Local-reducible compression method and device for industrial CT (Computed Tomography) data

Abstract

The invention relates to a compression method capable of local reduction for industrial CT data, and the method comprises the steps: traversing a whole preset product, obtaining the pixel value of each pixel point in each section of the preset product, and obtaining the to-be-compressed three-dimensional data of the preset product; and inputting the to-be-compressed three-dimensional data into a neural network to obtain target three-dimensional data after the preset product is compressed. The method has the beneficial effects that when the local data is restored, the local information of the specified position can be directly restored from the compressed data, the requirement is quickly met, and the efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of CT image processing, and in particular to a compression method and device for industrial CT data capable of being partially restored. Background Art

[0002] With the widespread application of CT technology and the continuous improvement of scanning accuracy, the demand for efficient and high-quality CT data compression technology is becoming increasingly urgent. Industrial CT images usually contain more structural details and texture information, which places higher requirements on compression algorithms.

[0003] However, although the current industrial CT data compression technology can achieve a high compression rate, it will lose some image quality and is not suitable for application scenarios such as industrial CT that require high-precision measurement.

[0004] Based on this technology, CN104244012A-A CT data compression method, the data compression method includes the following steps: original image data acquisition, caching the data collected by the detector; differential encoding, differential encoding the cached data; area division, dividing the data matrix into N small matrices of equal size; comparison and selection, comparing and screening the divided data blocks with the previous data blocks; data compression, secondary compression of the parts with large data differences; for large-scale data acquisition and transmission of industrial CT, the invention provides a lossless compression method for industrial CT projection raw data, which compresses the data in a CT device using a high-resolution large-area flat-panel detector, which can effectively improve the transmission efficiency and data compression ratio, but the compression algorithm of the scheme is to compress the data as a whole, and the data of each cross section needs to be completely restored during decompression. In actual applications, it is generally only necessary to pay attention to the local situation of the workpiece to check whether there are defects such as cracks, pores, looseness, inclusions, etc., and when restoring local data, the scheme needs to restore all compressed data first, and then intercept the required local information, which increases the engineering workload and is very inefficient. Summary of the invention

[0005] The main purpose of the present invention is to provide a compression method and device for industrial CT data that can be partially restored, so that when local data is restored, local information at a specified location can be directly restored from the compressed data, quickly meeting the needs and improving efficiency.

[0006] To achieve the above object, the present invention provides a compression method for industrial CT data that can be partially restored, comprising the following steps:

[0007] Traversing the entire preset product, obtaining the pixel value of each pixel point in each cross section of the preset product, thereby obtaining the three-dimensional data to be compressed of the preset product;

[0008] The three-dimensional data to be compressed is input into a neural network to obtain target three-dimensional data of the preset product after compression.

[0009] Furthermore, the step of inputting the three-dimensional data to be compressed into a neural network to obtain the target three-dimensional data after compression of the preset product includes:

[0010] Presetting the size of target three-dimensional data after the three-dimensional data to be compressed is performed, and setting the structure of the neural network according to the size of the target three-dimensional data and the three-dimensional data to be compressed to obtain an initial neural network;

[0011] Randomly initializing the weights of the initial neural network to obtain a temporary neural network;

[0012] The three-dimensional data to be compressed is input into the temporary neural network to obtain compressed target three-dimensional data.

[0013] Furthermore, the step of inputting the to-be-compressed three-dimensional data into the temporary neural network to obtain compressed target three-dimensional data comprises:

[0014] A. Randomly extract a batch of simulation coordinates, pass them into the temporary neural network, perform forward propagation, and obtain the simulation values ​​of the data under the simulation coordinates;

[0015] B. Comparing the simulated value with the real value of the three-dimensional data to be compressed at the corresponding coordinates and calculating the loss, and obtaining the gradient direction by calculating the gradient through the loss function;

[0016] C. updating the weight of the temporary neural network according to the reverse direction of the gradient direction and the current learning rate of the temporary neural network to obtain a final neural network;

[0017] D. Repeat steps AC until the set number of training rounds is completed to obtain compressed target three-dimensional data.

[0018] Furthermore, after the step of inputting the three-dimensional data to be compressed into the neural network to obtain the target three-dimensional data after compression of the preset product, it also includes:

[0019] Determine whether information for performing partial restoration is received;

[0020] If the information of partial restoration is received, the coordinates of the partial area to be restored in the preset product are determined;

[0021] The coordinates are input into the neural network and the local restoration area is obtained through calculation.

[0022] Furthermore, the step of inputting the coordinates into the neural network to obtain the local restoration area through calculation includes:

[0023] Inputting the coordinates of each pixel point in the local area to be restored into the neural network one by one;

[0024] Calculate the restored pixel value of each pixel point in the local area to be restored by forward propagation of the neural network;

[0025] The restored pixel values ​​are coordinate spliced, and the image model is reconstructed based on the spliced ​​coordinates, thereby obtaining image information of the local restored area.

[0026] The present invention also provides a compression device capable of partial restoration of industrial CT data, comprising:

[0027] A traversal module, used to traverse the entire preset product, obtain the pixel value of each pixel point in each cross section of the preset product, so as to obtain the three-dimensional data to be compressed of the preset product;

[0028] The input module is used to input the three-dimensional data to be compressed into the neural network to obtain the target three-dimensional data after compression of the preset product.

[0029] Wherein, the input module includes:

[0030] A neural network calculation submodule, used to pre-set the size of the target three-dimensional data after the three-dimensional data to be compressed is compressed, and to set the structure of the neural network according to the size of the target three-dimensional data and the three-dimensional data to be compressed, so as to obtain an initial neural network;

[0031] A weight initialization submodule, used to randomly initialize the weights of the initial neural network to obtain a temporary neural network;

[0032] The data compression submodule is used to input the three-dimensional data to be compressed into the temporary neural network to obtain compressed target three-dimensional data.

[0033] Wherein, the compression submodule includes:

[0034] A coordinate processing unit, used for randomly extracting a batch of simulation coordinates, transferring them into the temporary neural network, performing forward propagation, and obtaining simulation values ​​of the data under the simulation coordinates;

[0035] A coordinate calculation unit, used for comparing the simulated value with the real value of the three-dimensional data to be compressed at the corresponding coordinate and calculating the loss, and obtaining the gradient direction by calculating the gradient through the loss function;

[0036] A neural network updating unit, used for updating the weight of the temporary neural network according to the reverse direction of the gradient direction and the current learning rate of the temporary neural network to obtain a final neural network;

[0037] The data compression unit is used to repeat the steps of the coordinate processing unit, the coordinate calculation unit, and the neural network updating unit until a set number of training rounds are completed to obtain compressed target three-dimensional data.

[0038] A compression device capable of partial restoration of industrial CT data, further comprising:

[0039] A receiving module, used to determine whether information for performing partial restoration is received;

[0040] A judgment module, configured to receive information on partial restoration and determine the coordinates of the partial area to be restored in the preset product;

[0041] The restoration module inputs the coordinates into the neural network and obtains the local restoration area through calculation.

[0042] The restoration module comprises:

[0043] A coordinate input submodule, used to input the coordinates of each pixel point in the local area to be restored into the neural network one by one;

[0044] A pixel calculation submodule, used for calculating the restored pixel value of each pixel point in the local area to be restored through the forward propagation of the neural network;

[0045] The coordinate stitching submodule is used to perform coordinate stitching on the restored pixel values ​​and reconstruct the image model based on the stitched coordinates, thereby obtaining image information of the local restored area.

[0046] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned compression method for industrial CT data that can be partially restored when executing the computer program.

[0047] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the above-mentioned compression method for industrial CT data that can be partially restored are implemented.

[0048] The specific beneficial effects of the locally restorable compression method for industrial CT data provided by the present invention are as follows:

[0049] (1) Compress and map industrial CT data through a preset neural network. Compared with traditional compression methods, neural networks can capture the intrinsic characteristics of data, achieve a more efficient compression rate, and reduce the cost of data storage and transmission;

[0050] (2) The present invention provides a coordinate-based local restoration function, which can accurately restore detailed information of a specific area from the compressed data when necessary, without decompressing the entire compressed data, greatly improving the flexibility and efficiency of data use;

[0051] (3) The neural network architecture used in the present invention can be adjusted according to the preset compression value, and the complexity of the compression process and the size of the compressed data can be affected by changing the number, type or parameters of the connection layers in the neural network; the user can select a suitable neural network configuration to achieve the desired compression ratio and freely control the size of the compressed data;

[0052] (4) Through iterative training of the neural network, the network weights are continuously optimized, and the original information of the data is retained as much as possible during the compression process, so that high-quality image information can be obtained during local restoration to meet the accuracy requirements of industrial detection, analysis and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a flowchart of a compression method for industrial CT data that can be partially restored in one embodiment of the present invention;

[0054] Figure 2 is a flow chart of a method for partially restoring industrial CT data in another embodiment of the present invention;

[0055] Figure 3 It is a structural block diagram of a compression device capable of partial restoration of industrial CT data in one embodiment of the present invention;

[0056] Figure 4 It is a schematic diagram of industrial CT data in the present invention;

[0057] Figure 5 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0059] The locally restorable compression method for industrial CT data provided by the present invention aims to realize the compression of industrial CT data while supporting local restoration, so as to quickly meet the needs of industrial inspection; at the same time, the size of the final compressed data can also be freely controlled, wherein the technical principle is:

[0060] Industrial CT data such as Figure 4As shown, the industrial CT data includes multiple sections, each of which can be regarded as a picture. We regard the long direction of the picture as x, the wide direction as y, and the vertical direction of different sections (i.e., the stacking direction) as the z direction. In this way, the industrial CT data can be regarded as a three-dimensional data; let the information stored in a certain pixel of a certain section in the industrial CT data be c (actually the color of the pixel), if we can obtain a mapping c=f(x,y,z) from x, y and z to c, we can calculate the color of each section and each pixel through multiple layers of loops, and restore the entire three-dimensional data, that is, the mapping itself already contains the information of the entire industrial CT data; considering that adjacent pixel values ​​are generally similar, it is better to choose a continuous function for the mapping, and the more parameters the mapping has, the stronger the expression ability, so this application uses a neural network as a mapping. In specific experiments, the use of a neural network is better than other methods of industrial CT storage.

[0061] Reference Figure 1 , a flow chart of a method for partially restoring compression of industrial CT data in one embodiment of the present invention includes the following steps:

[0062] S1: traverse the entire preset product, obtain the pixel value of each pixel point in each cross section of the preset product, and thus obtain the three-dimensional data to be compressed of the preset product;

[0063] S2: Inputting the three-dimensional data to be compressed into a neural network to obtain target three-dimensional data after compression of the preset product.

[0064] For step S1, specifically:

[0065] S11: Access each pixel point in each cross section of the preset product in a preset loop order;

[0066] S12: After an iterative access operation, pixel values ​​are calculated for each section and each pixel to obtain the three-dimensional data to be compressed of the preset product.

[0067] In the specific implementation process: based on the storage method of the three-dimensional data to be compressed in the storage space, access efficiency, processing algorithm requirements and other factors, the loop sequence is preset. In the specific implementation, starting from the first section of the preset product, the pixels in each section are traversed one by one, and after completing one section, it moves to the next section until all sections and pixels are traversed, and the position information of each pixel point (i.e., the coordinates of the pixel point in the three-dimensional space) is read at the same time; after the traversal is completed, the iterative loop operation is repeatedly performed until the traversal of the entire three-dimensional data to be compressed is completed, and each iteration accesses a pixel point according to the preset loop sequence, and performs the operation of calculating or confirming the pixel value; after the above iterative loop and pixel value calculation or confirmation process, the three-dimensional data to be compressed of the preset product is obtained.

[0068] For step S2, specifically:

[0069] S21: presetting the size of target three-dimensional data after the three-dimensional data to be compressed is compressed, and setting the structure of the neural network according to the size of the target three-dimensional data and the three-dimensional data to be compressed to obtain an initial neural network;

[0070] S22: Randomly initialize the weights of the initial neural network to obtain a temporary neural network;

[0071] S23: Inputting the three-dimensional data to be compressed into the temporary neural network to obtain compressed target three-dimensional data.

[0072] In the specific implementation process: first determine the size of the target three-dimensional data after the three-dimensional data to be compressed is compressed. In the specific experiment, the float32 format is selected to save the parameters. Each parameter occupies 4B of space. It is recommended to use 512 neurons per hidden layer, and the last hidden layer can be reduced due to the compression size limit. Considering that the input and output dimensions are both 3, the following formula can be obtained:

[0073] [(3+1)*512+(512+1)*512*(n-1)+(512+1)*y+(y+1)*3]*4 = compressed size

[0074] Where n is the number of hidden layers, y is the size of the last hidden layer. Taking the target size of 10MB as an example, we can calculate n=10, y=495, thereby determining the structure of the neural network.

[0075] A normal distribution with a mean of 0 and a standard deviation of a small value (such as 0.01) is selected to ensure that the weights are random and close to 0 before training begins; the weights of each connection layer of the neural network are independently randomly assigned; after completing the random assignment of weights, the neural network completes the initialization process and obtains the initial neural network; randomly assign values ​​in a normal distribution manner, initialize the weights of the neural network layer by layer, and obtain a temporary neural network; input the three-dimensional data to be compressed into the temporary neural network, and perform an iterative process through the existing network forward propagation of the neural network; each iteration sends the pixel data of the three-dimensional data to be compressed in batches to the network of the temporary neural network for compression, and generates corresponding output; after multiple iterations, the neural network converts the three-dimensional data to be compressed into compressed target three-dimensional data; the coordinate information of the three-dimensional data to be compressed is mapped to the compressed target three-dimensional data, and a unique coordinate is assigned to each pixel point for position marking.

[0076] For step S23, specifically:

[0077] A. randomly extract a batch of simulation coordinates, pass them into the temporary neural network, perform forward propagation, and obtain the simulation values ​​of the data under the simulation coordinates;

[0078] B. comparing the simulated value with the real value of the three-dimensional data to be compressed at the corresponding coordinates and calculating the loss, and obtaining the gradient direction by calculating the gradient through the loss function;

[0079] C. updating the weight of the temporary neural network according to the reverse direction of the gradient direction and the current learning rate of the temporary neural network to obtain the final neural network;

[0080] D. Repeat steps AC until the set number of training rounds is completed to obtain compressed target three-dimensional data.

[0081] In the specific implementation process: a batch of pixels (the batch size can be defined by yourself, generally 16 or 32) is randomly selected, each pixel is represented by a three-dimensional coordinate (x, y, z), and the coordinates of the selected batch of pixels are transmitted as input to the temporary neural network; in the forward propagation process of the temporary neural network, the input data (i.e., the coordinates of the selected batch of pixels) pass through the network of the temporary neural network and generate a predicted output value, which is very close to the actual pixel value in the three-dimensional data to be compressed, and the performance of the network is evaluated by calculating the difference between the predicted output value and the true value; the difference is measured by a certain form of loss function (such as mean square error, cross entropy, etc.), and the smaller the value of the loss function, the more accurate the network prediction; after calculating the loss function value, the back propagation algorithm is used to update the weight of the temporary neural network; the whole process (AC step) is repeated many times, each time a new batch of data is iterated until a preset number of rounds is reached, during which the weight of the neural network is gradually optimized, and the compressed target three-dimensional data is obtained through the network output.

[0082] Reference Figure 2 , a flowchart of a method for partially restoring industrial CT data in another embodiment of the present invention further includes the following steps:

[0083] S3: Determine whether information for partial restoration is received;

[0084] S4: If the information of partial restoration is received, the coordinates of the partial area to be restored in the preset product are determined;

[0085] S5: Input the coordinates into the neural network and obtain the local restoration area through calculation.

[0086] In the specific implementation process: the system checks whether it has received instructions from the user, other systems or internal logic, requiring the restoration of a local area of ​​a preset product; once the system confirms that a local restoration operation needs to be performed, the position of the local area to be restored in the preset product image is determined through the three-dimensional data to be compressed; after determining the coordinates of the local area to be restored, the system will use these coordinates as input data and send them to the trained neural network. The neural network calculates the pixel value that should be restored for each pixel point in the area through its complex internal calculation process (such as forward propagation), and finally obtains the restored pixel value of the entire local restoration area.

[0087] For step S5, specifically:

[0088] S51: Inputting the coordinates of each pixel point in the local area to be restored into the neural network one by one;

[0089] S52: Calculating the restored pixel value of each pixel point in the local area to be restored by forward propagation of the neural network;

[0090] S53: performing coordinate splicing on the restored pixel values, and reconstructing the image model based on the spliced ​​coordinates, thereby obtaining image information of the local restored area.

[0091] In the specific implementation process: the system takes the coordinates of the local area to be restored as input data and sends them to the trained neural network; after the neural network receives the pixel coordinates, it performs forward propagation through a series of internal computing layers (such as convolutional layer, pooling layer, fully connected layer, etc.); during the forward propagation process, the neural network calculates the restored pixel value of each pixel based on the learned weights and biases, as well as the input pixel coordinates of the local area to be restored; each restored pixel value is associated with its corresponding coordinates in the three-dimensional data to be compressed, and splicing is performed according to the position of each pixel in the local area to be restored in the three-dimensional data to be compressed, and the image model is reconstructed based on the spliced ​​coordinates, and finally the image information of the local restored area is presented.

[0092] Specific beneficial effects of the partially restorable compression method for industrial CT data provided by the embodiment of the present invention are as follows:

[0093] (1) Compress and map industrial CT data through a preset neural network. Compared with traditional compression methods, neural networks can capture the intrinsic characteristics of data, achieve a more efficient compression rate, and reduce the cost of data storage and transmission;

[0094] (2) The present invention provides a coordinate-based local restoration function, which can accurately restore detailed information of a specific area from the compressed data when necessary, without decompressing the entire compressed data, greatly improving the flexibility and efficiency of data use;

[0095] (3) The neural network architecture used in the present invention can be adjusted according to the preset compression value, and the complexity of the compression process and the size of the compressed data can be affected by changing the number, type or parameters of the connection layers in the neural network; the user can select a suitable neural network configuration to achieve the desired compression ratio and freely control the size of the compressed data;

[0096] (4) Through iterative training of the neural network, the network weights are continuously optimized, and the original information of the data is retained as much as possible during the compression process, so that high-quality image information can be obtained during local restoration to meet the accuracy requirements of industrial detection, analysis and other fields.

[0097] Reference Figure 3The embodiment of the present invention further provides a compression device capable of partial restoration of industrial CT data, comprising:

[0098] The traversal module 1 is used to traverse the entire preset product and obtain the pixel value of each pixel point in each cross section of the preset product, so as to obtain the three-dimensional data to be compressed of the preset product.

[0099] The input module 2 is used to input the three-dimensional data to be compressed into the neural network to obtain the target three-dimensional data after compression of the preset product.

[0100] Wherein, the input module 2 comprises:

[0101] A neural network calculation submodule, used to pre-set the size of the target three-dimensional data after the three-dimensional data to be compressed is compressed, and to set the structure of the neural network according to the size of the target three-dimensional data and the three-dimensional data to be compressed, so as to obtain an initial neural network;

[0102] A weight initialization submodule, used to randomly initialize the weights of the initial neural network to obtain a temporary neural network;

[0103] The data compression submodule is used to input the three-dimensional data to be compressed into the temporary neural network to obtain compressed target three-dimensional data.

[0104] Wherein, the compression submodule includes:

[0105] A coordinate processing unit, used for randomly extracting a batch of simulation coordinates, transferring them into the temporary neural network, performing forward propagation, and obtaining simulation values ​​of the data under the simulation coordinates;

[0106] A coordinate calculation unit, used for comparing the simulated value with the real value of the three-dimensional data to be compressed at the corresponding coordinate and calculating the loss, and obtaining the gradient direction by calculating the gradient through the loss function;

[0107] A neural network updating unit, used for updating the weight of the temporary neural network according to the reverse direction of the gradient direction and the current learning rate of the temporary neural network to obtain a final neural network;

[0108] The data compression unit is used to repeat the steps of the coordinate processing unit, the coordinate calculation unit, and the neural network updating unit until a set number of training rounds are completed to obtain compressed target three-dimensional data.

[0109] A compression device capable of partial restoration of industrial CT data, further comprising:

[0110] A receiving module, used to determine whether information for performing partial restoration is received;

[0111] A judgment module, configured to receive information on partial restoration and determine the coordinates of the partial area to be restored in the preset product;

[0112] The restoration module is used to input the coordinates into the neural network and obtain a local restoration area through calculation.

[0113] The restoration module comprises:

[0114] A coordinate input submodule, used to input the coordinates of each pixel point in the local area to be restored into the neural network one by one;

[0115] A pixel calculation submodule, used for calculating the restored pixel value of each pixel point in the local area to be restored through the forward propagation of the neural network;

[0116] The coordinate stitching submodule is used to perform coordinate stitching on the restored pixel values ​​and reconstruct the image model based on the stitched coordinates, thereby obtaining image information of the local restored area.

[0117] Reference Figure 5 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 5 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0118] Those skilled in the art will understand that Figure 5 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0119] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0120] In summary, the entire preset product is traversed to obtain the pixel value of each pixel in each cross section of the preset product, thereby obtaining the three-dimensional data to be compressed of the preset product; the three-dimensional data to be compressed is compressed and mapped by a preset neural network to obtain the target three-dimensional data of the preset product after compression; when the target three-dimensional data is needed to perform local restoration information, the coordinates of the pixel points in the preset product are determined, and input into the neural network to obtain the required local restoration information through calculation, so that when the local data is restored, the local information of the specified position can be directly restored from the compressed data, so as to quickly meet the needs and improve efficiency.

[0121] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0122] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0123] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A locally restorable compression method for industrial CT data, characterized in that: The following steps are involved: Traversing the entire preset product, obtaining the pixel value of each pixel point in each cross section of the preset product, thereby obtaining the three-dimensional data to be compressed of the preset product; The three-dimensional data to be compressed is input into a neural network to obtain target three-dimensional data of the preset product after compression.

2. The locally restorable compression method for industrial CT data according to claim 1, characterized in that: The step of inputting the three-dimensional data to be compressed into a neural network to obtain the target three-dimensional data after compression of the preset product includes: Presetting the size of target three-dimensional data after the three-dimensional data to be compressed is performed, and setting the structure of the neural network according to the size of the target three-dimensional data and the three-dimensional data to be compressed to obtain an initial neural network; Randomly initializing the weights of the initial neural network to obtain a temporary neural network; The three-dimensional data to be compressed is input into the temporary neural network to obtain compressed target three-dimensional data.

3. The locally restorable compression method for industrial CT data according to claim 2, characterized in that: The step of inputting the to-be-compressed three-dimensional data into the temporary neural network to obtain compressed target three-dimensional data comprises: A. randomly extract a batch of simulation coordinates, pass them into the temporary neural network, perform forward propagation, and obtain the simulation values ​​of the data under the simulation coordinates; B. comparing the simulated value with the real value of the three-dimensional data to be compressed at the corresponding coordinates and calculating the loss, and obtaining the gradient direction by calculating the gradient through the loss function; C. updating the weight of the temporary neural network according to the reverse direction of the gradient direction and the current learning rate of the temporary neural network to obtain the final neural network; D. Repeat steps AC until the set number of training rounds is completed to obtain compressed target three-dimensional data.

4. The locally restorable compression method for industrial CT data according to claim 1, characterized in that: After the step of inputting the three-dimensional data to be compressed into the neural network to obtain the target three-dimensional data after compression of the preset product, the method further includes: Determine whether information for performing partial restoration is received; If the information of partial restoration is received, the coordinates of the partial area to be restored in the preset product are determined; The coordinates are input into the neural network and the local restoration area is obtained through calculation.

5. The locally restorable compression method for industrial CT data according to claim 4, characterized in that: The step of inputting the coordinates into the neural network and calculating to obtain the local restoration area comprises: Inputting the coordinates of each pixel point in the local area to be restored into the neural network one by one; Calculate the restored pixel value of each pixel point in the local area to be restored by forward propagation of the neural network; The restored pixel values ​​are coordinate spliced, and the image model is reconstructed based on the spliced ​​coordinates, thereby obtaining image information of the local restored area.

6. A compression device for industrial CT data capable of partial restoration, characterized in that: include: A traversal module, used to traverse the entire preset product, obtain the pixel value of each pixel point in each cross section of the preset product, so as to obtain the three-dimensional data to be compressed of the preset product; The input module is used to input the three-dimensional data to be compressed into the neural network to obtain the target three-dimensional data after compression of the preset product.

7. The partially restorable compression device for industrial CT data according to claim 6, characterized in that: The input module comprises: A neural network calculation submodule, used to pre-set the size of the target three-dimensional data after the three-dimensional data to be compressed is compressed, and to set the structure of the neural network according to the size of the target three-dimensional data and the three-dimensional data to be compressed, so as to obtain an initial neural network; A weight initialization submodule, used to randomly initialize the weights of the initial neural network to obtain a temporary neural network; The data compression submodule is used to input the three-dimensional data to be compressed into the temporary neural network to obtain compressed target three-dimensional data.

8. The partially restorable compression device for industrial CT data according to claim 6, characterized in that: Also includes: A receiving module, used to determine whether information for performing partial restoration is received; A judgment module, configured to determine the coordinates of the local area to be restored in the preset product if the information of the local restoration is received; The restoration module inputs the coordinates into the neural network and obtains the local restoration area through calculation.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the processor implements the steps of the partially restorable compression method for industrial CT data as described in any one of claims 1 to 5.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the locally restorable compression method for industrial CT data described in any one of claims 1 to 5 are implemented.

Images