Dual denoising method and device for electromagnetic imaging detection based on N2V and EMI removal algorithm

Abstract

The application discloses a kind of based on N2V and EMI removal algorithm's electromagnetic imaging detection dual denoising method and device, the present application is by introducing N2V algorithm to carry out image noise to the image after initial denoising, and using the image after adding noise to carry out secondary denoising training;Thus, it can utilize secondary denoising model, to the initial denoising nuclear magnetic resonance image secondary denoising processing, to obtain the nuclear magnetic resonance image with better denoising effect, based on this, the present application can be in the case of lacking real image secondary denoising training, realizes the secondary denoising processing of nuclear magnetic resonance image, solves the restriction that image secondary denoising training cannot be carried out in traditional technology due to lack of clean image and noise image pair, whereby, the present application can further improve image quality, applicable in the field of ultra-low field MRI equipment imaging image denoising Large-scale application and popularization.

Description

Technical Field

[0001] This invention belongs to the field of image denoising technology, specifically relating to a dual denoising method and apparatus for electromagnetic imaging detection based on N2V and EMI removal algorithms. Background Technology

[0002] In recent years, many companies have been dedicated to developing low-field and ultra-low-field (ULF) MRI equipment. Ultra-low-field MRI is conceptually similar to high-field MRI. Its development aims not to replace high-field MRI equipment on the same track, but to fill gaps in the market by utilizing its unique characteristics. Among them, ultra-low-field MRI equipment has the advantages of low cost and portability, and does not require a dedicated operating space. Therefore, it is particularly suitable for imaging in remote and economically underdeveloped locations (clinics, hospitals in remote areas, etc.).

[0003] Under low magnetic fields, the signal-to-noise ratio of nuclear magnetic resonance is proportional to the 3 / 2 power of the magnetic field strength. Therefore, the image quality obtained under low magnetic fields is often inferior to that under high magnetic fields, and the imaging is often affected by noise. Thus, improving image quality and pursuing a higher signal-to-noise ratio has become an unavoidable issue in the imaging process of ultra-low field MRI equipment.

[0004] In MRI scans, methods to improve image quality can be broadly categorized into the following three types:

[0005] The first approach is to increase the number of signal averaging iterations to obtain images with a higher signal-to-noise ratio. However, this method increases scan time, causing patient discomfort, and motion artifacts are exacerbated by patient movement. Therefore, it is not well-suited for MRI image denoising. The second approach is to use radio frequency shielding to block noise. Using radio frequency shielding to resist external electromagnetic noise has always been a traditional method for reducing environmental noise sources. However, this method increases system cost and reduces system portability, thus it is also not well-suited for image denoising of ultra-low field MRI equipment. The third approach is image post-processing technology. Image post-processing technology does not affect scan time and does not require additional hardware. Based on this, it has become a potential competitor for removing external electromagnetic noise in ultra-low field MRI images. Deep learning methods, as a promising tool, are very suitable for denoising images captured by ultra-low field MRI equipment.

[0006] To date, most deep learning EMI removal algorithms applied to ULF MRI scanners primarily utilize CNN networks to learn the mapping relationship between EMI signals detected by EMI sensing coils and EMI signals detected by MRI receiving coils. However, when environmental noise is high and the number of EMI sensing coils used is low, the noise removal level that EMI removal algorithms can remove is limited, and the denoising effect does not meet the expected requirements. Furthermore, in ULF MRI applications, the lack of clean and noisy image pairs often limits the application of deep learning for secondary image denoising. Therefore, how to provide a method to achieve secondary denoising of ultra-low field MRI images even in the absence of clean and noisy image pairs, thereby improving image denoising performance, has become an urgent problem to be solved. Summary of the Invention

[0007] The purpose of this invention is to provide a dual denoising method and apparatus for electromagnetic imaging detection based on N2V and EMI removal algorithms, in order to solve the problems of poor denoising effect of existing technologies using EMI removal algorithms, and the inability to perform secondary denoising on ultra-low field MRI images when clean and noisy image pairs are lacking.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] Firstly, a dual denoising method for electromagnetic imaging detection based on N2V and EMI removal algorithms is provided, including:

[0010] An initial denoised MRI image is obtained, wherein the initial denoised MRI image is generated based on a denoised MRI signal, and the denoised MRI signal is obtained by denoising the MRI signal using an EMI removal algorithm;

[0011] A secondary denoising model is obtained, wherein the secondary denoising model is trained by taking several sample noisy NMR images and the corresponding mask images of each sample noisy NMR image as input, and the denoised images corresponding to each sample noisy NMR image as output. Each sample noisy NMR image corresponds to an initial denoised NMR image of the sample. Any sample noisy NMR image and its corresponding mask image are obtained by adding noise to the target image using the N2V algorithm, and the target image is the initial denoised NMR image of the sample corresponding to any sample noisy NMR image.

[0012] The initial denoised NMR image is subjected to secondary denoising processing using the aforementioned secondary denoising model, so that a denoised NMR image is obtained after the secondary denoising processing.

[0013] Based on the above disclosure, this invention pre-trains a secondary denoising model. During training, the input data for this model consists of several sample noisy MRI images and their corresponding mask images. Each sample noisy MRI image corresponds to an initial denoised MRI image, and both the noisy MRI image and its corresponding mask image are obtained by applying noise to the initial denoised MRI image using the N2V algorithm. Thus, this invention essentially adds different noises to each initial denoised image and utilizes a deep learning network. The initial denoised image with added noise is trained to denoise the image. After training, the model can denoise images containing different levels of noise. In practical applications, since the MRI image obtained after initial denoising still contains environmental noise, it is equivalent to treating the initial denoised MRI image as the aforementioned image with added noise. Finally, it is input into the secondary denoising model to remove the environmental noise from the initial denoised MRI image, thereby achieving secondary denoising and obtaining an MRI image with better denoising effect.

[0014] Through the aforementioned design, this invention introduces the N2V algorithm to add noise to the initially denoised image and uses the denoised image for secondary denoising training. In this way, the secondary denoising model can be used to perform secondary denoising processing on the initially denoised MRI image, resulting in an MRI image with better denoising performance. Based on this, this invention can perform secondary denoising training even in the absence of real images, realizing secondary denoising processing of MRI images. This overcomes the limitation of traditional techniques that cannot perform secondary denoising training due to the lack of clean and noisy image pairs. Therefore, this invention can further improve image quality and is suitable for large-scale application and promotion in the field of denoising of ultra-low field MRI imaging images.

[0015] In one possible design, the N2V algorithm is used to add noise to the target image, resulting in a noisy sample NMR image and a mask image corresponding to the target image, including:

[0016] Obtain random noise data and a mask image corresponding to the target image, wherein the size of the mask image is the same as the size of the target image, and the grayscale value of any pixel in the mask image is 0 or 1;

[0017] The random noise data is superimposed onto the target image to obtain an initial noisy image;

[0018] The grayscale values ​​of each pixel in the initial noisy image are normalized to obtain a normalized image;

[0019] Obtain blind spot pixel information, wherein the blind spot pixel information includes the proportion of blind spot pixels and the neighborhood size of the blind spot pixels, and the proportion of blind spot pixels is the ratio between the blind spot pixels and the total number of pixels in the normalized image.

[0020] Based on the blind spot pixel information, blind spot pixels are determined from the normalized image, and the blind spot pixels are assigned values ​​to obtain the sample noisy MRI image corresponding to the target image after the assignment process.

[0021] Based on the blind spot pixels determined in the normalized image, the mask image is assigned values ​​to obtain the mask image corresponding to the target image after the assignment process.

[0022] In one possible design, the blind spot pixels are determined from the normalized image based on the blind spot pixel information, including:

[0023] The number of blind pixels in the normalized image is determined based on the proportion of blind pixels in the blind pixel information.

[0024] Randomly select a number of pixels from the normalized image that are equal to the number of blind spot pixels, and use them as blind spot pixels;

[0025] Accordingly, the blind spot pixels are assigned values ​​to obtain a sample-added noisy MRI image corresponding to the target image after the assignment process, which includes:

[0026] Based on the neighborhood size of the blind pixel, the blind neighborhood region of each blind pixel in the normalized image is determined.

[0027] Based on the pixel values ​​of pixels in each blind spot neighborhood region, the blind spot pixels corresponding to each blind spot neighborhood region are assigned values ​​to obtain the sample noisy MRI image corresponding to the target image after the assignment process.

[0028] In one possible design, based on the blind spot pixels determined in the normalized image, the mask image is assigned values ​​to obtain a mask image corresponding to the target image after the assignment process, including:

[0029] Based on the blind pixel in the normalized image, sampling pixel points are determined from the mask image, wherein the pixel coordinates of the sampling pixel points correspond one-to-one with the pixel coordinates of the blind pixel determined in the normalized image;

[0030] The sampled pixels are set to 0 to obtain the mask image corresponding to the target image.

[0031] In one possible design, the loss function of the secondary denoising model is:

[0032]

[0033] In formula (1) above, L represents the loss function, x i Let represent the grayscale value of the i-th pixel in the label image corresponding to the input noisy NMR image, where the label image corresponding to the input noisy NMR image is the initial denoised NMR image of the sample corresponding to the noisy NMR image. z,i represents the gray value of the i-th pixel in the denoised image corresponding to the input noisy NMR image, mask represents the mask image corresponding to the input noisy NMR image, 1-mask represents the inversion operation of the mask image corresponding to the input noisy NMR image, and n represents the total number of pixels in the input noisy NMR image.

[0034] In one possible design, the secondary denoising model employs a trained residual network, wherein the residual network comprises four residual blocks arranged sequentially in the image processing direction and an output layer, and each residual block comprises a convolutional structure layer, a batch normalization layer and a corrected linear layer.

[0035] In one possible design, the convolutional structure layer comprises nine convolutional layers, wherein the kernel sizes used by the nine convolutional layers are 11×11, 11×11, 9×9, 9×9, 5×5, 5×5, 1×1, 1×1 and 1×1, respectively.

[0036] Secondly, a dual denoising device for electromagnetic imaging detection based on N2V and EMI removal algorithms is provided, comprising:

[0037] An initial denoising unit is used to acquire an initial denoised MRI image, wherein the initial denoised MRI image is generated based on a denoised MRI signal, and the denoised MRI signal is obtained by denoising the MRI signal using an EMI removal algorithm;

[0038] A secondary denoising unit is used to obtain a secondary denoising model, wherein the secondary denoising model is trained by taking several sample noisy NMR images and the corresponding mask images of each sample noisy NMR image as input, and the denoised images corresponding to each sample noisy NMR image as output. Each sample noisy NMR image corresponds to an initial denoised NMR image of the sample. Any sample noisy NMR image and its corresponding mask image are obtained by adding noise to the target image using the N2V algorithm. The target image is the initial denoised NMR image of the sample corresponding to any sample noisy NMR image.

[0039] The secondary denoising unit is further configured to perform secondary denoising processing on the initial denoised NMR image using the secondary denoising model, so as to obtain a denoised NMR image after the secondary denoising processing.

[0040] Thirdly, another dual denoising device for electromagnetic imaging detection based on N2V and EMI removal algorithms is provided. Taking the device as an electronic device as an example, it includes a memory, a processor, and a transceiver that are connected in sequence. The memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the dual denoising method for electromagnetic imaging detection based on N2V and EMI removal algorithms as described in the first aspect or any possible design of the first aspect.

[0041] Fourthly, a computer-readable storage medium is provided, on which instructions are stored, which, when executed on a computer, perform the electromagnetic imaging detection dual denoising method based on N2V and EMI removal algorithms as described in the first aspect or any possible design of the first aspect.

[0042] Fifthly, a computer program product containing instructions is provided, which, when executed on a computer, causes the computer to perform the electromagnetic imaging detection dual denoising method based on N2V and EMI removal algorithms as described in the first aspect or any possible design of the first aspect.

[0043] Beneficial effects:

[0044] (1) This invention introduces the N2V algorithm to add noise to the initially denoised image and uses the denoised image for secondary denoising training. In this way, the secondary denoising model can be used to perform secondary denoising on the initially denoised MRI image, thereby obtaining an MRI image with better denoising effect. Based on this, this invention can perform secondary denoising training in the absence of real images, realizing secondary denoising processing of MRI images. It solves the limitation of traditional technology that it is impossible to perform secondary denoising training due to the lack of clean image and noisy image pairs. Therefore, this invention can further improve image quality and is suitable for large-scale application and promotion in the field of denoising of imaging images of ultra-low field MRI equipment. Attached Figure Description

[0045] Figure 1 A schematic flowchart of the steps of the dual denoising method for electromagnetic imaging detection based on N2V and EMI removal algorithms provided in an embodiment of the present invention;

[0046] Figure 2 A schematic diagram of the arrangement structure of the EMI induction coil provided in an embodiment of the present invention;

[0047] Figure 3 This is a schematic diagram of the structure of the noise mapping model provided in an embodiment of the present invention;

[0048] Figure 4 The nuclear magnetic resonance image obtained by the conventional EMI removal algorithm provided in this embodiment of the invention;

[0049] Figure 5 The nuclear magnetic resonance image obtained after secondary denoising using this method is provided in the embodiments of the present invention;

[0050] Figure 6 A schematic diagram of the structure of the dual denoising device for electromagnetic imaging detection based on N2V and EMI removal algorithms provided in an embodiment of the present invention;

[0051] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0053] It should be understood that although the terms first, second, etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, without departing from the scope of the exemplary embodiments of the invention.

[0054] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0055] Example:

[0056] See Figures 1-5As shown, the dual denoising method for electromagnetic imaging detection based on N2V and EMI removal algorithms provided in this embodiment introduces the N2V algorithm to add noise to the initially denoised image, and uses the denoised image to train the neural network for secondary denoising, thereby obtaining a secondary denoising model. Based on this, in practical applications, this secondary denoising model can be used to perform secondary denoising processing on the initially denoised MRI image, thereby obtaining an MRI image with better denoising effect. Thus, this method can achieve secondary denoising training of MRI images without relying on clean images, solving the training limitations of traditional techniques and further improving the quality of MRI images. For example, this method can be run on the medical image denoising end-user. Optionally, the medical image denoising end-user can be run on, but is not limited to, a personal computer. It is understood that the aforementioned execution subject does not constitute a limitation on the embodiments of this application. Accordingly, the operation steps of this method can be, but are not limited to, the steps S1 to S3 below.

[0057] S1. Obtain an initial denoised MRI image, wherein the initial denoised MRI image is generated based on a denoised MRI signal, and the denoised MRI signal is obtained by denoising the MRI signal using an EMI removal algorithm.

[0058] In this embodiment, for example, but not limited to, the following steps S11 to S15 can be used to achieve initial denoising of the image captured by the MRI scanner.

[0059] S11. Acquire the MRI signal and the first EMI signal, wherein the first EMI signal is a noise signal detected by an EMI induction coil located in front of the MRI scanner; in specific applications, for example, the number of EMI induction coils may be, but is not limited to, two, that is, two EMI induction coils are set in front of the MRI scanner to detect electromagnetic interference around the MRI scanner. See also... Figure 2 As shown, for example, two EMI induction coils are selected along the gradient direction of the MRI scanner slice (i.e., Figure 2 The EMI signal detected by the EMI induction coil is orthogonally placed in front of the MRI scanner (in the z-axis direction). With the aforementioned placement structure, on the one hand, the EMI signal detected by the EMI induction coil can be better correlated with the noise signal received by the MRI scanner (i.e., the EMI signal it receives itself, which is the target noise signal mentioned below), thereby facilitating the subsequent learning of the mapping relationship between the two by the first residual network. On the other hand, compared with the traditional use of 4-10 coils to sense EMI signals, the present invention only requires 2 induction coils, which can significantly reduce the cost of noise reduction.

[0060] In addition, for example, in actual use, the MRI scanner is set to radio frequency on state, and in this state, the first EMI signal detected by the EMI induction coil in front of it and the MRI signal generated by itself are acquired; then, the initial denoising processing of the MRI image can be completed based on the aforementioned two signals.

[0061] After acquiring the MRI signal generated by the MRI scanner itself during actual use, and the first EMI signal detected by the EMI induction coil in front of it, a pre-trained noise mapping module can be obtained to predict the target noise signal (i.e. the EMI signal received by the MRI scanner itself) received by the MRI scanner under the interference of the first EMI signal, based on the first EMI signal detected by the EMI induction coil. This allows for the initial denoising of the MRI image based on the predicted target noise signal. The specific acquisition process of the noise mapping model can be, but is not limited to, the steps shown in step S12 below.

[0062] S12. Obtain a noise mapping model, wherein the noise mapping model is a trained first residual network. The noise mapping model is used to determine the mapping relationship between the second EMI signal detected by the EMI induction coil and the target noise signal received by the MRI scanner, and the target noise signal is generated by interference from the second EMI signal. In this embodiment, the noise mapping model is essentially trained using the EMI signal detected by the EMI induction coil as input and the EMI signal received by the MRI scanner as output. In layman's terms, the first residual network is used to learn the mapping relationship between the EMI signals detected by the EMI induction coil and the MRI scanner, so that the EMI signal mixed in by the MRI scanner during magnetic resonance imaging can be predicted based on the EMI signal detected by the EMI induction coil. Then, the MRI signal can be denoised based on the predicted EMI signal, thereby completing the entire initial denoising process.

[0063] Optionally, this embodiment can use the trained first residual network as the aforementioned noise mapping model; and in specific applications, this embodiment improves the network structure of the aforementioned first residual network, as shown below:

[0064] In practical applications, for example, the improved first residual network (i.e., the noise mapping model) can include, but is not limited to, following the signal processing direction (i.e., Figure 3 The four first residual blocks and one first output layer are arranged sequentially from left to right (as indicated by the arrows in the image). Each first residual block includes a first convolutional structure layer, a first batch normalization layer, and a first corrected linear layer. Figure 3In this embodiment, the two squares connected by each jump arrow represent a first residual block. The first output layer includes a first convolutional layer, and the activation function used by the first output layer is the hyperbolic tangent function. In this embodiment, the four residual blocks are mainly used to extract features from the input second EMI signal and to perform feature learning based on the extracted feature information. Finally, the first output layer outputs the output signal corresponding to the input signal (i.e., the mapping relationship between the EMI signal detected by the EMI induction coil and the EMI signal detected by the MRI scan signal).

[0065] Furthermore, in this embodiment, the first convolutional structure layer includes nine second convolutional layers, wherein the sizes of the convolutional kernels used in the nine second convolutional layers are 11×11, 11×11, 9×9, 9×9, 5×5, 5×5, 1×1, 1×1 and 1×1, respectively. Thus, through the aforementioned design, a different convolutional kernel than that used in traditional residual networks is used in the residual network. Based on this, the model learning task (i.e., the mapping relationship between the EMI signal detected by the EMI induction coil and the EMI signal received by the MRI scanner) can be better met, thereby improving the accuracy of predicting EMI signals mixed in the MRI signal.

[0066] After describing the specific architecture of the aforementioned improved first residual network, the following disclosure of one training method for the aforementioned improved first residual network may include, but is not limited to, the steps one, two, and three described below.

[0067] Step 1: Obtain the training dataset, which includes multiple second EMI signals and the corresponding real target noise signals for each second EMI signal. The real target noise signal corresponding to any second EMI signal refers to the electromagnetic interference signal received by the MRI scanner caused by the interference of any second EMI signal. In this embodiment, for example, but not limited to, a 3D gradient echo (GRE) sequence with EMI noise sampling function can be used to collect training data in order to obtain a more accurate dataset and provide a more reliable data foundation for model training.

[0068] Furthermore, for example, data can be collected from the MRI scanner when the RF (radio frequency) is on and off within a repetition time (TR). The second EMI signals of the EMI induction coils collected when the RF is off, as well as the corresponding EMI signals (i.e., real target noise signals) received by the MRI scanner, are used as the aforementioned training data. The data collected when the RF is on are used as test data.

[0069] In this embodiment, since the second EMI signal detected by the EMI induction coil in front of the MRI scanner is acquired during a TR time when the RF is on and off, as well as the EMI signal (i.e., the real target noise signal) received by the MRI scanner due to interference from each second EMI signal, the second EMI signal when the RF is off can be used to train the GAN model. After the model training is completed, the target noise signal corresponding to the second EMI signal when the RF is on can be tested (i.e., the second EMI signal when the RF is on is used as input to obtain the EMI signal received by the MRI scanner due to its interference).

[0070] Optionally, the training process is as shown in steps two and three below.

[0071] Step 2: Use each real target noise signal as label data; In this embodiment, it is equivalent to the MRI scanner receiving the electromagnetic interference signal generated by the second EMI signal (i.e., the EMI signal it receives itself, which is also the aforementioned real target noise signal) when the EMI induction coil detects a second EMI signal. In this way, the real noise target signal can be used as the true value to train the model; The specific training process is shown in Step 3 below.

[0072] Step 3: Using each second EMI signal and its corresponding label data in the training dataset as input and the target noise signal corresponding to each second EMI signal as output, train the first residual network to obtain the noise mapping model after training. In this embodiment, the FE (frequency encoding) line is used as input and label for both training and testing to obtain better training results.

[0073] Meanwhile, the training process described above is explained by referring to the specific architecture of the improved first residual network: the improved first residual network continuously extracts features through four first residual blocks (where the first first residual block convolves the image from a single channel to 128 channels, and the remaining first residual blocks gradually restore the image to a single channel), and finally outputs the network through the first convolutional layer and activation function in the first output layer to obtain the target noise signal corresponding to the input second EMI signal. This target noise signal refers to the EMI signal predicted by the model and received at the MRI scanner, which is the electromagnetic interference noise in the MRI signal. Based on this, through continuous training, the mapping relationship between the EMI signal detected by the EMI induction coil and the EMI signal received by the MRI scanner can be learned.

[0074] Optionally, for example, the loss function of the improved first residual network is:

[0075] L′=MSE(x′,x′ z (2)

[0076] In the above formula (2), L′ represents the loss function of the noise mapping model (i.e., the improved residual network), x′ represents the input second EMI signal, and x′ z The input second EMI signal corresponds to the actual target noise signal, and MSE() represents the mean square error.

[0077] Therefore, as explained above, using the first residual network as a noise mapping model to achieve initial denoising has several advantages. First, since the first residual network itself uses residual connections, it can effectively solve the gradient vanishing and exploding problems during training, resulting in better fitting performance. Second, this embodiment features a lightweight network structure, which improves network convergence speed while achieving better fitting performance. Furthermore, this embodiment uses different convolutional kernels for feature extraction, which better adapts to the learning task and improves learning accuracy.

[0078] After training the improved first residual network and obtaining the noise mapping model, the model can be used to predict the EMI signal (i.e. the target noise signal corresponding to the first EMI signal) received by the MRI scanner under the interference of the aforementioned first EMI signal. Finally, based on the original MRI signal and the predicted target noise signal, the initial denoising of the signal can be performed to obtain the initial denoised MRI image. The denoising process is as shown in steps S13 to S15 below.

[0079] S13. Input the first EMI signal into the noise mapping model to obtain the target noise signal corresponding to the first EMI signal; in this embodiment, it is equivalent to using the noise mapping model to predict the electromagnetic interference signal in the MRI signal generated by the MRI scanner affected by the first EMI signal; that is, the target noise signal output by the model refers to the electromagnetic interference signal in the MRI signal; after predicting the electromagnetic interference signal in the MRI signal, initial denoising processing can be performed, as shown in step S14 below.

[0080] S14. Based on the target noise signal corresponding to the MRI signal and the first EMI signal, a denoised MRI signal is obtained; in this embodiment, the denoised MRI signal is obtained by subtracting the target noise signal corresponding to the first EMI signal from the MRI signal (of course, the subtraction is performed in K space, that is, in Fourier space).

[0081] After the denoising of the MRI signal is completed, image reconstruction can be performed to obtain the initial denoised MRI image; the reconstruction process is as shown in step S15 below.

[0082] S15. Perform an inverse Fourier transform on the denoised MRI signal to obtain the initial denoised MRI image after the inverse Fourier transform; in this embodiment, the inverse Fourier transform algorithm is a commonly used method for image reconstruction, and its principle will not be elaborated here.

[0083] Thus, through steps S11 to S15, the initial denoising of the MRI signal can be completed, and the initial denoised MRI image can be obtained. Then, in order to improve the denoising effect, this embodiment also sets up a secondary denoising process, as shown in steps S2 and S3 below.

[0084] S2. Obtain a secondary denoising model, wherein the secondary denoising model is trained by taking several sample noisy NMR images and the corresponding mask images of each sample noisy NMR image as input, and the denoised images corresponding to each sample noisy NMR image as output. Each sample noisy NMR image corresponds to an initial denoised NMR image of the sample. Any sample noisy NMR image and its corresponding mask image are obtained by adding noise to the target image using the N2V algorithm. The target image is the initial denoised NMR image of the sample corresponding to any sample noisy NMR image.

[0085] In practical applications, due to the lack of clean images in the application scenarios of ultra-low field MRI equipment (i.e., the images still contain noise after initial denoising), this embodiment uses the N2V algorithm to generate a dataset that can be trained for secondary denoising. The core idea of ​​the N2V algorithm is to add noise to the image and generate a corresponding mask image. Then, the blind spot mask (i.e., the mask image) and the noisy image are used to train the neural network model, so that the model can denoise images containing different levels of noise.

[0086] Optionally, in this embodiment, several initial denoised MRI images obtained after initial denoising using the aforementioned steps S11-S15 are used as sample initial denoised MRI images. Then, the NV2 algorithm is used to add noise to each of the aforementioned sample initial denoised MRI images, thereby obtaining several sample denoised MRI images and corresponding mask images. At the same time, during the noise addition process, the noise added to each sample initial denoised MRI image is different, and the added noise includes at least the types of noise contained in the sample initial denoised MRI images. In this way, it is equivalent to using the noise signal contained in the initial denoised MRI images to train the model, so that after the training is completed, the model can identify the noise contained in the initial denoised MRI images and perform noise removal processing.

[0087] In this embodiment, the specific construction process of the aforementioned sample noisy MRI image and the corresponding mask image is disclosed below. Since the construction process of each sample noisy MRI image and the corresponding mask image is the same, the following description takes any sample noisy MRI image as an example. The construction process can be, but is not limited to, the steps S21 to S26 below.

[0088] S21. Obtain random noise data and a mask image corresponding to the target image, wherein the size of the mask image is the same as the size of the target image, and the gray value of any pixel in the mask image is 0 or 1; in this embodiment, the target image refers to the initial denoised MRI image corresponding to any of the aforementioned sample denoised MRI images, and the random noise data is preset in the medical image denoising end, and all the random noise data packets set contain all types of noise signals contained in the initial denoised MRI image; similarly, the mask image is also preset in the medical image denoising end.

[0089] After obtaining the random noise data and the mask image, the sample noise-added NMR image and the mask image can be constructed. The construction process is shown in steps S22 to S26 below.

[0090] S22. The random noise data is superimposed onto the target image to obtain an initial noisy image.

[0091] S23. Normalize the gray values ​​of each pixel in the initial noisy image to obtain a normalized image; in this embodiment, for example, but not limited to, normalizing the gray value of any pixel in the initial noisy image to between 0 and 1; of course, gray value normalization is also a common technique in image processing, and its principle will not be elaborated here.

[0092] After normalizing the initial noisy image, blind pixel points in the image can be selected and their neighborhood values ​​assigned. The blind pixel selection and assignment process is shown in steps S24 and S25 below.

[0093] S24. Obtain blind spot pixel information, wherein the blind spot pixel information includes the proportion of blind spot pixels and the neighborhood size of the blind spot pixels, and the proportion of blind spot pixels is the ratio between the blind spot pixels and the total number of pixels in the normalized image; in this embodiment, the blind spot pixel information is also preset in the medical image denoising end, and the proportion of blind spot pixels and the neighborhood size can be specifically set according to actual use; optionally, the neighborhood size can be set to 3×3; furthermore, the blind spot pixel is the pixel corresponding to the effective signal in the image, and its neighboring pixels are regarded as the pixel corresponding to the invalid signal; therefore, using the noisy image and masked image constructed by the N2V algorithm for denoising training is equivalent to using the blind spot mask to learn the correlation between the selected effective signal and invalid signal, thereby realizing the removal of invalid signal.

[0094] After obtaining the blind spot pixel information, the blind spot pixels can be determined from the normalized image and their neighborhood values ​​can be assigned, as shown in step S25 below.

[0095] S25. Based on the blind spot pixel information, determine the blind spot pixels from the normalized image and assign values ​​to the blind spot pixels to obtain the sample noisy MRI image corresponding to the target image after the assignment process. In this embodiment, for example, but not limited to, the number of blind spot pixels in the normalized image can be determined according to the proportion of blind spot pixels in the blind spot pixel information. Then, a number of pixels equal to the number of blind spot pixels are randomly selected from the normalized image as blind spot pixels. For example, assuming that the proportion of blind spot pixels is one-tenth and the total number of pixels in the normalized image is 100, then 10 pixels are randomly selected from the 100 pixels as blind spot pixels. Of course, when the proportion of blind spot pixels is different, the principle of determining blind spot pixels is the same as the above example, and will not be repeated here.

[0096] Furthermore, after identifying the blind spot pixels, neighborhood assignment can be performed based on the neighborhood size of the blind spot pixels in the blind spot pixel information. This neighborhood assignment involves first determining the blind spot neighborhood region of each blind spot pixel in the normalized image based on the neighborhood size of the blind spot pixels; then, based on the pixel values ​​of the pixels in each blind spot neighborhood region, the blind spot pixels corresponding to each blind spot neighborhood region are assigned values. After the assignment process, a sample noisy NMR image corresponding to the target image is obtained. In practical applications, for any blind spot pixel, the grayscale value of any pixel is randomly selected from the blind spot neighborhood region corresponding to that blind spot pixel, and the selected grayscale value is assigned to that blind spot pixel, thus completing the neighborhood assignment for that blind spot pixel.

[0097] Therefore, after the aforementioned step S25, the selection and assignment of blind spot pixels in the normalized image can be completed, thereby obtaining the sample noisy NMR image corresponding to the target image; then, it is necessary to construct the mask image corresponding to the sample noisy NMR image based on the mask image, the process of which is shown in step S26 below.

[0098] S26. Based on the blind pixel identified in the normalized image, the mask image is assigned a value to obtain the mask image corresponding to the target image. In specific applications, sampling pixels can be determined from the mask image based on the blind pixel in the normalized image first; then, the pixel value of the sampling pixel is set to 0, thus obtaining the mask image corresponding to the target image. In this embodiment, it has been explained that the size of the mask image is the same as that of the target image, and the image size has not changed during the aforementioned image processing. Therefore, the size of the normalized image is also the same as that of the mask image. Based on this, after knowing the position of each blind pixel in the normalized image, the corresponding pixel can be found in the mask image according to its position (i.e., pixel coordinates) to serve as the sampling pixel. Thus, in this embodiment, the pixel coordinates of the sampling pixel correspond one-to-one with the pixel coordinates of the blind pixel identified in the normalized image. After obtaining the sampling pixel, its amplitude value is set to 0 to obtain the mask image corresponding to the target image.

[0099] Thus, through the aforementioned steps S21 to S26, the sample-denoised NMR image and mask image corresponding to the initial denoised NMR image of each sample can be constructed based on the N2V algorithm; then, the neural network can be trained based on the constructed data.

[0100] In this embodiment, the training process of secondary denoising is disclosed below, as shown in the following steps: (1) Obtain a secondary denoising training dataset, wherein the secondary denoising training dataset includes several sample noisy NMR images, and the sample initial denoised NMR image and mask image corresponding to each sample noisy NMR image; (2) Use the sample initial denoised NMR image corresponding to each sample noisy NMR image as the label image; (3) Use the sample noisy NMR image, the mask image and label image corresponding to each sample noisy NMR image as the input, and the denoised image corresponding to each sample noisy NMR image as the output. During the training process, the loss function value of the residual network (hereinafter referred to as the second residual network for easy distinction from the first residual network) is calculated using the denoised image corresponding to each sample noisy NMR image, the label image and the mask image corresponding to each sample noisy NMR image. The loss function value is used to determine whether the second residual network has converged based on the loss function value, and the secondary denoising model is obtained when it converges.

[0101] Therefore, as explained above, this embodiment does not require a clean MRI image when performing secondary denoising training. It can directly add noise to the noisy MRI image and then use the noisy image to perform secondary denoising training. This solves the limitation of traditional techniques that cannot perform secondary denoising training due to the lack of clean and noisy image pairs.

[0102] In practical applications, the secondary denoising model uses the trained residual network (i.e., the trained second residual network), and this embodiment improves the network architecture of the aforementioned secondary denoising model to further improve denoising efficiency.

[0103] Optionally, the aforementioned second residual network includes four residual blocks arranged sequentially according to the image processing direction and an output layer (hereinafter referred to as the second residual block and the second output layer for easy distinction from the residual blocks and the output layer in the aforementioned first residual network). In this embodiment, the second residual network has the same structure as the first residual network, that is, each second residual block also includes a second convolutional structure layer, a second batch normalization layer and a second corrected linear layer, and the second output layer also includes a second convolutional layer, and the activation function used by the second output layer is also the hyperbolic tangent function. Similarly, the second convolutional structure layer in the second residual network also contains nine convolutional layers, and the sizes of the convolutional kernels are 11×11, 11×11, 9×9, 9×9, 5×5, 5×5, 1×1, 1×1 and 1×1, respectively. Thus, by using the same network structure and convolutional parameters as the first residual network, the convergence efficiency can be improved while the fitting effect can be improved.

[0104] Furthermore, for example, the loss function of the aforementioned quadratic denoising model is:

[0105]

[0106] In formula (1) above, L represents the loss function, x i Let represent the grayscale value of the i-th pixel in the label image corresponding to the input noisy NMR image, where the label image corresponding to the input noisy NMR image is the initial denoised NMR image of the sample corresponding to the noisy NMR image. z,i The input sample denoised NMR image represents the gray value of the i-th pixel in the denoised image. The mask represents the mask image corresponding to the input sample denoised NMR image. The 1-mask represents the inversion operation of the mask image corresponding to the input sample denoised NMR image. n represents the total number of pixels in the input sample denoised NMR image. In this embodiment, the 1-mask is essentially the inversion of the gray value of the i-th pixel in the mask image (i.e., 0 becomes 1, or 1 becomes 0).

[0107] Of course, in this embodiment, if the grayscale value of the aforementioned sampled pixel is set to 1, then formula (1) becomes:

[0108]

[0109] Based on the loss function of the aforementioned secondary denoising model, it can be seen that when calculating the loss function value, pixels other than the blind pixel can be excluded from the loss calculation. Thus, for the loss function to converge, the selected blind pixel must be the pixel corresponding to the effective signal. In this way, the aforementioned second residual network learns the correlation between the effective signal and the noise signal in the image during the training process, so as to realize the identification and removal of the noise signal in the image.

[0110] Therefore, through the aforementioned design, this invention constructs noisy NMR images and corresponding mask images for each sample based on the N2V algorithm, and uses the constructed noisy images and mask images to train the second residual network for denoising. Based on this, a denoising model capable of performing secondary denoising on the image obtained after initial denoising can be obtained. Then, in practical applications, this secondary denoising model can be used to perform secondary denoising on the initial denoised NMR image in step S1, thereby obtaining an NMR image with better denoising effect. The secondary denoising process is as shown in step S3 below.

[0111] S3. The initial denoised NMR image is subjected to secondary denoising processing using the secondary denoising model to obtain a denoised NMR image after secondary denoising processing; in this embodiment, the initial denoised NMR image is input into the secondary denoising model to obtain the denoised NMR image.

[0112] Therefore, through the dual denoising method for medical images described in detail in steps S1 to S3 above, this invention introduces the N2V algorithm to add noise to the initially denoised image, and uses the denoised image to train the neural network for secondary denoising, thereby obtaining a secondary denoising model. Based on this, in practical applications, this secondary denoising model can be used to perform secondary denoising processing on the initially denoised MRI image, thereby obtaining an MRI image with better denoising effect. Thus, this invention can achieve secondary denoising training of MRI images without relying on clean images, solving the training limitations of traditional techniques and further improving the quality of MRI images.

[0113] In one possible design, the second aspect of this embodiment provides an example of using the method provided in the first aspect of the embodiment to denoise an MRI image, to illustrate that the method provided in this embodiment can achieve a secondary denoising effect.

[0114] This embodiment first performs initial denoising on the water film acquired using a ULF-MRI scanner, specifically using the EMI removal algorithm to obtain an initial denoised MRI image. Adam is selected as the optimizer, and the hyperparameters for training the first residual network are set as follows: learning rate α = 3.0 × e⁻⁴ and two exponential decay factors β₁ = 0.5 and β₂ = 0.999. The batch sizes for training and testing are set to 8 and 1, respectively. Figure 4 The image shows the reconstructed images before and after the EMI removal algorithm. The top image is the reconstructed image of the original data, and the bottom image is the reconstructed image after the EMI removal algorithm. It can be seen that some noise still remains.

[0115] Then, in this embodiment, the NV2 algorithm is used to construct the dataset for secondary denoising training and to train the second residual network. In the dataset preparation based on the N2V algorithm, the blind spot ratio and neighborhood size are set to 0.1 and 3×3, respectively. When training the second residual network, Adam is still selected as the optimizer, and the training hyperparameters are set as follows: learning rate α = 1.0×e⁻³ and two exponential decay factors β₁ = 0.5 and β₂ = 0.999, with training and testing batch sizes set to 2 and 2, respectively. After training, the trained second residual network is used to perform secondary denoising on the initially denoised MRI images, and the denoising effect is as follows: Figure 5 As shown ( Figure 5 The image in the upper middle is Figure 4 The image in the lower middle, Figure 5 The image below is the image after secondary denoising. Figure 5 As can be seen, the secondary denoising removed the noise that the EMI removal algorithm failed to remove, and the effect was good.

[0116] like Figure 6As shown, the third aspect of this embodiment provides a hardware device for implementing the dual denoising method for electromagnetic imaging detection based on N2V and EMI removal algorithms described in the first aspect of the embodiment, comprising:

[0117] An initial denoising unit is used to acquire an initial denoised MRI image, wherein the initial denoised MRI image is generated based on a denoised MRI signal, and the denoised MRI signal is obtained by denoising the MRI signal using an EMI removal algorithm.

[0118] A secondary denoising unit is used to obtain a secondary denoising model, wherein the secondary denoising model is trained by taking several sample noisy NMR images and the corresponding mask images of each sample noisy NMR image as input, and the corresponding denoised images of each sample noisy NMR image as output. Each sample noisy NMR image corresponds to an initial denoised NMR image of the sample. Any sample noisy NMR image and its corresponding mask image are obtained by adding noise to the target image using the N2V algorithm, and the target image is the initial denoised NMR image of the sample corresponding to any sample noisy NMR image.

[0119] The secondary denoising unit is further configured to perform secondary denoising processing on the initial denoised NMR image using the secondary denoising model, so as to obtain a denoised NMR image after the secondary denoising processing.

[0120] The working process, working details and technical effects of the device provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0121] like Figure 7 As shown, the fourth aspect of this embodiment provides another dual denoising device for electromagnetic imaging detection based on N2V and EMI removal algorithms. Taking the device as an electronic device as an example, it includes: a memory, a processor, and a transceiver that are connected in sequence. The memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the dual denoising method for electromagnetic imaging detection based on N2V and EMI removal algorithms as described in the first aspect of the embodiment.

[0122] For specific examples, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.; specifically, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), PLA (Programmable Logic Array). The processor may also include a main processor and a coprocessor. The main processor, also known as the CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state.

[0123] In some embodiments, the processor may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. For example, the processor may not be limited to microprocessors of the STM32F105 series, reduced instruction set computer (RISC) microprocessors, x86 architecture processors, or processors with integrated neural network processing units (NPUs). The transceiver may be, but is not limited to, a Wi-Fi transceiver, a Bluetooth transceiver, a General Packet Radio Service (GPRS) transceiver, a ZigBee (a low-power LAN protocol based on the IEEE 802.15.4 standard) transceiver, a 3G transceiver, a 4G transceiver, and / or a 5G transceiver. Furthermore, the device may also include, but is not limited to, a power module, a display screen, and other necessary components.

[0124] The working process, working details and technical effects of the electronic device provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0125] The fifth aspect of this embodiment provides a computer-readable storage medium storing instructions containing the instructions of the dual denoising method for electromagnetic imaging detection based on N2V and EMI removal algorithms as described in the first aspect of this embodiment. That is, the computer-readable storage medium stores instructions that, when executed on a computer, perform the dual denoising method for electromagnetic imaging detection based on N2V and EMI removal algorithms as described in the first aspect of this embodiment.

[0126] The storage medium refers to a carrier for storing data, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or memory sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0127] The working process, working details and technical effects of the storage medium provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0128] The sixth aspect of this embodiment provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the dual denoising method for electromagnetic imaging detection based on N2V and EMI removal algorithms as described in the first aspect of this embodiment, wherein the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0129] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A dual denoising method for electromagnetic imaging detection based on N2V and EMI removal algorithms, characterized in that, include: An initial denoised MRI image is obtained, wherein the initial denoised MRI image is generated based on a denoised MRI signal, and the denoised MRI signal is obtained by denoising the MRI signal using an EMI removal algorithm; A secondary denoising model is obtained, wherein the secondary denoising model is trained by taking several sample noisy NMR images and the corresponding mask images of each sample noisy NMR image as input, and the denoised images corresponding to each sample noisy NMR image as output. Each sample noisy NMR image corresponds to an initial denoised NMR image of the sample. Any sample noisy NMR image and its corresponding mask image are obtained by adding noise to the target image using the N2V algorithm, and the target image is the initial denoised NMR image of the sample corresponding to any sample noisy NMR image. The initial denoised NMR image is subjected to secondary denoising processing using the aforementioned secondary denoising model, so that a denoised NMR image is obtained after the secondary denoising processing. The N2V algorithm is used to add noise to the target image, resulting in a noisy NMR image and a mask image of the corresponding target image, including: Obtain random noise data corresponding to the target image and an initial mask image, wherein the size of the initial mask image is the same as the size of the target image, and the grayscale value of any pixel in the initial mask image is 0 or 1; The random noise data is superimposed onto the target image to obtain an initial noisy image; The grayscale values ​​of each pixel in the initial noisy image are normalized to obtain a normalized image; Obtain blind spot pixel information, wherein the blind spot pixel information includes the proportion of blind spot pixels and the neighborhood size of the blind spot pixels, and the proportion of blind spot pixels is the ratio between the blind spot pixels and the total number of pixels in the normalized image. Based on the blind spot pixel information, blind spot pixels are determined from the normalized image, and the blind spot pixels are assigned values ​​to obtain the sample noisy MRI image corresponding to the target image after the assignment process. Based on the blind spot pixels determined in the normalized image, the initial masking image is assigned a value so that the masking image corresponding to the target image is obtained after the assignment process. Based on the blind spot pixel information, the blind spot pixels are determined from the normalized image, including: The number of blind pixels in the normalized image is determined based on the proportion of blind pixels in the blind pixel information. Randomly select a number of pixels from the normalized image that are equal to the number of blind spot pixels, and use them as blind spot pixels; Accordingly, the blind spot pixels are assigned values ​​to obtain a sample-added noisy MRI image corresponding to the target image after the assignment process, which includes: Based on the neighborhood size of the blind pixel, the blind neighborhood region of each blind pixel in the normalized image is determined. Based on the pixel values ​​of pixels in each blind spot neighborhood region, the blind spot pixels corresponding to each blind spot neighborhood region are assigned values ​​to obtain the sample noisy MRI image corresponding to the target image after the assignment process.

2. The method according to claim 1, characterized in that, Based on the blind spot pixels determined in the normalized image, the initial masking image is assigned values ​​to obtain the masking image corresponding to the target image after the assignment process, including: Based on the blind spot pixels in the normalized image, sampling pixels are determined from the initial masking image, wherein the pixel coordinates of the sampling pixels correspond one-to-one with the pixel coordinates of the blind spot pixels determined in the normalized image. The pixel value of the sampled pixel is set to 0 to obtain the mask image corresponding to the target image.

3. The method according to claim 1, characterized in that, The loss function of the quadratic denoising model is: （1） In the above formula (1), Represents the loss function. This represents the grayscale value of the i-th pixel in the label image corresponding to the input noisy NMR image, where the label image corresponding to the input noisy NMR image is the initial denoised NMR image of the sample corresponding to the noisy NMR image. represents the grayscale value of the i-th pixel in the denoised image corresponding to the input noisy NMR image, and mask represents the mask image corresponding to the input noisy NMR image. This indicates that the mask image corresponding to the input sample noisy NMR image is inverted, and n represents the total number of pixels in the input sample noisy NMR image.

4. The method according to claim 1, characterized in that, The secondary denoising model uses a trained residual network, which includes four residual blocks arranged sequentially according to the image processing direction and an output layer. Each residual block includes a convolutional structure layer, a batch normalization layer, and a corrected linear layer.

5. The method according to claim 4, characterized in that, The convolutional structure layer includes nine convolutional layers, wherein the kernel sizes used in the nine convolutional layers are 11×11, 11×11, 9×9, 9×9, 5×5, 5×5, 1×1, 1×1 and 1×1 respectively.

6. A dual denoising device for electromagnetic imaging detection based on N2V and EMI removal algorithms, characterized in that, An apparatus for performing the dual denoising method for electromagnetic imaging detection based on N2V and EMI removal algorithms as described in any one of claims 1 to 5, wherein the apparatus comprises: An initial denoising unit is used to acquire an initial denoised MRI image, wherein the initial denoised MRI image is generated based on a denoised MRI signal, and the denoised MRI signal is obtained by denoising the MRI signal using an EMI removal algorithm; A secondary denoising unit is used to obtain a secondary denoising model, wherein the secondary denoising model is trained by taking several sample noisy NMR images and the corresponding mask images of each sample noisy NMR image as input, and the denoised images corresponding to each sample noisy NMR image as output. Each sample noisy NMR image corresponds to an initial denoised NMR image of the sample. Any sample noisy NMR image and its corresponding mask image are obtained by adding noise to the target image using the N2V algorithm, and the target image is the initial denoised NMR image of the sample corresponding to any sample noisy NMR image. The secondary denoising unit is further configured to perform secondary denoising processing on the initial denoised NMR image using the secondary denoising model, so as to obtain a denoised NMR image after the secondary denoising processing.

7. An electronic device, characterized in that, include: A memory, a processor, and a transceiver are sequentially connected in communication, wherein the memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer programs and execute the dual denoising method for electromagnetic imaging detection based on N2V and EMI removal algorithms as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a computer, perform the dual denoising method for electromagnetic imaging detection based on N2V and EMI removal algorithms as described in any one of claims 1 to 5.

Images