Image moire judgment model construction method and terminal
By using multi-scale convolution to extract the input model and data preprocessing methods, the overfitting problem of existing moiré pattern recognition models is solved, achieving efficient and accurate moiré pattern feature extraction and image analysis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- STATE GRID FUJIAN ELECTRIC POWER CO LTD
- Filing Date
- 2023-12-22
- Publication Date
- 2026-06-12
Smart Images
Figure CN117876757B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of deep learning and image quality assessment of electricity marketing metering boxes, and particularly to a method and terminal for constructing an image moiré pattern assessment model. Background Technology
[0002] In marketing meter box inspection scenarios, photos of the meter boxes taken on-site are typically submitted. The images are then used to assess for defects and other issues. Currently, some personnel responsible for on-site inspections use their phones to take photos of the meter boxes instead of physically inspecting the environment. These photos often contain unique moiré patterns, which can be extracted to determine if the photos are genuine on-site shots or copied images, thus improving closed-loop management capabilities.
[0003] A patent with publication number CN115984582A discloses a moiré pattern recognition model training method and apparatus, and an image reproduction recognition method. This method involves acquiring a training sample set, including image samples and label information indicating the presence or absence of moiré patterns in the image samples; dividing each image sample into multiple blocks and recording the position information of each block within the image sample; training a pre-built network model based on the training sample set; extracting features from each block in the image sample based on the position information to obtain block features, obtaining the recognition result for each block based on the block features, and updating the model parameters based on the recognition result and label information of each block until the model converges, thus obtaining a moiré pattern recognition model. However, the model construction in this patent requires a large amount of moiré pattern data and data without moiré patterns, making the trained model prone to overfitting. The trained model struggles to extract true moiré pattern information and instead tends to extract other redundant information from the image.
[0004] Another patent, CN116503883A, discloses a method and device for identifying reproduced ID cards, as well as a computer-readable medium and electronic device. This method obtains a time-domain feature map by preprocessing the target ID card image and a frequency-domain feature map by performing a Fourier transform on the image. By extracting the time-domain and frequency-domain feature maps and distributing the feature points on a two-dimensional coordinate system, the feature distributions in these two spaces are compared to determine if there are overlapping regions with gently changing peaks, thus identifying whether the ID card image is a reproduced image. However, this method requires using Fourier transform to extract moiré pattern features for comparison. Since it is a mathematical method, it is difficult to implement and the results are unpredictable. It requires comparison with the original image and cannot directly output whether the image is a reproduced image based on the current image. Summary of the Invention
[0005] The technical problem to be solved by the present invention is to provide a method and terminal for constructing an image moiré pattern judgment model, which is easy to implement moiré pattern judgment, achieves high-performance extraction of moiré pattern features, and effectively improves the model's judgment capability.
[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:
[0007] A method for constructing an image moiré pattern analysis model includes the following steps:
[0008] S1. Obtain the image to be processed and construct a multi-scale convolutional model to extract the input.
[0009] S2. The image to be processed with moiré pattern features is blurred and transformed, and the original image to be processed is subtracted from the transformed image to obtain the moiré pattern feature image.
[0010] S3. Input the moiré pattern feature image and the image to be processed without moiré pattern features into the multi-scale convolution extraction input model, carry out moiré pattern model training, and obtain the moiré pattern judgment model.
[0011] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is as follows:
[0012] A terminal for constructing an image moiré pattern analysis model includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it performs the following steps:
[0013] S1. Obtain the image to be processed and construct a multi-scale convolutional model to extract the input.
[0014] S2. The image to be processed with moiré pattern features is blurred and transformed, and the original image to be processed is subtracted from the transformed image to obtain the moiré pattern feature image.
[0015] S3. Input the moiré pattern feature image and the image to be processed without moiré pattern features into the multi-scale convolution extraction input model, carry out moiré pattern model training, and obtain the moiré pattern judgment model.
[0016] The beneficial effects of this invention are as follows: It provides a method and terminal for constructing an image moiré pattern judgment model. By performing a blur transformation on an image with moiré pattern features in the image to be processed and subtracting the original image, a moiré pattern feature image is obtained. The moiré pattern features are highlighted by data preprocessing, which greatly reduces image feature redundancy. Then, it is input together with the image to be processed without moiré pattern features into a pre-constructed multi-scale convolutional extraction input model. That is, the moiré pattern judgment model is trained based on a multi-scale moiré pattern feature extraction network, which can extract moiré pattern information at different scales. While making it easy to implement moiré pattern judgment, it improves the overall accuracy of moiré pattern extraction and effectively enhances the model's judgment capability. Attached Figure Description
[0017] Figure 1 This is a flowchart of an image moiré pattern analysis model construction method according to an embodiment of the present invention;
[0018] Figure 2 This is a schematic diagram of the structure of a terminal for constructing an image moiré pattern analysis model according to an embodiment of the present invention.
[0019] Label Explanation:
[0020] 1. A terminal for constructing an image moiré pattern analysis model; 2. Memory; 3. Processor. Detailed Implementation
[0021] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.
[0022] Please refer to Figure 1 A method for constructing an image moiré pattern analysis model, comprising the following steps:
[0023] S1. Obtain the image to be processed and construct a multi-scale convolutional model to extract the input.
[0024] S2. The image to be processed with moiré pattern features is blurred and transformed, and the original image to be processed is subtracted from the transformed image to obtain the moiré pattern feature image.
[0025] S3. Input the moiré pattern feature image and the image to be processed without moiré pattern features into the multi-scale convolution extraction input model, carry out moiré pattern model training, and obtain the moiré pattern judgment model.
[0026] As can be seen from the above description, the beneficial effects of the present invention are as follows: It provides a method and terminal for constructing an image moiré pattern judgment model. By performing a blur transformation on the image with moiré pattern features in the image to be processed and subtracting the original image, a moiré pattern feature image is obtained. The moiré pattern features are highlighted by the data preprocessing method, which greatly reduces the redundancy of image features. Then, it is input together with the image to be processed without moiré pattern features into a pre-constructed multi-scale convolutional extraction input model. That is, the moiré pattern judgment model is trained based on the multi-scale moiré pattern feature extraction network, which can extract moiré pattern information at different scales. While making it easy to implement moiré pattern judgment, it improves the overall accuracy of moiré pattern extraction and effectively enhances the model's judgment capability.
[0027] Furthermore, the multi-scale convolutional input extraction model consists of three 3-channel input convolutional groups, and the number of convolutional kernels in each of the three convolutional groups is 1.
[0028] The first convolutional group has a scale of 3×3, the second convolutional group has a scale of 5×5, and the third convolutional group has a scale of 7×7.
[0029] As described above, three convolutional groups with three channels and one kernel but different scales are constructed to achieve different scales for moiré pattern extraction. That is, the larger the scale of the convolutional kernel, the larger the moiré pattern can be extracted, and the smaller the scale of the convolutional kernel can be extracted for fine moiré patterns, thus achieving comprehensive extraction of moiré pattern information at different scales in the image.
[0030] Further, step S2 specifically includes:
[0031] S21. Constructing a variant fuzzy operator formula:
[0032]
[0033] Where G(x',y') is the blurred pixel value, π is pi, and σ is the standard deviation of the Gaussian kernel. This indicates summation within the neighborhood centered on the current pixel, where I(x,y) is the pixel value at coordinates (x,y) in the original image;
[0034] S22. Define the image to be processed with moiré pattern features as F. After performing G(x',y') fuzzy transformation on image F, the feature image K is obtained.
[0035] S23. Subtract image F from feature image K to obtain moiré feature image M with moiré pattern characteristics:
[0036] M = KF (2).
[0037] As described above, performing a blur transformation on an image with moiré patterns and subtracting the original image can enhance the moiré pattern features, significantly reduce image feature redundancy, and improve the model's judgment ability.
[0038] Further, step S3 specifically includes:
[0039] S31. Uniformly sample the moiré pattern feature image and the non-moiré pattern feature image, with a ratio of 1:1. The non-moiré pattern feature image is the image to be processed that does not have moiré pattern features.
[0040] S32. Stack the moiré pattern feature image and the non-moiré pattern feature image in multiple batches and input them into the multi-scale convolutional extraction input model, and predict the output value between 0 and 1;
[0041] S33. Mark the output values between 0 and 1 with target labels, where 0 represents no moiré pattern and 1 represents moiré pattern.
[0042] S34. Calculate the regression error using the target label and the output;
[0043] S35. The error is backpropagated to train the model and obtain the moiré pattern judgment model.
[0044] As described above, multi-scale convolution input extracts moiré pattern feature information at different scales, and then the output results are used to calculate the regression error. The moiré pattern judgment model finally trained using the error and conventional backpropagation training method can be directly deployed without much parameter adjustment. It can directly and accurately judge the moiré patterns in subsequent images, effectively reducing the implementation difficulty.
[0045] Further, step S34 specifically includes:
[0046] The mean squared error weighting method is used to calculate the regression error between the target label and the output:
[0047]
[0048] Where i represents the lower coordinate of the output, y i Indicates the predicted output, t i denoted by , m represents the upper limit of the number of outputs, and α and β are the decay coefficients of the predicted output and the label value, respectively.
[0049] As described above, the regression error between the target label (i.e., the target value) and the output (i.e., the predicted value) is calculated using the mean square error weighting method. Both the predicted output and the target value have attenuation coefficients, which can effectively reduce overfitting.
[0050] Please refer to Figure 2A terminal for constructing an image moiré pattern analysis model includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it performs the following steps:
[0051] S1. Obtain the image to be processed and construct a multi-scale convolutional model to extract the input.
[0052] S2. The image to be processed with moiré pattern features is blurred and transformed, and the original image to be processed is subtracted from the transformed image to obtain the moiré pattern feature image.
[0053] S3. Input the moiré pattern feature image and the image to be processed without moiré pattern features into the multi-scale convolution extraction input model, carry out moiré pattern model training, and obtain the moiré pattern judgment model.
[0054] As can be seen from the above description, the beneficial effects of the present invention are as follows: Based on the same technical concept, and in conjunction with the above-mentioned image moiré pattern judgment model construction method, a 3D convex lens method terminal is provided. By performing a blur transformation on the image with moiré pattern features in the image to be processed and subtracting the original image, a moiré pattern feature image is obtained. The moiré pattern features are highlighted by the data preprocessing method, which greatly reduces image feature redundancy. Then, it is input together with the image to be processed without moiré pattern features into a pre-constructed multi-scale convolutional extraction input model. That is, the moiré pattern judgment model is trained based on the multi-scale moiré pattern feature extraction network, which can extract moiré pattern information at different scales. While making it easy to implement moiré pattern judgment, it improves the overall accuracy of moiré pattern extraction and effectively enhances the model's judgment capability.
[0055] Furthermore, the multi-scale convolutional input extraction model consists of three 3-channel input convolutional groups, and the number of convolutional kernels in each of the three convolutional groups is 1.
[0056] The first convolutional group has a scale of 3×3, the second convolutional group has a scale of 5×5, and the third convolutional group has a scale of 7×7.
[0057] As described above, three convolutional groups with three channels and one kernel but different scales are constructed to achieve different scales for moiré pattern extraction. That is, the larger the scale of the convolutional kernel, the larger the moiré pattern can be extracted, and the smaller the scale of the convolutional kernel can be extracted for fine moiré patterns, thus achieving comprehensive extraction of moiré pattern information at different scales in the image.
[0058] Further, step S2 specifically includes:
[0059] S21. Constructing a variant fuzzy operator formula:
[0060]
[0061] Where G(x',y') is the blurred pixel value, π is pi, and σ is the standard deviation of the Gaussian kernel. This indicates summation within a neighborhood centered on the current pixel, where I(x,y) is the pixel value at coordinates (x,y) in the original image.
[0062] S22. Define the image to be processed with moiré pattern features as F. After performing G(x',y') fuzzy transformation on image F, the feature image K is obtained.
[0063] S23. Subtract image F from feature image K to obtain moiré feature image M with moiré pattern characteristics:
[0064] M = KF (2).
[0065] As described above, performing a blur transformation on an image with moiré patterns and subtracting the original image can enhance the moiré pattern features, significantly reduce image feature redundancy, and improve the model's judgment ability.
[0066] Further, step S3 specifically includes:
[0067] S31. Uniformly sample the moiré pattern feature image and the non-moiré pattern feature image, with a ratio of 1:1. The non-moiré pattern feature image is the image to be processed that does not have moiré pattern features.
[0068] S32. Stack the moiré pattern feature image and the non-moiré pattern feature image in multiple batches and input them into the multi-scale convolutional extraction input model, and predict the output value between 0 and 1;
[0069] S33. Mark the output values between 0 and 1 with target labels, where 0 represents no moiré pattern and 1 represents moiré pattern.
[0070] S34. Calculate the regression error using the target label and the output;
[0071] S35. The error is backpropagated to train the model and obtain the moiré pattern judgment model.
[0072] As described above, multi-scale convolution input extracts moiré pattern feature information at different scales, and then the output results are used to calculate the regression error. The moiré pattern judgment model finally trained using the error and conventional backpropagation training method can be directly deployed without much parameter adjustment. It can directly and accurately judge the moiré patterns in subsequent images, effectively reducing the implementation difficulty.
[0073] Further, step S34 specifically includes:
[0074] The mean squared error weighting method is used to calculate the regression error between the target label and the output:
[0075]
[0076] Where i represents the lower coordinate of the output, y i Indicates the predicted output, t i denoted by , m represents the upper limit of the number of outputs, and α and β are the decay coefficients of the predicted output and the label value, respectively.
[0077] As described above, the regression error between the target label (i.e., the target value) and the output (i.e., the predicted value) is calculated using the mean square error weighting method. Both the predicted output and the target value have attenuation coefficients, which can effectively reduce overfitting.
[0078] This invention provides a method and terminal for constructing an image moiré pattern judgment model, which is mainly applied to the image quality judgment scenario of electricity marketing metering boxes. The following is a detailed description with reference to specific embodiments:
[0079] Please refer to Figure 1 Embodiment 1 of the present invention is as follows:
[0080] A method for constructing an image moiré pattern analysis model includes the following steps:
[0081] S1. Obtain the image to be processed and construct a multi-scale convolutional model to extract the input.
[0082] The multi-scale convolutional input extraction model consists of three convolutional groups with 3 channels of input, and each of the three convolutional groups has 1 kernel. The first convolutional group has a 3×3 scale, the second convolutional group has a 5×5 scale, and the third convolutional group has a 7×7 scale.
[0083] This involves constructing three convolutional groups with 3 channels and 1 kernel, but at different scales. This allows for the extraction of moiré patterns at different scales. Larger-scale convolutional kernels can extract large-scale moiré patterns, while smaller-scale convolutional kernels can extract moiré patterns with finer textures, enabling comprehensive extraction of moiré pattern information at different scales in the image.
[0084] S2. Perform a blur transformation on the image to be processed with moiré pattern features, and subtract the original image from the transformed image to obtain the moiré pattern feature image.
[0085] S3. Input the moiré pattern feature image and the image to be processed without moiré pattern features into the multi-scale convolution to extract the input model, carry out the moiré pattern model training, and obtain the moiré pattern judgment model.
[0086] In this embodiment, the moiré pattern feature image is obtained by blurring the image with moiré pattern features in the image to be processed and subtracting the original image. The moiré pattern feature image is highlighted by the data preprocessing method, which greatly reduces the redundancy of image features. Then, it is input together with the image to be processed without moiré pattern features into the pre-constructed multi-scale convolutional extraction input model. That is, the moiré pattern judgment model is trained based on the multi-scale moiré pattern feature extraction network, which can extract moiré pattern information at different scales. While making it easy to implement moiré pattern judgment, it improves the overall accuracy of moiré pattern extraction and effectively enhances the model's judgment ability.
[0087] Embodiment 2 of the present invention is as follows:
[0088] A method for constructing an image moiré pattern analysis model, based on the above embodiment one, specifically includes step S2 as follows:
[0089] S21. Constructing a variant fuzzy operator formula:
[0090]
[0091] Where G(x',y') is the blurred pixel value, π is pi, and σ is the standard deviation of the Gaussian kernel. This indicates summation within a neighborhood centered on the current pixel, where I(x,y) is the pixel value at coordinates (x,y) in the original image.
[0092] S22. Define the image to be processed with moiré pattern features as F. After performing G(x',y') fuzzy transformation on image F, the feature image K is obtained.
[0093] S23. Subtract image F from feature image K to obtain moiré feature image M with moiré pattern characteristics:
[0094] M = KF (2).
[0095] In this embodiment, blurring the image to be processed with moiré patterns and subtracting the original image can enhance the moiré pattern features, significantly reduce image feature redundancy, and improve the model's judgment ability.
[0096] In this embodiment, step S3 specifically includes:
[0097] S31. Uniformly sample the moiré feature image and the non-moiré feature image, with a ratio of 1:1. The non-moiré feature image is the image to be processed that does not have moiré features.
[0098] S32. Stack multiple batches of moiré pattern feature images and non-moiré pattern feature images into a multi-scale convolutional extraction input model, and predict the output value between 0 and 1.
[0099] S33. Mark the output values between 0 and 1 with target labels, where 0 represents no moiré pattern and 1 represents moiré pattern.
[0100] S34. Calculate the regression error using the target label and output, specifically as follows:
[0101] The mean squared error weighting method is used to calculate the regression error between the target label and the output:
[0102]
[0103] Where i represents the lower coordinate of the output, y i Indicates the predicted output, t i denoted by , m represents the upper limit of the number of outputs, and α and β are the decay coefficients of the predicted output and the label value, respectively.
[0104] That is, the regression error between the target label (i.e., the target value) and the output (i.e., the predicted value) is calculated by the mean square error weighting method. Both the predicted output and the target value have attenuation coefficients, which can effectively reduce overfitting. In this embodiment, α is 0.99 and β is 0.9.
[0105] S35. The error is backpropagated to train the model and obtain the moiré pattern judgment model.
[0106] In other words, multi-scale convolution input extracts moiré pattern feature information at different scales, and then uses the output results to calculate regression error. The moiré pattern judgment model trained by the conventional backpropagation training method can be directly deployed without too many parameter adjustments, and can directly perform accurate judgment of subsequent image moiré patterns, effectively reducing the difficulty of implementation.
[0107] Please refer to Figure 2 Embodiment four of the present invention is as follows:
[0108] A terminal 1 for constructing an image moiré pattern analysis model includes a memory 2, a processor 3, and a computer program stored on the memory 2 and executable on the processor 3. When the processor 3 executes the computer program, it completes the steps in the image moiré pattern analysis model construction method of the above embodiment 1 or embodiment 2.
[0109] In summary, the image moiré pattern analysis model construction method and terminal provided by this invention have the following beneficial effects:
[0110] 1. The constructed model adopts data augmentation and multi-scale feature extraction techniques. Through multi-scale convolution input, moiré pattern information at different scales is extracted. The data preprocessing method, namely moiré pattern feature clipping, can enhance moiré pattern features, significantly reduce image feature redundancy, and improve the model's judgment ability.
[0111] 2. Easy to implement moiré pattern analysis: Once the model is trained, it can be deployed directly without much parameter adjustment, which effectively reduces the difficulty of implementation.
[0112] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.
Claims
1. A method for constructing a moire artifact judgment model of an image, characterized in that: Including the following steps: S1. Obtain the image to be processed and construct a multi-scale convolutional model to extract the input. S2. The image to be processed with moiré pattern features is blurred and transformed, and the original image to be processed is subtracted from the transformed image to obtain the moiré pattern feature image. Step S2 specifically involves: S21. Constructing fuzzy operator formulas: （1）； in, G ( x ', y ') represents the blurred pixel value, π is pi, and σ is the standard deviation of the Gaussian kernel. This indicates summation within a neighborhood centered on the current pixel. I ( x , y ) is the coordinate in the original image ( x , y x' and y' are the pixel values of the blurred image, and x' and y' are the coordinates of the pixels in the blurred image. S22. Define the image to be processed with moiré pattern features as F, and process image F... G ( x ', y The feature image K is obtained after fuzzy transformation. S23. Subtract image F from feature image K to obtain moiré feature image M with moiré pattern characteristics: M=KF(2) S3. Input the moiré pattern feature image and the image to be processed without moiré pattern features into the multi-scale convolution extraction input model, carry out moiré pattern model training, and obtain the moiré pattern judgment model.
2. The method for constructing an image moiré pattern analysis model according to claim 1, characterized in that, The multi-scale convolutional input extraction model consists of three 3-channel input convolutional groups, and the number of convolutional kernels in each of the three convolutional groups is 1. The first convolutional group has a scale of 3×3, the second convolutional group has a scale of 5×5, and the third convolutional group has a scale of 7×7.
3. The method for constructing an image moiré pattern analysis model according to claim 1, characterized in that, Step S3 specifically involves: S31. Uniformly sample the moiré pattern feature image and the non-moiré pattern feature image, with a ratio of 1:
1. The non-moiré pattern feature image is the image to be processed that does not have moiré pattern features. S32. Stack the moiré pattern feature image and the non-moiré pattern feature image in multiple batches and input them into the multi-scale convolutional extraction input model, and predict the output value between 0 and 1; S33. Mark the output values between 0 and 1 with target labels, where 0 represents no moiré pattern and 1 represents moiré pattern. S34. Calculate the regression error using the target label and the output; S35. The error is backpropagated to train the model and obtain the moiré pattern judgment model.
4. The method for constructing an image moiré pattern analysis model according to claim 3, characterized in that, Step S34 specifically involves: The mean squared error weighting method is used to calculate the regression error between the target label and the output: （3）； in i This indicates the lower coordinate of the output. y i Indicates the predicted output. t i Indicates the tag value. m Indicates the upper limit of the output quantity. α and β These are the decay coefficients for the predicted output and the label value, respectively.
5. A terminal for constructing an image moiré pattern analysis model, characterized in that, Includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, performs the following steps: S1. Obtain the image to be processed and construct a multi-scale convolutional model to extract the input. S2. The image to be processed with moiré pattern features is blurred and transformed, and the original image to be processed is subtracted from the transformed image to obtain the moiré pattern feature image. Step S2 specifically involves: S21. Constructing fuzzy operator formulas: （1）； in, G ( x ', y ') represents the blurred pixel value, π is pi, and σ is the standard deviation of the Gaussian kernel. This indicates summation within a neighborhood centered on the current pixel. I ( x , y ) is the coordinate in the original image ( x , y x' and y' are the pixel values of the blurred image, and x' and y' are the coordinates of the pixels in the blurred image. S22. Define the image to be processed with moiré pattern features as F, and process image F... G ( x ', y The feature image K is obtained after fuzzy transformation. S23. Subtract image F from feature image K to obtain moiré feature image M with moiré pattern characteristics: M=KF(2) S3. Input the moiré pattern feature image and the image to be processed without moiré pattern features into the multi-scale convolution extraction input model, carry out moiré pattern model training, and obtain the moiré pattern judgment model.
6. The image moiré pattern analysis model construction terminal according to claim 5, characterized in that, The multi-scale convolutional input extraction model consists of three 3-channel input convolutional groups, and the number of convolutional kernels in each of the three convolutional groups is 1. The first convolutional group has a scale of 3×3, the second convolutional group has a scale of 5×5, and the third convolutional group has a scale of 7×7.
7. A terminal for constructing an image moiré pattern analysis model according to claim 5, characterized in that, Step S3 specifically involves: S31. Uniformly sample the moiré pattern feature image and the non-moiré pattern feature image, with a ratio of 1:
1. The non-moiré pattern feature image is the image to be processed that does not have moiré pattern features. S32. Stack the moiré pattern feature image and the non-moiré pattern feature image in multiple batches and input them into the multi-scale convolutional extraction input model, and predict the output value between 0 and 1; S33. Mark the output values between 0 and 1 with target labels, where 0 represents no moiré pattern and 1 represents moiré pattern. S34. Calculate the regression error using the target label and the output; S35. The error is backpropagated to train the model and obtain the moiré pattern judgment model.
8. A terminal for constructing an image moiré pattern analysis model according to claim 7, characterized in that, Step S34 specifically involves: The mean squared error weighting method is used to calculate the regression error between the target label and the output: （3）； in i This indicates the lower coordinate of the output. y i Indicates the predicted output. t i Indicates the tag value. m Indicates the upper limit of the output quantity. α and β These are the decay coefficients for the predicted output and the label value, respectively.