image of the cylindrical lithium battery are relatively clear, the present invention uses an edge extraction algorithm to extract the vertical boundaries of the cylindrical lithium battery, and then calculates the inclination angle between the vertical boundaries to achieve vertical adjustment of the projection image.

[0023] like Figure 4As shown, since the grayscale distribution of the X-ray projection image and the neutron projection image is different, it is necessary to perform grayscale normalization on the input image pair (neutron projection image and X-ray projection image) and normalize the grayscale value to the range of [0,1]. For the normalized projection image, the OSTU algorithm (Otsu algorithm) is used to perform threshold segmentation on the projection image to extract the edge contour of the projection image. The key to the threshold segmentation of the OSTU algorithm is to determine the segmentation threshold. The mathematical principle of the segmentation threshold calculation is shown in formula (1). By selecting a segmentation threshold T, the projection image is divided into two parts, the foreground and the background, so that the inter-class variance between the foreground and the background is maximum.

[0024] (1); In formula (1), when the segmentation threshold is T, and Respectively represent the proportion of the number of pixels in the foreground area and the background area to the total number; and Represents the grayscale mean of the foreground and background areas.

[0025] (2); The segmentation threshold is determined by formula (1), and the projection image can be segmented by threshold according to formula (2) to obtain the edge contour image. .in Represents the grayscale value of the pixel at row i and column j. For the acquired edge contour image, the Sobel operator is used to extract the vertical boundary of the foreground area.

[0026] Because the accuracy of edge extraction algorithms is limited, if the inclination angle of the vertical boundary is calculated directly, the calculation result will be less accurate. Figure 5 It can be seen that the vertical boundary image of the lithium battery is a binary image, and the image is bilaterally symmetrical. If the symmetry axes on the left and right sides are calculated, and then the inclination angle of the symmetry axes is calculated, the vertical adjustment of the cylindrical lithium battery can be achieved. Based on this consideration, the present invention first counts the coordinates of the boundary pixels (i.e., pixels with a grayscale value of 1), and then uses PCA to determine the first principal component, which is the direction of the symmetry axis. The principle of PCA is shown in formula (3). PCA is a statistical method that aims to project high-dimensional data into a new coordinate system through linear transformation, and in this new coordinate system, the variance of the data is arranged in descending order. The goal of this method is to find the direction with the largest variance in the data set, that is, the direction of the principal component. The first principal component is the direction with the largest variance in the data set, so it can explain the largest changes in the data.

[0027] (3); In formula (3), Is the image data The grayscale mean, is the image data after removing the mean, yes The transpose of is the covariance matrix, is the eigenvector of the covariance matrix, indicating the direction of the i-th direction (principal component) in the data set; is the eigenvalue in the i-th direction, indicating the variance of the data in the i-th direction; is the eigenvector corresponding to the largest eigenvalue; is the position coordinate vector of the i-th pixel, for exist After determining the symmetry axis of the boundary image, the tilt angle of the symmetry axis can be calculated. By rotating the projected image according to the tilt angle, the vertical adjustment of the projected image can be completed.

[0028] S2. The edge contour extraction module determines the height mapping of the heterogeneous tomographic image.

[0029] Aiming at the problem of slice position differences caused by different resolutions of neutron / X-ray CT devices, the present invention designs a slice image height mapping method.

[0030] In order to align heterogeneous tomographic images, the height corresponding to the tomographic images of the front and rear of the cylindrical lithium battery can be determined by determining the horizontal boundary of the projected image. To achieve this goal, an adaptive edge extraction operator is used to detect the horizontal boundary of the projected image. Due to the limited accuracy of edge detection and the fact that the projected images of the upper and lower boundaries of the cylindrical lithium battery are not strictly straight lines, the row coordinates of the pixels on the edge are not the same. In order to improve the calculation accuracy, the pixels at the horizontal boundary are sorted, and the average of the row coordinates of the first 20% of the pixels is used as the height of the horizontal boundary. The specific principle is shown in formula (4).

[0031] (4); In formula (4), Represents the binary image after edge extraction, i is the row index of the pixel, j is the column index of the pixel, Represents the matrix after index association; flatten() represents flattening a two-dimensional image into a one-dimensional array; is the sorted array, Represents descending sorting of a one-dimensional array; represents the mode extraction function, Indicates the height of the first fault, represents the height of the tail fault, Indicates filtering the top N data. It means to filter the bottom N data, where N is equal to 20% of the image width. and Respectively represent the row indices of the first 20% and last 20% after sorting the one-dimensional array.

[0032] After the head and tail tomographic image positions of the neutron and X-ray sources are determined according to formula (4), since the height of the cylindrical lithium battery is fixed and the projection image has been adjusted for verticality, the height positions of all tomographic images can be accurately aligned and mapped with the head and tail tomographic image positions as the standard.

[0033] S3. Use a circle detection algorithm to detect the center of the circle and determine the translation amount to achieve translation alignment of heterogeneous tomographic images.

[0034] During CT imaging, changes in the distance of an object relative to the rotation center of the turntable will cause the object's position in the projected image to change. Both neutron and X-ray systems require 360-degree projection sampling around the object during CT imaging. The difference in the initial irradiation angle during sampling will cause rotation between the reconstructed images. The principle can be explained by formulas (5)-(7): (5); in, is the tomographic image to be reconstructed, x is the pixel horizontal coordinate, y is the pixel vertical coordinate; Indicates the angle The projection data along the direction of line t; is the Dirac function, which is used to select the projection direction. As can be seen from formula (5), at the angle Projection data at It can be regarded as the cumulative projection of the object at this angle.

[0035] Assume that the initial projection angle becomes , then: (6); in, is the rotation angle, is the projection data after rotation; Tomographic images are usually obtained by using a filtered back-projection algorithm. The essence is to accumulate projection values ​​at different angles to the pixels traversed by the ray path. The calculation principle is as follows: (7); in, Indicates the angle Along the Directional projection data; When the initial projection angle changes, the angles of all projected images change relative to the original projection. The tomographic image reconstruction formula becomes: (8); in, is the reconstructed tomographic image after rotation; Let the projection angle after rotation be : (9); It can be seen from formula (9) that when the projection data occurs After the change, the reconstructed tomographic image will also produce the same rotation angle relative to the original tomographic image.

[0036] Based on the above analysis, after completing the vertical adjustment of the projected image and the height alignment of the tomographic images, it is still necessary to consider the scale differences, translation, and rotation in the tomographic image alignment. Considering that the foreground area of ​​the cylindrical lithium battery tomographic image is circular, this paper designs a differentiable Lie group Hough circle detection algorithm to extract the circular boundary of the tomographic image.

[0037] like Figure 6 As shown, the present invention uses an adaptive edge extraction operator to extract the horizontal boundaries of the tomographic images. Because the operator is sensitive to noise, a 5×5 Gaussian smoothing template is used to smooth the input tomographic images before edge detection to achieve noise reduction. Edge detection is then performed. After edge detection, many unnecessary structures are detected within the outermost closed circle. To improve circle detection accuracy, holes in the tomographic images are filled based on the principle of region growing. After hole filling, the differentiable Lie group Hough circle detection algorithm is used to detect the circle center. The coordinates of the circle center and the circle diameter are calculated based on regional connectivity. The translation amount is determined based on the center position (the default is to move the circle center to the exact center of the image). After the centers of the heterogeneous tomographic images are moved to the center, the translation alignment of the heterogeneous tomographic images is completed.

[0038] S4. Using an image super-resolution reconstruction algorithm according to the scale ratio to achieve scale transformation of the tomographic image.

[0039] Neutron and X-ray CT imaging systems differ in their geometry and imaging methods. Neutron beams are parallel, so the image produced is the same size as the actual object. X-ray beams are cone-shaped, and the size of the projected image increases as the distance between the object and the radiation source decreases. The projected image is always magnified relative to the actual object. Furthermore, the camera in a neutron imaging system and the detector in an X-ray imaging system generally have different pixel sizes and numbers. These factors directly lead to differences in the resolution of the reconstructed tomographic images and the size of the foreground area of ​​the image.

[0040] Step S3 can move the foreground area in the tomographic image to the center of the tomographic image (i.e. the center of the circle coincides with the center of the image), and the diameter of the tomographic image can also be determined. Assume that the diameter of the lithium battery in the X-ray tomographic image is , the diameter of the lithium battery in the neutron tomography image is , calculate the size ratio: In order to achieve image resolution matching, super-resolution reconstruction of low-resolution images is required.

[0041] The essence of super-resolution reconstruction is grayscale interpolation based on the original image information using various interpolation methods, such as nearest neighbor interpolation, linear interpolation, B-spline interpolation, multi-scale interpolation, and deep learning-based interpolation methods. When using interpolation algorithms for image super-resolution reconstruction, the presence of noise significantly affects the reconstruction quality. Noise can cause artifacts and blurring during the interpolation process, resulting in loss of detail or blurring. Since interpolation algorithms generally assume smooth images, denoising should be performed in the presence of significant noise to improve the quality of the final reconstructed image. During CT imaging, the turntable rotates 360 degrees while the detector samples, resulting in symmetrical projections 180 degrees apart. Given that noise is random (both in intensity and location), the noise distribution in two projections 180 degrees apart will differ. This characteristic allows for the two images to be superimposed and averaged to reduce the noise content.

[0042] To improve the quality of super-resolution reconstructed images, highlight details, and improve contrast and clarity, image enhancement is performed before and after upsampling the low-resolution images. In addition, to avoid over-enhancement, the grayscale of the tomographic images is adjusted using an adaptive histogram equalization method.

[0043] The algorithm flow of super-resolution image reconstruction is as follows Figure 7 As shown in the figure, sinusoidal image stacking is used for noise reduction, adaptive histogram equalization is used to improve image clarity, homomorphic filtering is used to sharpen details, bilinear interpolation algorithm is used for upsampling, and homomorphic filtering is used to enhance contrast, thus achieving super-resolution reconstruction and scale registration of tomographic images.

[0044] After the image is subjected to adaptive histogram equalization, the clarity of the image is improved, and after being processed with homomorphic filtering, the detailed features of the image are sharpened.

[0045] The bilinear interpolation algorithm can better preserve image details, achieve smooth edge transitions, effectively reduce aliasing effects, and make images look more natural and realistic. In addition, compared with nearest neighbor interpolation, the designed method can avoid interpolation distortion in high-frequency areas (such as image edges), reduce visual distortion problems, and improve the overall clarity and visual quality of the image. When interpolating and upsampling the image, the present invention uses this method for upsampling. Figure 8 As can be seen, the edges and details of the interpolated image are more prominent, and the image resolution is significantly improved. Bilinear interpolation inevitably introduces some blur, especially in areas with high detail, which may produce slight artifacts or jagged edges. Filtering can smooth these imperfect areas and improve edge appearance. With this in mind, homomorphic filtering is used again on the upsampled image to enhance detail contrast.

[0046] S5. Image rotation angle matching.

[0047] By analyzing the structure of cylindrical lithium batteries, we found that the battery shell has more distinct features, and that heterogeneous tomographic images share a certain degree of similarity. With this in mind, we selected the tomographic image of the cylindrical lithium battery's top shell as the registration target for rotational matching. After calculating the rotation angle, we then performed rotational registration on all tomographic images. To further improve similarity, we first calculated the image gradient and selectively retained the image's central region.

[0048] The phase cross-correlation method can utilize the phase information of the image and has strong robustness to image rotation. Phase cross-correlation converts the image from the spatial domain to the frequency domain through Fourier transform, which can globally consider the frequency components of the image without the need for local pixel-by-pixel optimization. In addition, since phase cross-correlation mainly relies on the phase information of the image, it has good anti-interference ability against noise and local artifacts in the image. Considering the low signal-to-noise ratio and similarity in neutron and X-ray tomography images, the present invention adopts this method to match heterogeneous gradient images. The principle is as follows: First, the tomographic image to be matched is subjected to Fourier transform according to formula (10) to obtain the spectrum: (10); in, is the neutron tomographic function in the spatial domain, is the X-ray tomographic image function in the spatial domain, is the frequency domain representation of the neutron tomographic image after Fourier transformation, is the frequency domain representation of the X-ray tomographic image after Fourier transformation, x is the spatial abscissa, y is the spatial ordinate, u is the frequency domain abscissa, v is the frequency domain ordinate, e is a natural constant, and j is an imaginary unit; Then calculate the cross power spectrum based on the spectrum data: (11); in, is the cross power spectrum, for The complex conjugate of After performing an inverse Fourier transform on the power spectrum, relevant information of the image to be matched can be obtained, where the angle corresponding to the largest correlation coefficient is the desired rotation angle: (12); in, is the correlation coefficient, is the inverse Fourier transform, is the rotation angle.

[0049] The registration of heterogeneous tomographic images can be completed by rotating the image at the rotation angle determined by the above process. The registration result is as follows: Figure 8 In order to more intuitively show the effect of image registration, Figure 8 The registered images are superimposed together. It can be seen from the superimposed images that the registration method designed in the present invention has high registration accuracy.

[0050] The present invention uses neutron CT equipment and X-ray CT equipment to image two cylindrical lithium batteries and obtain real projection images and tomographic images. The main imaging parameters are shown in Table 1.

[0051] Table 1. Cylindrical lithium battery X-ray CT imaging parameters

[0052] Neutron CT imaging experiments on cylindrical lithium batteries were conducted at the China Advanced Research Reactor (CARR) at the China Institute of Atomic Energy. Key system parameters included a full reactor power of 60 MW, a collimation ratio (L / D) of 70-400, and a neutron flux of 4 × 10⁷ ncm⁻² s⁻¹ (20 MW). Key imaging parameters are listed in Table 2.

[0053] Table 2. Neutron CT imaging parameters of cylindrical lithium batteries

[0054] The registration effect is as follows Figure 8 As shown in the image. As can be seen from the image, the superposition effect of the registered neutron tomography image and the X-ray tomography image shows a high degree of consistency. In fault 1, the structure of the multi-layer spiral coil is perfectly aligned, and the boundary of each layer of the coil is displayed as a continuous and clear circular ring shape in the fusion image, indicating the high accuracy of the registration method in correcting small curves and interlayer distances. In fault 2, due to the presence of local material changes and heterogeneous regions, the superposition image successfully shows the transition details between different materials. In particular, under the combined effect of the penetration information of the neutron tomography image on thick materials and the high-resolution information of the X-ray, the heterogeneous region appears more realistic and complete. The registration result not only clearly retains the precise boundary depiction of the metal area by the X-ray, but also enhances the characterization capability of the neutron ray for non-metallic areas (such as electrolyte or diaphragm layers).

[0055] In order to observe and analyze the effect of registration more clearly, Figure 8 The superimposed images in the figure are rendered in pseudo color, as shown in Figure 9 As shown. Pseudo-color processing further highlights the differences in properties of different materials, such as the distribution and interface clarity of key parts such as current collector, positive electrode, negative electrode, electrolyte, etc., significantly improving the visualization effect of the image. Figure 9 As can be seen, the light blue area corresponds to the electrolyte distribution, which is particularly prominent in neutron radiography. After registration, it can be accurately superimposed with the metal current collector (highlighted area) and electrode structure (positive and negative electrodes) of X-ray imaging. The circular area in the middle is displayed in black or dark gray, corresponding to the central core of the battery, and the material boundary between it and the peripheral area can be clearly distinguished. The positive and negative electrodes are represented by pseudo colors of different brightness, with obvious layering, and the alignment boundaries of each layer are highly consistent. In addition, the boundaries of the highlighted area of ​​the current collector are clear, showing its accurate geometric relationship with the spiral coil, while retaining the complementary information of neutron and X-ray imaging: neutron rays highlight non-metallic materials (such as electrolyte and negative electrode), while X-rays accurately depict the details of the metal current collector and electrode.

[0056] In order to more comprehensively evaluate the 3D registration effect, the tomographic image of the No. 1 lithium battery was displayed in 3D, and the 3D model was sectioned to further analyze the longitudinal registration accuracy of the image. Figure 10 Neutron tomography images, X-ray tomography images, and the resulting fused 3D model of a No. 1 lithium battery are displayed. In the neutron tomography image, the electrolyte 1 is clearly distributed, and the neutron imaging's penetrating properties for non-metallic materials ensure that the electrolyte's morphology is complete and its boundaries are distinct. In the X-ray tomography image, metallic structures such as the positive terminal 2 and current collector 3 are rendered in high-resolution detail. Through registration of the fused image, key components such as the electrolyte 1, positive terminal 2, current collector 3, and negative terminal 5 are precisely spatially aligned. In particular, the morphology and layering of the housing 4 and negative electrode sheet 7 are highly restored in the fused image. As can be seen from the longitudinal cross-sectional image, the fused image fully demonstrates the uniform distribution of the current collector and positive and negative electrode sheets throughout the entire lithium battery, while preserving the high penetrating properties of the neutron tomography image and the edge clarity of the X-ray tomography image. The fused 3D model effectively complements the material and structural properties of the disparate images, simultaneously demonstrating the spatial relationships between different materials and components within the lithium battery.

[0057] from Figure 11The registration results show that the X-ray tomography image clearly displays the boundary structure of the lithium battery's top metal components, while the neutron radiation highlights the distribution of the non-metallic areas. The registered neutron tomography image and the superimposed tomography image demonstrate precise spatial alignment of the top metal components with the non-metallic areas. The superimposed image clearly defines the layers and the relative positions of the components are highly consistent. The registration results demonstrate a high degree of consistency between the non-metallic distribution and the metallic structure in the superimposed tomography image.

[0058] Figure 12 The longitudinal and three-dimensional cross-section images of a No. 2 lithium battery are presented. The cross-section results demonstrate that the proposed method achieves superior alignment in both longitudinal and three-dimensional space. In the longitudinal cross-section image, the X-ray tomography image clearly reveals the boundary between the lithium battery's outer shell and the internal current collector, with particularly high-resolution details of the metal material being well captured. The neutron image highlights the distribution characteristics of non-metallic materials such as the electrolyte and separator within the battery. After registration, the neutron and X-ray tomography images are successfully spatially aligned, clearly displaying both the metal and non-metallic properties within the battery and revealing the complete internal structure. In the three-dimensional cross-section, the fused image exhibits significant layering and detail consistency, and the registered three-dimensional model demonstrates excellent spatial consistency. For example, in the top and bottom regions of the battery, complex geometric structures (such as the winding of the positive and negative electrodes and the distribution of the electrolyte) can be accurately represented in the fused image.

[0059] Table 3. Quantitative analysis of cylindrical lithium battery registration

[0060] As shown in Table 3, the proposed method's registration performance was quantitatively analyzed using DSC (Days Similarity Coefficient), NMI (Normalized Mutual Information), and GS (Gradient Similarity) as evaluation metrics. For battery No. 1, the DSC index was 96.78% and 97.72% at two fault locations, demonstrating high spatial alignment accuracy and demonstrating the high matching capability of the registration algorithm within complex battery structures. NMI was 41.31% and 43.17%, respectively, demonstrating significant success in aligning heterogeneous image information. GS was 79.91% and 81.25%, further verifying the consistency of image edge features and the accurate alignment of geometric details. From the perspective of DSC, the four fault positions of battery No. 2 all reach high values ​​(42.47%, 98.43%, 98.37%, and 98.67%, respectively). Among them, the improvement of fault 1 after registration is particularly obvious due to its large initial heterogeneity; the NMI ranges from 27.32% to 43.44%. As the fault moves from top to bottom, the information consistency gradually increases after registration; the GS ranges from 63.39% to 68.11%, indicating the stability of the proposed algorithm in maintaining geometric structure alignment and fusion feature restoration.

[0061] The present invention designs Figure 7 The super-resolution reconstruction module shown in Figure 4 performs image resizing and resolution calibration while also improving image quality. To quantitatively and objectively evaluate the module's performance in improving image quality, the peak signal-to-noise ratio (PSNR) and average gradient (AG) of the images before and after super-resolution are compared, as shown in Table 4.

[0062] Table 4. Quantitative analysis of super-resolution reconstruction

[0063] The data in Table 4 demonstrates that the designed super-resolution reconstruction module significantly improves image quality. Improvements in both PSNR and AG indicate that this method demonstrates high stability and adaptability in improving the resolution of cylindrical lithium battery tomographic images. The significant improvement in PSNR demonstrates the advantages of this reconstruction method in reducing image noise and improving signal quality. This is particularly true for images with low initial quality, as it effectively improves signal fidelity and ensures significant optimization of the visual and quantitative quality of the reconstructed image. Furthermore, the growth in average gradient demonstrates the method's superior performance in detail enhancement and edge feature enhancement. Analysis of multi-slice images reveals that the super-resolution reconstruction method effectively restores high-frequency information in the image, making the details of complex structures more clearly discernible. This effect not only improves image clarity and contrast but also enhances the ability to characterize multi-layered, complex geometric structures.

[0064] The image registration method designed in this paper is designed for cylindrical lithium batteries, but it should also be effective for samples with circular contours. To verify this hypothesis, a heterogeneous image 3D registration experiment was designed using a micro DC motor as the research object. The main experimental parameters are shown in Tables 5 and 6.

[0065] Table 5. X-ray CT scanning parameters

[0066] Table 6. Neutron X-ray CT scanning parameters

[0067] The size of the X-ray tomographic image is 370×370, and the size of the neutron tomographic image is 1058×1058. The registration results of the registration method designed by the present invention are as follows: Figure 13As shown, the detailed features of the registered images show that the image boundaries are highly consistently aligned, and the structural features show good consistency in the overlay image, avoiding misalignment or blurring caused by differences in heterogeneous imaging. The ROI region further verifies the high-precision characteristics of the registration algorithm. The geometric features and positions of this region are completely overlapped in the neutron and X-ray overlay images, and the boundaries are clear, indicating that the algorithm has excellent correction capabilities for subtle differences in heterogeneous materials in space. In addition, through regional analysis, non-metallic materials show their accurate position in the structure after fusion, while the metal parts show clear contrast and integrity. Further analyzing the material properties, neutron imaging highlights the thicker non-metallic areas, while X-ray imaging captures the high-resolution characteristics of the metal parts. The fused image fully retains the advantages of both imaging modes, achieving a comprehensive presentation of material properties and structural characteristics.

[0068] To quantitatively evaluate the image registration accuracy, we calculated the DSC, NMI, and GS of the registered images. The results show that the proposed method significantly improves image similarity after heterogeneous image registration. The numbers in parentheses in Table 7 represent the magnitude of the increase in the coefficients after registration. Table 7 shows that all three metrics for all faults have significantly improved, validating the high-precision registration capability of the proposed method. Specifically, DSC reached 88.79%, 81.18%, and 57.35% for the three faults, respectively, demonstrating a high degree of structural overlap between faults. Fault 1, in particular, achieved the highest registration accuracy due to its relatively regular geometric features. NMI reflects the degree of information sharing between images. The NMIs for faults 1 and 2 were 0.5674 and 0.4722, respectively, demonstrating the effectiveness of the algorithm in aligning information between heterogeneous images. GS further demonstrates the consistency of edge features after registration. GS for all three faults significantly improved, reaching a maximum of 83.85%.

[0069] Table 7. Quantitative analysis of micro DC motor registration

[0070] In order to more intuitively display the effect of 3D registration of heterogeneous images, we visualize the tomographic images in 3D, perform weighted fusion on the 3D models, and render them in pseudo-color.

[0071] Figure 14 The heterogeneous fusion model generated by the proposed method is demonstrated. The weighted fusion image clearly demonstrates the integrity of the micro DC motor's internal structure and the accurate restoration of its material distribution. This demonstrates the proposed method's powerful ability to characterize heterogeneous materials and its exceptional performance in capturing details.

[0072] Metallic materials exhibit high contrast and uniform distribution in the fused image. The proposed method accurately reproduces complex geometric forms, further validating the algorithm's effectiveness in registering metal and non-metal boundaries. Compared to single-modality imaging, the fused model retains both the high resolution of X-ray imaging and the penetrating power of neutron imaging. This not only enhances the representation of multiple materials in a single view, but also improves the hierarchical representation of tiny components.

[0073] From the above processing results, it can be seen that the three-dimensional registration method of neutron / X-ray heterogeneous images of cylindrical multi-layer structures provided by the present invention can solve the problems of inconsistent image size, grayscale distribution and spatial position caused by differences in geometric characteristics and parameters in heterogeneous imaging systems, and realize high-precision registration of heterogeneous cross-modal three-dimensional images of cylindrical workpieces.

[0074] The following describes the cylindrical multi-layer structure neutron / X-ray heterogeneous image three-dimensional registration device provided by the present invention. The cylindrical multi-layer structure neutron / X-ray heterogeneous image three-dimensional registration device described below and the heterogeneous image three-dimensional registration method described above can be referenced to each other.

[0075] like Figure 15 As shown, the cylindrical multi-layer structure neutron / X-ray heterogeneous image three-dimensional registration device provided by the present invention includes: The projection image verticality adjustment module 231 is used to adjust the inclination of the projection image; a tomographic image height mapping module 232 for mapping the heights of neutron and X-ray tomographic images; The translation matching module 233 is used to perform two-dimensional translation registration on the tomographic image; A super-resolution reconstruction module 234 is used to upsample the low-resolution image to achieve image scale registration; The image rotation matching module 235 is used to achieve matching of rotation angles of heterogeneous tomographic images.

[0076] The heterogeneous image three-dimensional registration device provided by the present invention can be used to execute the above-mentioned cylindrical multi-layer structure neutron / X-ray heterogeneous image three-dimensional registration method to solve the problems of inconsistent image size, grayscale distribution and spatial position caused by geometric characteristics and parameter differences in related technologies, effectively correct the geometric and modal differences between heterogeneous images, achieve accurate alignment of metal and non-metal areas, and significantly improve the robustness and accuracy of the registration.

[0077] Figure 16A schematic diagram of the physical structure of an electronic device is provided. The electronic device may include: a processor 2410, a communications interface 2420, a memory 2430, and a communications bus 2440. The processor 2410, the communications interface 2420, and the memory 2430 communicate with each other via the communications bus 2440. The processor 2410 may call the logic instructions in the memory 2430 to execute the various steps of the method for three-dimensional registration of neutron / X-ray heterogeneous images of a cylindrical multi-layer structure. Furthermore, the logic instructions in the memory 2430 may be implemented as software functional units and, when sold or used as a standalone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or the portion of the technical solution, may be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (such as a personal computer, server, or network device) to execute all or part of the steps of the method described in various embodiments of the present invention. The aforementioned storage media include: USB flash drives, mobile hard drives, read-only memories (ROM), random access memories (RAM), magnetic disks or optical disks, and other media that can store program codes.

[0078] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the various steps of the above-mentioned cylindrical multi-layer structure neutron / X-ray heterogeneous image three-dimensional registration method.

[0079] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the various steps of the above-mentioned cylindrical multi-layer structure neutron / X-ray heterogeneous image three-dimensional registration method.

[0080] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0081] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A 3D registration method for neutron / X-ray heterogeneous images of cylindrical multi-layer structures, characterized in that: include: Edge extraction operator and PCA principal component analysis are used to adjust the verticality of the image pairs; The edge contour extraction module determines the height map of the heterogeneous tomographic image; Use circle detection algorithm to detect the circle center and determine the translation amount to achieve translation alignment of heterogeneous tomographic images; According to the scale ratio, an image super-resolution reconstruction algorithm is used to realize the scale transformation super-resolution reconstruction of the tomographic image; Image rotation matching is achieved based on feature tomography and phase cross-correlation methods.

2. The heterogeneous image 3D registration method according to claim 1, characterized in that: The verticality adjustment is implemented based on the projection image; the image pair to be registered is a neutron projection image and an X-ray projection image; the verticality adjustment requires grayscale normalization of the input image pair, normalizing the grayscale value to the range of [0, 1]; for the normalized projection image, the OSTU algorithm is used to perform threshold segmentation on the projection image to extract the edge contour of the projection image; The PCA principal component analysis means: firstly counting the coordinates of the boundary pixels, and then using PCA to determine that the first principal component is the direction of the symmetry axis.

3. The heterogeneous image 3D registration method according to claim 2, characterized in that: The key to the threshold segmentation of OSTU algorithm is to determine the segmentation threshold. By selecting a segmentation threshold T, the projected image is divided into two parts: foreground and background, so that the inter-class variance between the foreground and background is Maximum; for the acquired edge contour image, the Sobel operator is used to extract the vertical boundary of the foreground area.

4. The heterogeneous image 3D registration method according to claim 2, characterized in that: The edge contour extraction module determines the purpose of the height mapping of heterogeneous tomographic images to map the neutron and X-ray tomographic images to the same height; the height mapping determines the height corresponding to the head and tail tomographic images of the cylindrical multi-layer structure by determining the horizontal boundary of the projected image.

5. The heterogeneous image 3D registration method according to claim 4, characterized in that: The height mapping method is: using an adaptive edge extraction operator to detect the horizontal boundary of the projected image, sorting the pixels at the horizontal boundary, and taking the average of the row coordinates of the first 20% of the pixels as the height of the horizontal boundary.

6. The heterogeneous image 3D registration method according to claim 3, characterized in that: Before edge detection, the input tomographic image pair is Gaussian smoothed using a Gaussian smoothing template, and then edge detection is performed. After edge detection, the holes of the tomographic image are filled based on the region growing principle. After the hole filling is completed, the center of the circle is detected using the differentiable Lie group Hough circle detection algorithm. The coordinates of the center of the circle and the diameter of the circle can be calculated based on the regional connectivity. The translation amount is determined according to the position of the center of the circle. After the centers of the heterogeneous tomographic images are all moved to the center of the circle, the translation alignment of the heterogeneous tomographic images is completed.

7. The heterogeneous image 3D registration method according to claim 3, characterized in that: The image super-resolution reconstruction algorithm includes using sinusoidal graph stacking to reduce noise, adaptive histogram equalization to improve image clarity, homomorphic filtering to sharpen details, bilinear interpolation algorithm upsampling, and homomorphic filtering to enhance contrast, thereby achieving super-resolution reconstruction and scale alignment of tomographic images.

8. The heterogeneous image 3D registration method according to claim 3, characterized in that: The tomographic image of the top shell area of ​​the cylindrical multilayer structure is used as the registration object. After the rotation angle is calculated, all the tomographic images are rotationally registered. The gradient of the tomographic image is first calculated and the central area of ​​the tomographic image is selectively retained. The phase correlation method is used to match the heterogeneous gradient images.

9. An electronic device comprising: A processor, a communication interface and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; characterized in that the processor calls the logic instructions in the memory to execute the various steps of the cylindrical multi-layer structure neutron / X-ray heterogeneous image three-dimensional registration method as described in any one of claims 1-8.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the computer program is implemented to perform the various steps of the method for three-dimensional registration of neutron / X-ray heterogeneous images of a cylindrical multi-layer structure as described in any one of claims 1 to 8.

Images