A GPU-based lithium battery pole piece CT image reconstruction system and method

Abstract

The invention discloses a GPU-based lithium battery pole piece CT image reconstruction system and method. The system comprises a CT device, a CPU device and at least two GPU devices; the CT device collects images of the lithium battery pole piece at multiple angles to obtain multiple X-ray transmission images at different angles; at least two GPU devices respectively establish n streams; at least two GPU devices receive multiple X-ray transmission images in parallel, each stream receives an X-ray transmission image, projection values ​​of the X-ray transmission images are calculated in each stream to obtain a positively correlated projection image of the thickness of the lithium battery pole piece, the positively correlated projection image is convoluted to obtain a convoluted image, the convoluted image is back-projected to obtain a reconstructed CT image result image; the system can effectively shorten the time of CT detection of lithium battery pole pieces.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery pole piece detection, and in particular to a GPU-based lithium battery pole piece CT image reconstruction system and method. Background Art

[0002] Computed tomography (CT) technology is mainly used in the medical field to detect the internal health status of the human body. With the development of related technologies, especially the resolution of radiation sources and detection devices, micron-level X-ray computed tomography technology has been gradually applied in geological exploration, biomimetic and material science. The application of lithium-ion battery cell technology has matured and is more or less at the end of its development. The next obstacle facing manufacturers is to achieve the reliability and high quality of these batteries in large-scale production and reduce costs. Use 3D CT technology for more detailed inspections and more accurate evaluation of batteries. This can complete the test quickly and reduce the failure rate of fault tests and waste. CT is able to analyze the various internal structural characteristics of lithium-ion power battery cells, and analyzes the characteristics of problems such as electrode fracture, electrode wrinkles, pole piece alignment and internal foreign matter in computed tomography images. For example, patent document CN114062403A discloses a CT device for scanning and detecting new energy batteries based on X-rays, including a base, a radiation source assembly installed on the base, a carrier platform assembly and a detector assembly, the carrier platform assembly is located between the radiation source assembly and the detector assembly, the radiation source assembly adjusts the position of the radiation source through a radiation source Z-axis moving module, the carrier platform assembly adjusts the position of the new energy battery through a carrier platform X-axis moving module, a carrier platform Y-axis moving module and an R-axis rotation module, and the detector assembly adjusts the position of the detector through a detector X-axis moving module, a detector Y-axis moving module and a detector Z-axis moving module.

[0003] CT technology is used to detect battery cells, usually offline or online sampling. This is because the CT detection time is too long and cannot keep up with the pace of the production line. Summary of the invention

[0004] The present invention provides a GPU-based lithium battery pole piece CT image reconstruction system and method, which can effectively shorten the time of lithium battery pole piece CT detection, thereby realizing online CT detection of lithium battery pole pieces.

[0005] A GPU-based lithium battery pole piece CT image reconstruction system, comprising a CT device, a CPU device and at least two GPU devices;

[0006] The CT device collects images of the lithium battery pole piece at multiple angles to obtain multiple X-ray transmission images at different angles; the at least two GPU devices respectively establish n streams; at least two GPU devices receive the multiple X-ray transmission images in parallel, each stream receives an X-ray transmission image, and in each stream, projection values ​​of the X-ray transmission images are calculated to obtain a positively correlated projection image of the thickness of the lithium battery pole piece, the positively correlated projection image is convolved to obtain a convolution image, the convolution image is back-projected to obtain an intermediate result image, the intermediate result images obtained from each stream are superimposed using a reduction algorithm to obtain a reconstructed CT image result image, and the reconstructed CT image result image is output to the CPU device for storage and display.

[0007] Furthermore, the CT device collects 360-degree images of the lithium battery pole piece, and collects an X-ray transmission image at each angle.

[0008] Furthermore, the CPU device allocates page-locked memory for the reconstructed CT image result map when initialized, and the at least two GPU devices allocate memory for the input multiple X-ray transmission images and intermediate variables in the reconstruction process when initialized.

[0009] Further, calculating the projection value of the X-ray transmission image includes:

[0010] Acquire transmission data of X-rays penetrating the lithium battery pole piece at various angles through a flat panel detector of a CT device;

[0011] According to the transmission data of the X-ray, the attenuation of the X-ray penetrating the lithium battery pole piece is calculated, and according to the attenuation, the thickness integral of the lithium battery pole piece at different angles is calculated;

[0012] According to the thickness integral of the lithium battery pole piece at different angles, the projection value of the X-ray transmission image of the lithium battery pole piece is calculated.

[0013] Furthermore, the projection value is calculated by the following formula:

[0014] b = u * lgP;

[0015] Among them, u is the absorption coefficient of the lithium battery electrode, b is the projection value, and P is the thickness integral.

[0016] Further, performing convolution processing on the positive correlation projection image includes:

[0017] The positive correlation projection image is convolved in the real domain according to the convolution kernel function to remove the superposition effect of low-frequency signals.

[0018] Furthermore, the convolution kernel function is as follows:

[0019] ;

[0020] Wherein, w is the width of the positively correlated projection image, where 0 ≤ i < w and i is an integer, and H(i) is the convolution kernel function.

[0021] Furthermore, perform back-projection on the convolution image to obtain an intermediate result image, including:

[0022] Duplicate and amplify each row of pixels of the convolution image to obtain an amplified image;

[0023] According to the shooting angle of the convolution image, perform an affine transformation on the amplified image to obtain a rotated image;

[0024] Add the pixels at the same position of the currently obtained rotated image and the previous rotated image to obtain the intermediate result image.

[0025] Furthermore, stack the intermediate result images obtained by each stream using a reduction algorithm, including:

[0026] In each round of calculation, perform pairwise parallel stacking of the pixel values at the same position in each intermediate result image until all pixel values at the same position are stacked, and then end the calculation to obtain the reconstructed CT image result.

[0027] A method for reconstructing CT images of lithium battery electrode sheets based on GPU using the above system, including:

[0028] The CT device collects images of the lithium battery electrode sheet at multiple angles to obtain multiple X-ray transmission images at different angles;

[0029] At least two GPU devices respectively establish n streams;

[0030] At least two GPU devices receive the multiple X-ray transmission images in parallel, and each stream receives one X-ray transmission image;

[0031] In each stream, calculate the projection value of the X-ray transmission image to obtain a positively correlated projection image of the thickness of the lithium battery electrode sheet, perform convolution processing on the positively correlated projection image to obtain a convolution image, and perform back-projection on the convolution image to obtain an intermediate result image;

[0032] Stack the intermediate result images obtained by each stream using a reduction algorithm to obtain the reconstructed CT image result, and output it to the CPU device for storage and display.

[0033] Further, the CT device performs 360-degree image acquisition on the lithium battery electrode sheet, and one X-ray transmission image is acquired at each angle.

[0034] Further, before the CT device performs image acquisition on the lithium battery electrode sheet at multiple angles, it further includes:

[0035] When the CPU device is initialized, it allocates pinned memory for the reconstructed CT image result map, and when the at least two GPU devices are initialized, they allocate memory for the input multiple X-ray transmission images and intermediate variables during the reconstruction process.

[0036] Further, calculating the projection value of the X-ray transmission image includes:

[0037] Obtaining the transmission data of the X-ray penetrating the lithium battery electrode sheet at each angle through the flat panel detector of the CT device;

[0038] According to the transmission data of the X-ray, calculating the attenuation amount of the X-ray penetrating the lithium battery electrode sheet, and calculating the thickness integral of the lithium battery electrode sheet at different angles according to the attenuation amount;

[0039] Calculating and obtaining the projection value of the X-ray transmission image of the lithium battery electrode sheet according to the thickness integral of the lithium battery electrode sheet at different angles.

[0040] Further, the projection value is calculated by the following formula:

[0041] b = u * lgP;

[0042] Wherein, u is the absorption coefficient of the lithium battery electrode sheet, b is the projection value, and P is the thickness integral.

[0043] Further, performing convolution processing on the positively correlated projection image includes:

[0044] Performing convolution processing on the positively correlated projection image in the real domain according to the convolution kernel function to remove the low-frequency signal superposition effect.

[0045] Further, the convolution kernel function is as follows:

[0046] ;

[0047] Wherein, w is the width of the positively correlated projection image, where 0 ≤ i < w and i is an integer, and H(i) is the convolution kernel function.

[0048] Further, performing backprojection on the convolution image to obtain an intermediate result map includes:

[0049] Copying and amplifying each row of pixels of the convolution image to obtain an amplified image;

[0050] According to the shooting angle of the convolution image, performing affine transformation on the amplified image to obtain a rotated image;

[0051] The pixels at the same position of the currently obtained rotated image and the previous rotated image are added to obtain the intermediate result image.

[0052] Furthermore, the intermediate result graphs obtained by each stream are superimposed using a reduction algorithm, including:

[0053] In each round of calculation, the pixel values ​​at the same position in each intermediate result image are superimposed in parallel, until all pixel values ​​at the same position are superimposed, and the calculation is terminated to obtain the reconstructed CT image result image.

[0054] The GPU-based lithium battery pole piece CT image reconstruction system and method provided by the present invention have at least the following beneficial effects:

[0055] (1) The X-ray transmission image is transmitted to the GPU, and a stream is allocated to the GPU for CT image reconstruction. After the result is obtained through calculation, it is output to the CPU, which fully utilizes the performance of the GPU and can effectively shorten the CT detection time of lithium battery pole pieces, making online CT detection of lithium battery pole pieces possible;

[0056] (2) In the image reconstruction process, the projection value is first calculated to obtain a positive correlation image of the battery electrode thickness, and then a convolution process is performed, which can effectively remove the superposition effect of low-frequency signals in the image, making the image clearer after subsequent back-projection processing;

[0057] (3) Using GPU to superimpose the intermediate result images using a reduction algorithm can effectively increase the image output speed and further shorten the CT detection time of lithium battery pole pieces:

[0058] (4) When the CPU device is initialized, it allocates page-locked memory for the reconstructed CT image result map. When the GPU device is initialized, it allocates memory for the input multiple X-ray transmission images and the intermediate variables in the reconstruction process, which can effectively reduce the detection time. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A structural schematic diagram of an embodiment of a GPU-based lithium battery pole piece CT image reconstruction system provided by the present invention.

[0060] Figure 2 A schematic diagram of stream allocation in an application scenario of the GPU-based lithium battery pole piece CT image reconstruction system provided by the present invention.

[0061] Figure 3A flowchart of an embodiment of a GPU-based lithium battery pole piece CT image reconstruction method provided by the present invention.

[0062] Figure 4 A flowchart of an embodiment of image reconstruction of a GPU-based lithium battery pole piece CT image reconstruction method provided by the present invention. DETAILED DESCRIPTION

[0063] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0064] refer to Figure 1 In some embodiments, a GPU-based lithium battery pole piece CT image reconstruction system is provided, comprising a CT device 101, a CPU device 102, and at least two GPU devices 103;

[0065] The CT device 101 collects images of the lithium battery pole piece at multiple angles to obtain multiple X-ray transmission images at different angles; at least two GPU devices 102 respectively establish n streams; at least two GPU devices 102 receive the multiple X-ray transmission images in parallel, each stream receives an X-ray transmission image, and in each stream, the projection value of the X-ray transmission image is calculated to obtain a positively correlated projection image of the thickness of the lithium battery pole piece, the positively correlated projection image is convolved to obtain a convolution image, the convolution image is back-projected to obtain an intermediate result image, the intermediate result images obtained from each stream are superimposed using a reduction algorithm to obtain a reconstructed CT image result image, and the result is output to the CPU device for storage and display.

[0066] The CT device 101 performs 360-degree image acquisition on the lithium battery electrode, and acquires an X-ray transmission image at each angle.

[0067] In some embodiments, the CPU device 102 allocates page-locked memory for the reconstructed CT image result map when initialized, and at least two GPU devices 103 allocate memory for the input multiple X-ray transmission images and intermediate variables in the reconstruction process when initialized.

[0068] Allocating memory on both the CPU and GPU devices takes time. In the CT reconstruction algorithm, due to the large number of input images and complex calculation process, more memory space is required. Before the memory is created, the processor's calculation core is idle. Therefore, creating memory during the calculation process will increase the detection time. Creating memory during initialization can effectively reduce the detection time.

[0069] Initialization memory allocation is mainly divided into the following parts:

[0070] 1. Allocate locked page memory of CPU device to reconstruct CT image result map. Due to the memory management characteristics of the operating system, memory pages will be replaced and locked page memory will be allocated to ensure that the memory address of CPU device memory is fixed when receiving GPU device data output.

[0071] 2. Allocate memory on the GPU device for the input X-ray transmission image.

[0072] 3. Allocate memory on the GPU device for the intermediate variables that need to be stored in the middle of the algorithm, including positive correlation projection images, convolution images, intermediate result images, etc.

[0073] Furthermore, n streams are respectively established in at least two GPU devices 102, and a total of m*n streams are established, where m is the number of GPU devices 102, and the number n is determined according to the shooting speed and the model of the GPU device.

[0074] Furthermore, after obtaining a plurality of X-ray transmission images at different angles collected by the CT device 101, they are sent to the GPU device.

[0075] In some application scenarios, refer to Figure 2 , taking two GPU devices as an example, create 2n streams, which are evenly located on two different GPU devices 102. Even numbers are located on GPU device 0, and odd numbers are located on GPU device 1. The number n is determined according to the shooting speed and GPU device model. The i-th (0≤i<2n) stream receives the i+2n*k-th X-ray transmission image starting from 0 degrees, where k is a natural number. After each X-ray transmission image is taken, it is immediately passed to the stream for processing. Using multiple GPU devices can multiply the computing performance. Using multiple streams can make full use of the efficiency of the GPU's copy engine and kernel function engine.

[0076] A stream (CUDA stream) represents a GPU operation queue, and the operations in the queue will be executed in the order in which they are added to the stream. A stream can be regarded as a task on a GPU device, and different tasks can be executed in parallel. Usually a task on a GPU device can be divided into three parts: copying data from the CPU device memory to the GPU device, running the kernel function on the GPU device, and copying the results of the GPU device to the CPU device. The copying of data and the running of the kernel function can be parallel, that is, the GPU device can perform the reconstruction of the CT image and the copying of data in parallel. Using multiple streams to make data copying and kernel functions parallel, improve the utilization of memory and core processors, and shorten the detection time.

[0077] CUDA (Compute Unified Device Architecture) It allows programmers to think in terms of memory and operations like traditional CPU programming. CUDA provides a simpler way to generate parallel code for execution on the GPU. Each GPU device is an independent device that runs asynchronously with the CPU device, which means that the CPU device and all GPU devices can perform calculations at the same time. Except for a few devices that share display memory with the CPU device, the GPU has its own physical memory. Although the size of GPU memory is limited, it can provide higher memory bandwidth than traditional CPU memory. CUDA kernels are functions that can be called from the CPU and executed asynchronously on CUDA devices, which means that the CPU queues the kernel and executes it only on the GPU device, does not wait for the kernel to complete, but continues to perform other work. Although the CUDA kernel cannot return a value due to the asynchronous mechanism, CUDA also provides some synchronization interfaces so that the host can determine when the kernel or pipeline is completed.

[0078] Furthermore, in each stream, the projection value of the X-ray transmission image is calculated, including:

[0079] Acquire transmission data of X-rays penetrating the lithium battery pole piece at various angles through a flat panel detector of a CT device;

[0080] According to the transmission data of the X-ray, the attenuation of the X-ray penetrating the lithium battery pole piece is calculated, and according to the attenuation, the thickness integral of the lithium battery pole piece at different angles is calculated;

[0081] According to the thickness integral of the lithium battery pole piece at different angles, the projection value of the X-ray transmission image of the lithium battery pole piece is calculated.

[0082] The projection value is calculated by the following formula:

[0083] b = u * lgP; (1)

[0084] Among them, u is the absorption coefficient of the lithium battery electrode, b is the projection value, and P is the thickness integral.

[0085] The absorption coefficient of the lithium battery pole piece is determined by its material and is expressed in mol / L or g / L. The transmission data is the cumulative result of the flat panel detector receiving X-rays within a certain integration time. These data are generally output in the form of images or matrices, and this part of the data is obtained through the X-ray transmission diagram. The thickness integral is obtained by integrating the attenuation.

[0086] The projection value of each pixel obtained by calculation is the pixel value of the positive correlation projection image of the lithium battery electrode thickness.

[0087] Further, the positive correlation projection image is subjected to convolution processing, including:

[0088] Perform convolution processing on the positively correlated projection image in the real domain according to the convolution kernel function to remove the low-frequency signal superposition effect.

[0089] The image reconstructed by the direct back-projection method is blurred. Although this blurred image can reproduce the original density function after correction, the correction process is very time-consuming. This is the problem of the back-projection first and then correction reconstruction method. Filtering is to perform high-pass filtering on the frequency-domain image to eliminate the influence brought by superposition during back-projection. Converting the image to the frequency domain, filtering and then converting back to the frequency domain is too time-consuming. According to the Fourier transform theory, multiplying in the frequency domain is equivalent to convolving in the real domain. Convert the R-L filter kernel into a convolution kernel in the real domain to reduce the influence brought by back-projection superposition.

[0090] The convolution kernel is as follows:

[0091] ; (2)

[0092] where w is the width of the positively correlated projection image, where 0 ≤ i < w and i is an integer, and H(i) is the convolution kernel function.

[0093] Further, perform back-projection on the convolution image to obtain an intermediate result image, including:

[0094] Duplicate and amplify each row of pixels of the convolution image to obtain an amplified image;

[0095] According to the shooting angle of the convolution image, perform affine transformation on the amplified image to obtain a rotated image;

[0096] Add the pixels at the same position of the currently obtained rotated image and the previous rotated image to obtain the intermediate result image.

[0097] Among them, duplicating and amplifying each row of pixels of the convolution image means amplifying the width and height of the convolution image to a preset multiple respectively.

[0098] Specifically, back-projection is a summation process that adds up all the data at all angles. It "smears" the data in the projection domain back along the original projection path without changing the value of the data. Duplicate and amplify each row of the image to max(h, w)*sqrt(2), where h and w are the height and width of the convolution image. According to the shooting angle of the current image, obtain the rotated image through affine transformation, and add and save it with the processing result of the same row of the previous picture.

[0099] Further, superimpose the intermediate result images obtained by each stream using a reduction algorithm, including:

[0100] In each round of calculation, the pixel values ​​at the same position in each intermediate result image are superimposed in parallel, until the calculation is terminated when all pixel values ​​at the same position are superimposed.

[0101] For example, for 360 intermediate result images at 360 angles, for the same position, in the first round of calculation, the corresponding pixel values ​​are added in parallel two by two to obtain 180 results. In the second round of calculation, the 180 results are added in parallel two by two again to obtain 90 results, and so on, until the calculation is terminated when all pixel values ​​at the same position are superimposed. In the last round, only one addition operation is required to obtain the final calculation result.

[0102] The first round requires N / 2 addition operations, the second round requires N / 4 addition operations, and the last round requires only 1 addition operation, for a total of log2(N) rounds of calculation. The total number of addition operations completed is N / 2+N / 4+N / 8+…+1=N-1. GPU devices can perform pairwise addition of different data in parallel, thus greatly improving the speed of image output.

[0103] refer to Figure 3 In some embodiments, a method for reconstructing a lithium battery pole piece CT image based on a GPU using the above system is provided, comprising:

[0104] S1. CT equipment collects images of lithium battery pole pieces at multiple angles to obtain multiple X-ray transmission images at different angles;

[0105] S2. At least two GPU devices establish n streams respectively;

[0106] S3. At least two GPU devices receive the multiple X-ray transmission images in parallel, and each stream receives one X-ray transmission image;

[0107] S4, calculating the projection value of the X-ray transmission image in each stream to obtain a positive correlation projection image of the thickness of the lithium battery electrode sheet, performing convolution processing on the positive correlation projection image to obtain a convolution image, and performing back-projection on the convolution image to obtain an intermediate result image;

[0108] S5. The intermediate result images obtained from each stream are superimposed using a reduction algorithm to obtain a reconstructed CT image result image, and the reconstructed CT image result image is output to the CPU device for storage and display.

[0109] Among them, in step S1, the CT equipment performs 360-degree image acquisition on the lithium battery electrode, and acquires an X-ray transmission image at each angle.

[0110] In addition, before step S1, it also includes: when the CPU device is initialized, a locked page memory is allocated for the reconstructed CT image result map; when the at least two GPU devices are initialized, a memory is allocated for the input multiple X-ray transmission images and intermediate variables in the reconstruction process.

[0111] Furthermore, in step S2, n streams are respectively established in at least two GPU devices 102, and a total of m*n streams are established, where m is the number of GPU devices 102, and the number n is determined according to the shooting speed and the model of the GPU device.

[0112] refer to Figure 4 In step S4, the X-ray transmission image is subjected to projection value calculation, including:

[0113] S41, obtaining transmission data of X-rays penetrating the lithium battery electrode at various angles through a flat panel detector of a CT device;

[0114] S42, calculating the attenuation of light penetrating the lithium battery pole piece according to the X-ray transmission data, and calculating the thickness integral of the lithium battery pole piece at different angles according to the attenuation;

[0115] S43, calculating and obtaining the projection value of the X-ray transmission image of the lithium battery electrode piece according to the thickness integral of the lithium battery electrode piece at different angles;

[0116] Please refer to formula (1) for the calculation formula of the projection value.

[0117] Further, in step S4, the positive correlation projection image is subjected to convolution processing, including:

[0118] S44, performing convolution processing on the positive correlation projection image in the real domain according to the convolution kernel function to remove the superposition effect of low-frequency signals.

[0119] Please refer to formula (2) for the convolution kernel function.

[0120] Furthermore, in step S4, the convolution image is back-projected to obtain an intermediate result image, including:

[0121] S45, copying and amplifying each row of pixels of the convolution image to obtain an amplified image;

[0122] S46, performing affine transformation on the amplified image according to the shooting angle of the convolution image to obtain a rotated image;

[0123] S47, adding the pixels at the same position of the currently obtained rotated image and the previous rotated image to obtain the intermediate result image.

[0124] Furthermore, in step S5, the intermediate result graphs obtained by each stream are superimposed using a reduction algorithm, including:

[0125] In each round of calculation, the pixel values ​​at the same position in each intermediate result image are superimposed in parallel, until all pixel values ​​at the same position are superimposed, and the calculation is terminated to obtain the reconstructed CT image result image.

[0126] Please refer to the above-mentioned embodiment for the specific calculation method, which will not be described in detail here.

[0127] The GPU-based lithium battery pole piece CT image reconstruction system and method provided by the present invention have at least the following beneficial effects:

[0128] (1) The X-ray transmission image is transmitted to the GPU, and a stream is allocated to the GPU for CT image reconstruction. After the result is obtained through calculation, it is output to the CPU, which fully utilizes the performance of the GPU and can effectively shorten the CT detection time of lithium battery pole pieces, making online CT detection of lithium battery pole pieces possible;

[0129] (2) In the image reconstruction process, the projection value is first calculated to obtain a positive correlation image of the battery electrode thickness, and then a convolution process is performed, which can effectively remove the superposition effect of low-frequency signals in the image, making the image clearer after subsequent back-projection processing;

[0130] (3) Using GPU to superimpose the intermediate result images using a reduction algorithm can effectively increase the image output speed and further shorten the CT detection time of lithium battery pole pieces:

[0131] (4) When the CPU device is initialized, it allocates page-locked memory for the reconstructed CT image result map. When the GPU device is initialized, it allocates memory for the input multiple X-ray transmission images and the intermediate variables in the reconstruction process, which can effectively reduce the detection time.

[0132] Although preferred embodiments of the present invention have been described, additional changes and modifications may be made to these embodiments by those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A GPU-based lithium battery pole piece CT image reconstruction system, characterized in that: It includes a CT device, a CPU device, and at least two GPU devices; The CT device performs image acquisition of the lithium battery electrode sheet at multiple angles to obtain multiple X-ray transmission images at different angles; each of the at least two GPU devices has established n streams; The at least two GPU devices receive the multiple X-ray transmission images in parallel, each stream receives one X-ray transmission image, calculates the positive correlation projection image of the lithium battery electrode sheet thickness by performing projection value calculation on the X-ray transmission image in each stream, performs convolution processing on the positive correlation projection image to obtain a convolution image, performs backprojection on the convolution image to obtain an intermediate result image, superimposes the intermediate result images obtained by each stream using a reduction algorithm to obtain a reconstructed CT image result image, and outputs it to the CPU device for storage and display; Performing backprojection on the convolution image to obtain an intermediate result image includes: Copying and amplifying each row of pixels of the convolution image to obtain an amplified image; Performing affine transformation on the amplified image according to the shooting angle of the convolution image to obtain a rotated image; Adding the pixels at the same position of the currently obtained rotated image and the previous rotated image to obtain the intermediate result image.

2. The system according to claim 1, characterized in that The CT device performs 360-degree image acquisition on the lithium battery electrode sheet, and one X-ray transmission image is acquired at each angle.

3. The system according to claim 1, characterized in that When initializing, the CPU device allocates pinned memory for the reconstructed CT image result image, and when initializing, the at least two GPU devices allocate memory for the input multiple X-ray transmission images and intermediate variables during the reconstruction process.

4. The system according to claim 1, characterized in that Performing projection value calculation on the X-ray transmission image includes: Obtaining the transmission data of the X-ray penetrating the lithium battery electrode sheet at each angle through the flat panel detector of the CT device; Calculating the attenuation amount of the X-ray penetrating the lithium battery electrode sheet according to the transmission data of the X-ray, and calculating the thickness integral of the lithium battery electrode sheet at different angles according to the attenuation amount; Calculating the projection value of the X-ray transmission image of the lithium battery electrode sheet according to the thickness integral of the lithium battery electrode sheet at different angles.

5. The system according to claim 4, characterized in that The projection value is calculated by the following formula: b = u * lgP; Where u is the absorption coefficient of the lithium battery electrode sheet, b is the projection value, and P is the thickness integral.

6. The system according to claim 1 or 4, characterized in that: Performing convolution processing on the positive correlation projection image includes: Performing convolution processing on the positive correlation projection image in the real domain according to the convolution kernel function to remove the low-frequency signal superposition effect.

7. The system according to claim 6, characterized in that The convolution kernel function is as follows: Where w is the width of the positive correlation projection image, where 0 ≤ i < w and i is an integer, and H(i) is the convolution kernel function.

8. The system according to claim 1, characterized in that Superimposing the intermediate result images obtained by each stream using a reduction algorithm includes: In each round of calculation, the pixel values at the same position in each intermediate result image are superimposed pairwise in parallel until all the pixel values at the same position are superimposed, and then the calculation ends to obtain a reconstructed CT image result image.

9. A method for reconstructing CT images of lithium battery pole pieces based on a GPU using the system as claimed in any one of claims 1 to 8, characterized in that: It includes: The CT device performs image acquisition of the lithium battery electrode sheet at multiple angles to obtain multiple X-ray transmission images at different angles; At least two GPU devices establish n streams respectively; At least two GPU devices receive the multiple X-ray transmission images in parallel, and each stream receives one X-ray transmission image; In each stream, the X-ray transmission image is projected and calculated to obtain a positive correlation projection image of the thickness of the lithium battery electrode sheet, the positive correlation projection image is convolved to obtain a convolution image, and the convolution image is back-projected to obtain an intermediate result image; The intermediate result images obtained from each stream are superimposed using a reduction algorithm to obtain a reconstructed CT image result image, and the reconstructed CT image result image is output to the CPU device for storage and display; Back-projecting the convolution image to obtain an intermediate result image includes: Copy and amplify each row of pixels of the convolution image to obtain an amplified image; According to the shooting angle of the convolution image, performing affine transformation on the amplified image to obtain a rotated image; The pixels at the same position of the currently obtained rotated image and the previous rotated image are added to obtain the intermediate result image.

Images