A model construction method and system for identifying dark stripe noise in high-speed TDI CCD camera images

Abstract

A model construction method and system for identifying dark stripe noise in high-speed TDI CCD camera images, belonging to the technical field of image stripe noise identification. The present invention solves the problems that the existing methods for identifying dark stripe noise in images have low efficiency and accuracy, need to be completed with the assistance of manual review, and the selection of thresholds is difficult. The method comprises the following steps: step S1, collecting sample image data of TDI CCD camera, and manually classifying and labeling the collected sample image data; step S2, obtaining a binary image after preprocessing the sample image data with manual classification and labeling, and using it as an input sample set of a convolutional neural network, and the sample set is divided into a training sample set and a verification sample set; step S3, constructing a convolutional neural network classifier, and inputting the training sample set and the verification sample set into the convolutional neural network for model training and parameter tuning. The present invention is applicable to the technical field of image stripe noise identification, and is used to identify dark interference stripes in images.

Description

Technical Field

[0001] The present invention relates to the technical field of image stripe noise recognition, and in particular to a model building method and system for high-speed TDI CCD camera image dark stripe noise recognition. Background Art

[0002] During the development of high-speed TDI CCD cameras, due to unreasonable power supply system and PCB design, the acquired images are affected by the crosstalk between high-speed signals in the system and the interference of each power module, resulting in the presence of dark stripe noise with unstable frequency and direction in the image data. In order to ensure timely verification of whether there is dark stripe noise in the imaging process in the early stage of camera development, it is necessary to analyze the images of the development process. The dark stripe noise is evaluated using images taken by the camera under darkroom conditions. In the images acquired under this condition, the magnitudes of the 1 / f noise, readout noise and dark stripe noise of the camera drive circuit are not much different. That is, it is necessary to identify whether there is dark stripe noise in a mixed noise with similar magnitudes. Although the dark stripe noise to be identified has certain regularity visually, its frequency and direction are not fixed, and there is no deterministic regularity in statistical distribution, which makes it difficult to identify.

[0003] At present, there are two methods for identifying weak stripe noise in images:

[0004] 1. Use Photoshop software to manually check whether the image contains the dark stripe noise. Since the dark stripe noise has poor visibility on the image, the image needs to be linearly stretched multiple times when checking. Therefore, this method is inefficient, and the average recognition time for a single image is about 60 seconds;

[0005] 2. Count the ratio of values ​​in the image that deviate from the mean by 3 times the standard deviation. If the ratio is higher than a certain set threshold, it is considered that there is weak stripe noise in the image. This method is slightly more efficient than method 1, but the accuracy is low, only about 75%, and after the statistics are completed, manual review is still required to further confirm whether the image contains the weak stripe noise. Therefore, the recognition efficiency has not been significantly improved. In addition, the selection of the threshold in this method is difficult. For image data from different cameras, the selection of the threshold will affect the final recognition accuracy, and it needs to be selected for different cameras.

[0006] 3. The weak stripe noise in the image is regarded as the edge feature of the image. The Sobel edge detection operator is used to binarize the image. The 1-0 ratio in the binarized image is counted, and a ratio threshold is set. When the 1-0 ratio is higher than the threshold, it is considered that the image contains the weak stripe noise. The recognition accuracy of this method is about 80%. After the statistics are completed, manual review is also required to further confirm whether there is weak stripe noise in the image. In addition, this method has similar problems to method 2. The selection of the threshold in this method is difficult. For image data of different cameras, the selection of the threshold will affect the final recognition accuracy, and it needs to be selected for different cameras.

[0007] 4. The convolutional neural network recognition method is used to identify the dark stripe noise in the image. Most of them are constructed based on existing classification models, such as GoogleNet and AlexNet models. The model complexity is high and is for specific image sizes. It cannot adapt to the image size of the camera described in the present invention. If the image size is forcibly adjusted to adapt to the existing network model, the dark stripe noise will be submerged after the adjustment due to the small amplitude of the dark stripe noise. It cannot be recognized, and if a highly complex model is used, the training and recognition efficiency will also be reduced. For the dark stripe noise without a fixed direction and without periodicity described in this article, only the convolutional neural network is used for classification and recognition, and the recognition accuracy is only about 60%.

[0008] Therefore, the existing image weak stripe noise recognition method has the following defects:

[0009] 1) Low recognition efficiency;

[0010] 2) The recognition accuracy is low and needs to be completed with the assistance of manual review;

[0011] 3) It is difficult to select the threshold.

[0012] In summary, although the existing technology can be used to identify dark stripe noise in images, the efficiency and accuracy of identifying dark stripe noise in images are low, and it needs to be assisted by manual review, and the selection of thresholds is difficult.

[0013] Therefore, in the prior art, the problem that the efficiency and accuracy of identifying weak stripe noise in images are low, and manual review is required to complete the task, and the difficulty of selecting the threshold value is not clearly improved. For example, the patent document CN113487547A discloses a "satellite remote sensing image stripe noise positioning method and device", which solves the problem that the image stripe noise detection method based on deep learning in the prior art can only process ordinary images of smaller size, but cannot process large-size satellite remote sensing images by constructing a convolutional neural network model. The technical solution described in the patent document can only solve the problem of processing large-size satellite remote sensing images, and does not provide a technical solution for the low efficiency and accuracy of identifying weak stripe noise in images, and manual review is required to complete the task, and the difficulty of selecting the threshold value is large. The patent document CN106469436B discloses an "image denoising system and image denoising method", which processes the image denoising system by using the Sobel operator, and can retain image details as much as possible while denoising the image, thereby improving the image signal-to-noise ratio. The technical solution described in this patent document can only improve the image signal-to-noise ratio, and its technical solution cannot provide an effective solution for identifying dark and weak stripe noise in images. The efficiency and accuracy are low and it needs to be completed with the assistance of manual review. In addition, the selection of thresholds is difficult. Summary of the invention

[0014] The present invention solves the problems that the existing methods for identifying dark and weak stripe noise in images have low efficiency and accuracy, require manual review and assistance, and are difficult to select thresholds.

[0015] The method for constructing a model for identifying dark stripe noise in a high-speed TDI CCD camera image according to the present invention comprises the following steps:

[0016] Step S1, collecting sample image data of a TDI CCD camera, and manually classifying and labeling the collected sample image data;

[0017] Step S2, the binary image obtained after preprocessing the sample image data with manual classification and annotation is used as the input sample set of the convolutional neural network, and the sample set is divided into a training sample set and a verification sample set;

[0018] Step S3, construct a convolutional neural network classifier, and input the training sample set and the verification sample set into the convolutional neural network for model training and parameter tuning.

[0019] Furthermore, in one embodiment of the present invention, in step S1, the condition for collecting sample image data of the TDI CCD camera is:

[0020] The image was processed in a dark room to reduce the interference of other noises in the image on the weak stripe noise.

[0021] Furthermore, in one embodiment of the present invention, in step S2, the sample image data preprocessing is to perform edge detection and extraction on the sample image data based on the Sobel operator to enhance the recognition of dark and weak stripe noise in the image.

[0022] Furthermore, in one embodiment of the present invention, the Sobel operator adopts four groups of operator templates: horizontal, vertical, 45°, and 135°.

[0023] Furthermore, in an embodiment of the present invention, in step S2, the sample set is divided into a training sample set and a verification sample set in a ratio of 7:2.

[0024] Furthermore, in one embodiment of the present invention, in step S3, the step of constructing a convolutional neural network includes:

[0025] Input layer, first convolution layer C1, first pooling layer Pool1, second convolution layer C2, second pooling layer Pool2, third convolution layer C3, third pooling layer Pool3, fully connected layer FC1, Softmax output layer.

[0026] The invention discloses a high-speed TDI CCD camera image dark stripe noise recognition system, the system comprising the following modules:

[0027] Module S1 is used to collect sample image data of the TDI CCD camera and manually classify and annotate the collected sample image data;

[0028] Module S2 is used to obtain the binary image after preprocessing the sample image data of manual classification and annotation, and use it as the input sample set of the convolutional neural network. This sample set is divided into a training sample set and a verification sample set;

[0029] Module S3 is used to build a convolutional neural network classifier and input the training sample set and the verification sample set into the convolutional neural network for model training and parameter tuning.

[0030] The computer-readable storage medium of the present invention stores a computer program, and when the computer program is executed by a processor, the steps of any one of the above methods are implemented.

[0031] A computer device described in the present invention includes a memory and a processor, wherein a computer program is stored in the memory, and when the processor runs the computer program stored in the memory, the steps described in any one of the above methods are executed.

[0032] The present invention solves the problem that the existing methods for identifying dark and weak stripe noise in images have low efficiency and accuracy, require manual review and assistance, and are difficult to select thresholds. Specific beneficial effects include:

[0033] 1. The present invention adopts an improved Sobel operator and a convolutional neural network recognition method, and the recognition accuracy of weak stripe noise can reach 99.15%, and the recognition efficiency can reach 200 images / 60s, without the need for manual intervention, thus realizing high-accuracy, high-efficiency, and fully automatic recognition of weak stripe noise in images.

[0034] 2. The Sobel operator used in the present invention designs an adaptive threshold algorithm, which automatically sets the threshold according to different image data, thereby solving the problem that it is difficult to manually set an appropriate threshold when the existing Sobel operator is used for binarization.

[0035] 3. The Sobel operator used in the present invention adds two sets of operator templates of 45° and 135° on the basis of the existing two sets of horizontal and vertical operators. It can not only detect the horizontal and vertical gradients of the image, but also detect gradients in multiple directions at the same time, effectively improving the detection effect.

[0036] 4. The present invention establishes a new convolutional neural network model suitable for the output image size of the TDI CCD camera described in the present invention, defines convolution kernels of different sizes for feature extraction, and obtains feature information of images at different scales, thereby solving the problem of low recognition efficiency of existing convolutional neural networks due to their inability to adapt to the image size of the camera described in the present invention.

[0037] The invention is applicable to the technical field of image noise recognition, and solves the technical problems of great difficulty in recognizing weak stripe noise and low recognition accuracy in high-speed TDI CCD camera images. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0039] Figure 1 The present invention discloses a flow chart of a model building method for identifying dark stripe noise in a high-speed TDI CCD camera image.

[0040] Figure 2 These are example images of the test and training samples described in Implementation Method 2, (a) a normal dark field image; (b) a dark field image containing striped noise.

[0041] Figure 3 This is the convolutional neural network structure diagram described in Implementation Method 6, 1 input layer; 2 convolution layer; 3 pooling layer; 4 fully connected layer FC1; 5 Softmax output layer.

[0042] Figure 4 This is a training process diagram described in Implementation Method 6. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe various embodiments of the present invention in conjunction with the accompanying drawings. The embodiments described with reference to the accompanying drawings are exemplary and intended to be used to explain the present invention, but should not be understood as limiting the present invention.

[0044] Implementation method 1: A model construction method for identifying dark stripe noise in a high-speed TDI CCD camera image described in this implementation method, referring to Figure 1 This embodiment can be better understood by including the following steps:

[0045] Step S1, collecting sample image data of a TDI CCD camera, and manually classifying and labeling the collected sample image data;

[0046] Step S2, the binary image obtained after preprocessing the sample image data with manual classification and annotation is used as the input sample set of the convolutional neural network. This sample set is divided into a training sample set, a verification sample set and a test sample set;

[0047] Step S3, constructing a convolutional neural network classifier, and inputting the training sample set and the verification sample set into the convolutional neural network for model training and parameter tuning;

[0048] Step S4: the test sample set is input into the convolutional neural network classification model obtained in step S3 for classification and recognition, and the classification accuracy of the model for the test sample set is tested.

[0049] In this implementation, the experimental environment is a platform built under the Windows 10 system, the CPU model is Intel Xeon 5222, the memory size is 128G, and the experimental software platform is MATLAB R2018b.

[0050] Embodiment 2: This embodiment further defines the model building method for identifying dark stripe noise in a high-speed TDI CCD camera image described in Embodiment 1. In this embodiment, in step S1, the condition for collecting sample image data of the TDI CCD camera is:

[0051] The image was processed in a dark room to reduce the interference of other noises in the image on the weak stripe noise.

[0052] In this embodiment, N>1000 camera dark field images are acquired under darkroom test conditions. The acquired sample images are as follows: Figure 2 As shown, since the amplitude of the weak interference fringes is small, it is difficult to directly distinguish them in the original image. Therefore, the example images given have undergone contrast stretching and brightness adjustment to enhance the recognition of the weak interference fringes in the image.

[0053] Specifically, the sample data is manually classified and labeled according to whether the dark field image contains the weak stripe noise. Now, 1000 camera dark field images are manually classified and labeled, and 500 images are obtained, including 500 images interfered by the weak stripe noise and 500 images are not interfered by the weak stripe noise.

[0054] Implementation method three: This implementation method further limits the model construction method for identifying dark stripe noise in a high-speed TDI CCD camera image described in implementation method one. In this implementation method, in step S2, the sample image data preprocessing is based on the Sobel operator to perform edge detection and extraction on the sample image data, thereby enhancing the recognition of dark stripe noise in the image.

[0055] In this embodiment, in order to solve the problem that the amplitude of the stripe interference in the image is usually smaller than the background signal, which affects the recognition effect of the algorithm, the edge detection and extraction of the sample data is performed based on the Sobel edge extraction operator to enhance the recognition of the interfering dark stripes in the image. The result of the sample image data after the Sobel operator operation is a binary edge image.

[0056] Implementation method 4: This implementation method further limits the model building method for identifying dark stripe noise in a high-speed TDI CCD camera image described in implementation method 3. In this implementation method, the Sobel operator uses four groups of operator templates: horizontal, vertical, 45°, and 135°.

[0057] In this implementation, the Sobel operator uses operator templates in four directions to improve the detection effect of the Sobel operator. The operator templates in the four directions are as follows:

[0058]

[0059]

[0060] Wherein, m1 represents the vertical operator template; m2 represents the horizontal operator template; m3 represents the 45° direction operator template; and m4 represents the 135° direction operator template.

[0061] An adaptive threshold algorithm is designed to automatically set the binarization threshold T of the Sobel operator according to the gradient matrix g(x, y), and set the pixel points with g(x, y)>T to 1 and the pixel points with g(x, y)≤T to 0 to complete the binarization of the image.

[0062] Specifically, a median filter is introduced into the adaptive threshold algorithm, a 3×3 pixel template is selected as the threshold judgment template, the maximum, median and minimum values ​​of the gradient matrix g(x,y) in the template window are extracted, and the average of the three is taken as the threshold T of the pixel points in the template window to realize the automatic adjustment of the binarization threshold T of the Sobel operator.

[0063] The adaptive threshold algorithm used automatically sets the threshold according to different image data, solving the problem that the existing Sobel operator needs to manually set the appropriate threshold when performing binarization, which is difficult.

[0064] The Sobel operator in this implementation adds two sets of operator templates of 45° and 135° on the basis of the existing two sets of horizontal and vertical operators. It can not only detect the horizontal and vertical gradients of the image, but also detect gradients in multiple directions at the same time, effectively improving the detection effect.

[0065] Implementation method five: This implementation method further limits the model building method for identifying dark stripe noise in a high-speed TDI CCD camera image described in implementation method one. In this implementation method, in step S2, the sample set is divided into a training sample set, a verification sample set, and a test sample set in a ratio of 7:2:1.

[0066] In this embodiment, the binary image is used as a convolutional neural network input sample set, and the sample set is divided into a training sample set, a validation sample set, and a test sample set in a ratio of 7:2:1. The training sample set is used to train the convolutional neural network classification model and determine the model parameters during the training process; the validation sample set is used to adjust the model's hyperparameters during the model iterative training process and to perform a preliminary evaluation of the model's classification ability, verifying the generalization ability of the current model; and the test sample set is used to test the trained model's ability to distinguish weak stripe noise.

[0067] Specifically, the TDI CCD camera used to obtain sample data includes five spectral bands, namely P, B1, B2, B3 and B4, and is optically spliced ​​from three TDI CCDs. A total of 15 image data can be obtained in a single imaging. Since the stripe noise only exists in some spectral bands and some CCDs of the camera, in order to ensure the diversity of samples, sample data are obtained from multiple TDI CCD cameras. For a large number of original sample images obtained, it is necessary to manually classify each image to confirm whether there is weak stripe noise in the current image. Therefore, the sample acquisition involved in this example is difficult and the number of sample sets is small. Therefore, before sending the samples to the convolutional neural network, the sample set is first augmented to increase the number of samples in the training set and improve the generalization ability of the model.

[0068] Embodiment 6: This embodiment further defines the model construction method for identifying dark stripe noise in a high-speed TDI CCD camera image described in Embodiment 1. In this embodiment, in step S3, the construction of a convolutional neural network includes:

[0069] Input layer, first convolution layer C1, first pooling layer Pool1, second convolution layer C2, second pooling layer Pool2, third convolution layer C3, third pooling layer Pool3, fully connected layer FC1, Softmax output layer.

[0070] In this embodiment, if Figure 3 As shown in the figure, the convolution layer uses the convolution kernel to extract image features and obtain the feature map of the image. It is the core layer for building a convolutional neural network. The pooling layer is used to generate a lower-dimensional feature map, reduce the sensitivity of the convolution layer to position, and reduce the amount of calculation. The fully connected layer is used to obtain the one-dimensional features of the image. This one-dimensional feature is input to the Softmax output layer to obtain the probability of each category of the image and output the specific category of the current image.

[0071] For the constructed neural network classifier, it is necessary to use the constructed training sample set and verification sample set to train the model and continuously tune the parameters to obtain a classification model with the best recognition accuracy. The main parameters involved in the training process include convolution kernel size, convolution kernel sliding step, pooling unit function, activation function, learning rate, batch size, etc. Through multiple training and tuning of the constructed neural network model, the parameters of the classification model finally established are as follows:

[0072] (1) Input layer: The input training sample of the convolutional neural network is a grayscale image with a size of 750×1536 pixels.

[0073] (2) The convolution kernel size of the first convolution layer C1 is 11×11, the number of convolution kernels is 3, the convolution sliding step is 1, and the feature map size obtained after the first convolution layer is 750×1536×3. C1 uses the ReLU activation function to activate neurons. The pooling layer Pool1 uses the maximum pooling function for downsampling. The maximum pooling can retain the maximum value in the area (2×2). After the pooling layer, the image feature map size becomes 375×768×3.

[0074] (3) The convolution kernel size of the second convolutional layer C2 is 5×5, and the sliding step size is 1. The output of the second convolutional layer is also activated using the ReLU activation function, and down-sampled using a pooling layer with a step size of 2 and a pooling window of 2×2 to obtain an image feature map size of 187×384×16.

[0075] (4) The third convolutional layer C3 has a convolution kernel size of 3×3, a sliding step size of 1, and uses the ReLU activation function for activation. It is downsampled by a pooling layer with a step size of 2 and a pooling window of 2×2, and the output image feature map size is 93×192×32.

[0076] (5) The image feature map output by the third pooling layer Pool3 is processed by the fully connected layer FC1 to obtain the one-dimensional features of the image, and then processed by Softmax to obtain the probability of each category and output the category to which the current input image belongs.

[0077] The established convolutional neural network contains only 9 layers, and the network model has low complexity. Convolution kernels of different sizes can be defined for feature extraction to obtain feature information of different scales of images and improve the recognition rate of images.

[0078] The convolutional neural network model suitable for the output image size of the TDI CCD camera established by the present embodiment defines convolution kernels of different sizes for feature extraction, and obtains feature information of different scales of the image, thereby solving the problem of low recognition efficiency caused by the inability of the existing convolutional neural network to adapt to the image size of the camera described in the present invention.

[0079] In summary, the above-built convolutional neural network model is trained. The training process is as follows: Figure 4 As shown in the figure, after 10 epochs, the accuracy of this model for the verification sample set can reach 99.15%, and the recognition efficiency can reach 200 images / 60s, without the need for human intervention, achieving high-accuracy, high-efficiency, and fully automatic recognition of dark stripe noise in images.

[0080] Embodiment 7: A high-speed TDI CCD camera image dark stripe noise recognition system described in this embodiment includes the following modules:

[0081] Module S1 is used to collect sample image data of the TDI CCD camera and manually classify and annotate the collected sample image data;

[0082] Module S2 is used to obtain the binary images obtained after preprocessing the sample image data of manual classification and annotation, and use them as the input sample set of the convolutional neural network. This sample set is divided into a training sample set, a verification sample set and a test sample set;

[0083] Module S3 is used to build a convolutional neural network classifier and input the training sample set and the verification sample set into the convolutional neural network for model training and parameter tuning;

[0084] Module S4 is used to input the test sample set into the convolutional neural network classification model obtained by module S3 to classify and identify it, and test the classification accuracy of the model for the test sample set.

[0085] Embodiment 8: A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of the method described in any one of embodiments 1 to 6 are implemented.

[0086] Embodiment 9: A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the steps of the method described in any one of embodiments 1 to 6 are executed.

Claims

1. A model construction method for identifying dark stripe noise in high-speed TDI CCD camera images, characterized in that: The following steps are involved: Step S1, collecting sample image data of a TDI CCD camera, and manually classifying and labeling the collected sample image data; Step S2, the binary image obtained after preprocessing the sample image data with manual classification and annotation is used as the input sample set of the convolutional neural network, and the sample set is divided into a training sample set and a verification sample set; Step S3, constructing a convolutional neural network classifier, and inputting the training sample set and the verification sample set into the convolutional neural network for model training and parameter tuning; Wherein, in the step S2, the sample image data preprocessing is to perform edge detection and extraction on the sample image data based on the Sobel operator to enhance the recognition of the dark and weak stripe noise in the image; Design an adaptive threshold algorithm to automatically set the Sobel operator's binarization threshold T according to the gradient matrix g(x,y), and set the pixel points with g(x,y)>T to 1, and the pixel points with g(x,y)≤T to 0, to complete the image binarization. Specifically: introduce median filtering into the adaptive threshold algorithm, select a 3×3 pixel template as the threshold judgment template, extract the maximum, median and minimum values ​​of the gradient matrix g(x,y) in the template window, take the average of the three as the threshold T of the pixel points in the template window, and realize the automatic adjustment of the Sobel operator's binarization threshold T; The Sobel operator adopts four groups of operator templates: horizontal, vertical, 45°, and 135°.

2. The model building method for identifying dark stripe noise in high-speed TDI CCD camera images according to claim 1 is characterized in that: In step S1, the conditions for collecting sample image data of the TDI CCD camera are: The image was processed in a dark room to reduce the interference of other noises in the image on the weak stripe noise.

3. The model building method for identifying dark stripe noise in high-speed TDI CCD camera images according to claim 1 is characterized in that: In step S2, the sample set is divided into a training sample set and a verification sample set in a ratio of 7:

2.

4. The model building method for identifying dark stripe noise in high-speed TDI CCD camera images according to claim 1 is characterized in that: In step S3, the construction of a convolutional neural network includes: Input layer, first convolution layer C1, first pooling layer Pool1, second convolution layer C2, second pooling layer Pool2, third convolution layer C3, third pooling layer Pool3, fully connected layer FC1, Softmax output layer.

5. A high-speed TDI CCD camera image dark stripe noise recognition system, characterized in that: The system includes the following modules: Module S1 is used to collect sample image data of the TDI CCD camera and manually classify and annotate the collected sample image data; Module S2 is used to obtain the binary image after preprocessing the sample image data of manual classification and annotation, and use it as the input sample set of the convolutional neural network. This sample set is divided into a training sample set and a verification sample set; Module S3 is used to build a convolutional neural network classifier and input the training sample set and the verification sample set into the convolutional neural network for model training and parameter tuning; Among them, in the module S2, the sample image data preprocessing is to perform edge detection and extraction on the sample image data based on the Sobel operator to enhance the recognition of dark and weak stripe noise in the image; Design an adaptive threshold algorithm to automatically set the Sobel operator's binarization threshold T according to the gradient matrix g(x,y), and set the pixel points with g(x,y)>T to 1, and the pixel points with g(x,y)≤T to 0, to complete the image binarization. Specifically: introduce median filtering into the adaptive threshold algorithm, select a 3×3 pixel template as the threshold judgment template, extract the maximum, median and minimum values ​​of the gradient matrix g(x,y) in the template window, take the average of the three as the threshold T of the pixel points in the template window, and realize the automatic adjustment of the Sobel operator's binarization threshold T; The Sobel operator adopts four groups of operator templates: horizontal, vertical, 45°, and 135°.

6. 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 method according to any one of claims 1 to 4 are implemented.

7. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor runs the computer program stored in the memory, the steps of the method according to any one of claims 1 to 4 are performed.

Images