Wheat moisture distribution THz image denoising method, equipment and medium

By combining information multi-distillation network and convolutional neural network, the wheat moisture distribution THz image is layered, which solves the problem of poor noise removal effect in the existing technology, realizes efficient noise denoising and accurate detection, and improves wheat The accuracy and image quality of moisture detection.

CN119090761APending Publication Date: 2024-12-06HENAN UNIVERSITY OF TECHNOLOGY

Patent Information

Application Number
CN202411057963.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-02
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The existing wheat moisture distribution THz image denoising method has poor effect and cannot effectively remove multiple noises, affecting the accuracy of image feature information extraction and detection.

Method used

Using a combination of information multi-distillation network IMDN and convolutional neural network DRUnet, the wheat moisture distribution THz image is layered, and noise is removed through cascade information multi-distillation blocks and convolution operations, combining the comparison perception channel attention mechanism and residual The difference block is processed in multiple scales.

Benefits of technology

It realizes efficient and high-quality denoising of wheat moisture distribution THz images, improves detection accuracy, effectively removes various noises, retains useful feature information, is suitable for deep learning models, and improves the data accuracy and fidelity of the image. .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119090761A_ABST
    Figure CN119090761A_ABST
Patent Text Reader

Abstract

The invention discloses a wheat moisture distribution THz image denoising method and device and a medium, and the method comprises the following steps: inputting THz image data obtained by scanning a wheat sample into an IMDN (Information Multi-Distillation Network), carrying out the layered extraction of image features through employing cascaded IMDBs in the IMDN, and fusing the image features into an output feature map; and inputting the feature map output by the IMDN into a convolutional neural network DRUnet, performing down-sampling and up-sampling by using a plurality of continuous residual blocks after performing convolution on the feature map, and outputting a de-noised wheat moisture distribution THz image after performing convolution. According to the method, the problems of insufficient applicability, poor denoising effect and the like of an existing method when the wheat moisture distribution THz image is denoised are solved, layering processing is performed on the image, various levels of noise can be removed by using a deep learning mode, and the method is suitable for processing the wheat moisture distribution THz image and video. The method can cooperate with a deep learning image detection model to improve the accuracy of wheat moisture distribution detection.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of agricultural detection, and in particular to a method, equipment and medium for denoising a THz image of wheat moisture distribution. Background Art

[0002] In the process of wheat quality testing, moisture detection is an important indicator. When the moisture content is too high, the wheat is prone to mold and cannot be used normally. Therefore, it is necessary to accurately test the moisture content and distribution of wheat in order to take targeted measures to control its moisture content within the most suitable range for storage.

[0003] The traditional drying method and chemical method for measuring moisture content require destructive treatment of the test samples, and the operation process is cumbersome and the detection cycle is long. The test results are easily affected by operational errors. Therefore, non-destructive spectral analysis is generally preferred. Among them, terahertz waves have the advantages of fast imaging speed, sensitivity to moisture, and non-destructive penetration, making them a better choice. However, the terahertz images obtained by terahertz imaging equipment scanning wheat samples will inevitably be mixed with system noise. In the process of using machine learning methods to analyze terahertz images to identify moisture content and distribution, the superimposed noise will affect the extraction of image feature information, and the accuracy of the test results will be reduced. Therefore, it is necessary to perform denoising preprocessing on the acquired terahertz images to ensure the accuracy of moisture detection. Chinese patent publication number CN114386456A, publication date April 22, 2022, entitled "A terahertz time-domain signal denoising method, image reconstruction method and system" discloses a terahertz signal denoising method, in which the noisy mode obtained by empirical mode decomposition is screened by threshold filtering to obtain a virtual noise channel, and then the noisy signal is separated into noise and denoised signal by independent component analysis. However, the signal separation method used in the method used in this patent cannot guarantee the deep separation of noise, and is prone to erroneous elimination of useful information and distortion. It is difficult to achieve a universal effect for the various types and levels of noise that may be involved in the image, and it is difficult to cooperate with the deep learning network model. The model generalization ability and denoising effect need to be further improved. Summary of the invention

[0004] The present invention overcomes the problems of insufficient applicability and poor denoising effect of existing methods for denoising THz images of wheat moisture distribution, and provides a method, device and medium for denoising THz images of wheat moisture distribution. The method uses a combination of an information multi-distillation network IMDN and a convolutional neural network DRUnet to perform layered processing on the image, thereby achieving efficient and high-quality denoising of the THz image of wheat moisture distribution. Various levels of noise can be removed by using a deep learning approach, and the method is suitable for processing THz images and videos, and can cooperate with a deep learning image detection model to improve the accuracy of wheat moisture distribution detection.

[0005] In order to achieve the above object, the present invention adopts the following scheme: The wheat moisture distribution THz image denoising method includes the following steps: The THz image data obtained by scanning wheat samples is input into the information multi-distillation network IMDN, and the image features are hierarchically extracted using the cascaded information multi-distillation block IMDB in IMDN and fused into the output feature map; The feature map output by IMDN is input into the convolutional neural network DRUnet. After convolution, the feature map is downsampled and upsampled using multiple consecutive residual blocks. After convolution, the output is the denoised THz image of wheat moisture distribution.

[0006] Preferably, the information multi-distillation network IMDN operates as follows: The input THz image is first subjected to a convolution operation of size 3×3, and then the feature information is hierarchically extracted through a cascade of multiple information multi-distillation blocks IMDB. The extracted multi-layer feature information is concatenated and then subjected to two convolution operations of size 1×1 and 3×3 respectively. The results of the two convolution operations and the result of the first convolution operation are aggregated and processed by an upsampler to obtain the output feature map.

[0007] Preferably, the upsampler comprises a cascade of a convolution of size 3×3 and a sub-pixel convolution.

[0008] Preferably, the information multi-distillation block IMDB is composed of a progressive refinement module PRM, a contrast-aware channel attention mechanism CCA and a convolution of size 1×1, which are connected in sequence; wherein the progressive refinement module PRM includes multiple layers of convolution of size 3×3, each layer of convolution is followed by a Leaky ReLU activation function, and a channel splitting operation is performed on the result of each layer of convolution processing, and the features of a part of the channels are retained as the refined features of the layer, and the features of the remaining channels are sent to the next layer, and the convolution processing results of the last layer and the refined features of other layers are channel-connected and aggregated, and the aggregation results are input into the CCA for processing.

[0009] Preferably, the contrast-aware channel attention mechanism CCA operates as follows: For input X=[ , …, , …, ], there are C spaces of size The feature map is used to calculate the contrast information value using the following expression: Among them, among them, Indicates the output elements, Represents the global contrast information evaluation function.

[0010] Preferably, the convolutional neural network DRUnet is provided with an input-end convolution and an output-end convolution, and 7 residual blocks are connected between the two convolutions, wherein the first 4 residual blocks are connected with 3 separation convolutions SConv at intervals to constitute a downsampling part; the last 4 residual blocks are connected with 3 transposed convolutions TConv at intervals to constitute an upsampling part, and a unit jump connection is adopted between a pair of symmetrically distributed separation convolutions SConv and transposed convolutions TConv centered on the 4th residual block.

[0011] Preferably, each of the residual blocks contains a ReLU activation function, and no activation function is set after the separation convolution SConv and the transposed convolution TConv.

[0012] The present application also provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the method described in any one of the above schemes.

[0013] The present application also provides a storage medium on which a computer program is stored, characterized in that when the program is executed by a processor, the method described in any one of the above schemes is implemented.

[0014] The present invention has at least the following beneficial effects: (1) by extracting image features in layers and fusing features according to their importance, the noise in the image that may interfere with wheat moisture detection can be removed more specifically and useless feature information in the image can be weakened; (2) by enhancing the image's texture, edge and other detail features through the attention mechanism, the original useful feature information in the image can be more completely retained, thereby ensuring the data accuracy of the denoised image when used for wheat moisture detection; (3) by utilizing the adaptive cropping strategy of the IMDN network and performing multi-scale downsampling and upsampling through the residual block of the DRUnet network, efficient denoising and high fidelity can be guaranteed when used for images of different sizes and noise intensities. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A schematic diagram of the IMDN-DRUnet network structure of the present invention; Figure 2 A schematic diagram of an IMDN network architecture of the present invention; Figure 3 A schematic diagram of the structure of an upsampler of an IMDN network of the present invention; Figure 4 A schematic diagram of an information multi-distillation block IMDB structure of the present invention; Figure 5 A schematic diagram of the structure of a CCA module of an IMDN network of the present invention; Figure 6 A schematic diagram of a DRUnet network structure of the present invention; Figure 7 This is a comparison chart of the denoising effect of the present invention. DETAILED DESCRIPTION

[0016] The present invention is further described in detail below in conjunction with the accompanying drawings so that those skilled in the art can implement the invention with reference to the description.

[0017] like Figure 1 As shown, the wheat moisture distribution THz image denoising method provided by the present invention comprises the following steps: The THz image data obtained by scanning wheat samples is input into the information multi-distillation network IMDN, and the image features are hierarchically extracted using the cascaded information multi-distillation block IMDB in IMDN and fused into the output feature map; like Figure 2As shown in the figure, the Information Multi-Distillation Network (IMDN) is a deep learning network for image super-resolution, which extracts and enhances image features through multiple distillations of information. IMDN and its basic component, the Information Multi-Distillation Block (IMDB), have a lightweight design and are suitable for real-time applications. IMDN uses the residual learning idea of ​​information distillation to gradually extract useful features through the residual structure and connect these features to enhance the network's expressiveness. IMDN uses the contrast-aware channel attention mechanism (CCA) to enhance the expressiveness of features by adjusting the weight of each channel, and adopts the adaptive cropping strategy (ACS) to make the input image adapt to the network structure and ensure that the size of the input image can be divided by the upsampling operation. IMDN can handle super-resolution tasks of different scales. Through downsampling operations, the network can flexibly handle images of various sizes. The network structure of IMDN includes feature extraction, detail learning, and reconstruction. The detail learning part consists of enhanced residual blocks (ERBs) and effective high-frequency attention blocks (HFABs), which can reduce memory consumption and speed up inference.

[0018] The feature map output by IMDN is input into the convolutional neural network DRUnet. After convolution, the feature map is downsampled and upsampled using multiple consecutive residual blocks. After convolution, the output is the denoised THz image of wheat moisture distribution.

[0019] The DRUnet network is an improvement on the FFDNet network. The structure is designed and implemented by using U-Net and ResNet. Compared with the traditional convolutional neural network (CNN), the U-Net network component can efficiently handle image-to-image conversion tasks, while the ResNet network implemented by multiple stacked residual blocks has powerful modeling capabilities. FFDNet takes the noise level map as input. DRUnet integrates the residual block into U-Net on its basis to perform efficient denoising prior modeling, which can be used for plug-and-play image restoration tasks. DRUnet adopts the classic encoder-decoder architecture of U-Net, which is conducive to capturing multi-scale features. Compared with ResNet, DRUnet adds additional jump connections, which helps to transfer gradients more effectively in the network. DRUnet uses fewer model parameters while maintaining performance, which makes it more efficient in terms of the number of parameters. By aggregating feature maps through a simple summation operation, DRUnet can achieve higher-precision image segmentation than other methods.

[0020] As an experimental example, different moisture contents were obtained by soaking each group of wheat samples for different times (0H, 2H, 4H, 6H, 8H, 10H), and the moisture contents of each group of wheat samples were measured to be 1.47%, 18.08%, 24.71%, 27.33%, 30.23%, and 31.91%, respectively. THz frequency domain images of six groups of wheat samples were obtained by terahertz spectroscopy equipment. In the frequency range of 0.5-2THz, the terahertz spectroscopy equipment extracted THz images at intervals of 0.01THz, and 150 THz images were obtained for each group of samples. The ratio of the training set to the test set was 8:2, and the original wheat moisture distribution THz frequency domain image was used as the model input for training, with the training batch set to 32 and the training cycle set to 200. The computer configuration used in the experiment: the operating system is Windows 10, the graphics card is NVIDIA GeForce RTX 3060, and the processor is Intel 12th generation i5-12490F. The denoising model was built and trained on the Pytorch deep learning framework. The Adam algorithm was used to optimize the model parameters, with a learning rate of 1e-4 and a loss function of mean absolute error (MAE). The denoising effect of wheat moisture THz images was evaluated.

[0021] The denoising effects of IMDN-DRUnet and the original DRUnet network on the THz image of wheat moisture distribution are compared. In order to observe the denoising effects of the two algorithms more clearly, the THz denoised image of wheat grains soaked in 0H is taken as an example, and the details are enlarged, such as Figure 7 As shown, the left group of pictures (a) are the DRUnet network denoising effect pictures and their local detail enlarged pictures, and the right group of pictures (b) are the IMDN-DRUnet network denoising effect pictures and their local detail enlarged pictures. It can be seen that there is still stripe noise in the edge area of ​​the wheat moisture distribution THz image denoised by DRUnet, while the wheat moisture distribution THz image denoised by IMDN-DRUnet used in this application can obviously suppress the stripe noise, and there is no loss of image feature information, and the image quality is higher.

[0022] In order to verify the denoising effect of the improved model, the denoising effects of various denoising algorithms on the wheat moisture distribution THz image are compared. The peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) are used to evaluate the image denoising quality. The experimental results are shown in Table 1. Table 1 From the data in Table 1, it can be seen that the IMDN-DRUnet network provided in this application has improved PSNR and SSIM when denoising the wheat moisture distribution THz image. Taking the wheat moisture distribution THz image soaked for 0H as an example, the PSNR and SSIM of IMDN-DRUnet are 56.08dB and 0.9971 respectively. In terms of PSNR, it is 16.66dB, 7.51dB, 1.85dB and 1.01dB higher than DnCNN, FFDnet, IMDN-DnCNN and original DRUnet respectively; in terms of SSIM, it is 0.0147, 0.0093, 0.0011 and 0.0008 higher than DnCNN, FFDnet, IMDN-DnCNN and original DRUnet respectively. The experimental results show that for the wheat moisture distribution THz image, the method provided in this application has a good denoising effect improvement, which can effectively improve the quality of the wheat moisture distribution THz image so that it can be better used in the detection of wheat moisture distribution.

[0023] Furthermore, considering that THz images are often affected by different levels of noise, it is necessary to denoise the wheat moisture distribution THz image with various levels of noise to facilitate the subsequent moisture detection process. Gaussian noise of different degrees (15dB, 25dB, 50dB) is added to the original wheat moisture distribution THz image, and denoising is performed using the method provided in this application and other denoising methods. The denoising effect is evaluated and compared using PSNR and SSIM. The experimental data show that with the increase of the noise level, the PSNR and SSIM of each method decrease accordingly, but at all the noise levels added in the experiment, the PSNR and SSIM when using the IMDN-DRUnet network denoising are better than other methods. The experimental results show that at different noise levels, the denoising effect of IMDN-DRUnet on the THz image of wheat moisture distribution is better than other methods.

[0024] The method provided in this application constructs an IMDN-DRUnet network. By extracting image features in layers and fusing features according to their importance, it can more effectively remove noise in the image that interferes with wheat moisture detection and weaken useless feature information in the image; by enhancing the image's texture, edge and other detail features through the attention mechanism, it can more completely retain the original useful feature information in the image, ensuring the accuracy of the denoised image for wheat moisture detection; using the adaptive cropping strategy of the IMDN network, the residual block of the DRUnet network is used for multi-scale downsampling and upsampling, which can ensure efficient denoising and high fidelity when used for images of different sizes and noise levels. After verification by the provided comparative experiments, the denoising method provided in this application has an obvious denoising effect on the THz image of wheat moisture distribution.

[0025] In another technical solution, Figure 2 As shown in the figure, Conv represents convolution, Sub-pixel represents sub-pixel convolution, and Upsampler represents upsampler. The operation mode of the information multi-distillation network IMDN is as follows: The input THz image is first subjected to a convolution operation of size 3×3, and then the feature information is hierarchically extracted through a cascade of multiple information multi-distillation blocks IMDB. The extracted multi-layer feature information is concatenated and then subjected to two convolution operations of size 1×1 and 3×3 respectively. The results of the two convolution operations and the result of the first convolution operation are aggregated and processed by an upsampler to obtain the output feature map.

[0026] like Figure 3 As shown, the upsampler includes a cascade of a 3×3 convolution and a sub-pixel convolution.

[0027] Given an input image, the corresponding output image is: The network is optimized using the mean absolute error (MAE) loss function, which is expressed as: in are updateable parameters of the model, for Norm.

[0028] In another technical solution, Figure 4 As shown in the figure, Channel Split represents channel separation, and concat represents connection. The information multi-distillation block IMDB is composed of a progressive refinement module PRM, a contrast-aware channel attention mechanism CCA and a convolution of size 1×1, which are connected in sequence; wherein the progressive refinement module PRM includes multiple layers of convolution of size 3×3, each layer of convolution is followed by a Leaky ReLU activation function, and the result of each layer of convolution processing is subjected to channel segmentation operation, and the features of a part of the channels are retained as the refined features of the layer, and the features of the remaining channels are sent to the next layer, and the convolution processing results of the last layer and the refined features of other layers are subjected to channel connection aggregation, and the aggregation results are input into CCA for processing.

[0029] like Figure 5 As shown, the contrast-aware channel attention mechanism CCA operates as follows: For input X=[ , …, , …, ], there are C spaces of size The feature map is used to calculate the contrast information value using the following expression: Among them, among them, Indicates the output elements, Represents the global contrast information evaluation function.

[0030] In another technical solution, Figure 1 and Figure 6 As shown in the figure, Residual Blocks (RB) is a residual block, SConv represents separation convolution, Skip Connection represents skip connection, TConv represents transposed convolution, Downscaling represents downsampling, and Upscaling represents upsampling. The convolutional neural network DRUnet is provided with input-end convolution and output-end convolution, and 7 residual blocks are connected between two convolutions, wherein the first 4 residual blocks are connected with 3 separation convolutions SConv at intervals to form the downsampling part; the last 4 residual blocks are connected with 3 transposed convolutions TConv at intervals to form the upsampling part, and a unit skip connection is used between a pair of symmetrically distributed separation convolutions SConv and transposed convolutions TConv with the 4th residual block as the center.

[0031] The backbone network of the DRUnet network adopts a U-Net structure and consists of four scales. Each scale has a unit skip connection between 2×2 leap convolution downscaling and 2×2 transposed convolution (TConv) upscaling. The number of channels from the first scale to the fourth scale is 64, 128, 256, and 512 respectively. The downsampling and upsampling of each scale consists of 4 consecutive residual blocks. There is no activation function at the beginning and end of the network, and after the SConv and TConv layers.

[0032] Each residual block contains a ReLU activation function, while no activation function is set after the separation convolution SConv and the transposed convolution TConv. This helps to enhance its scaling invariance in the wheat moisture distribution THz image denoising task, that is, , which for any scalar All are established.

[0033] The present application also provides an electronic device, comprising: at least one processor, and a memory connected to the at least one processor in communication, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor, so that the at least one processor performs the wheat moisture distribution THz image denoising method as described in any one of the above schemes. The electronic device can be any terminal device including a mobile phone, a laptop computer, a desktop computer, a tablet computer, a PDA (Personal Digital Assistant), a POS (Point of Sales), a car computer, etc.

[0034] The present application also provides a storage medium on which a computer program is stored, characterized in that when the program is executed by a processor, the wheat moisture distribution THz image denoising method as described in any of the above schemes is implemented.

[0035] Through the description of the above embodiments, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software plus necessary general hardware, and of course, it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. In general, all functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present invention, software program implementation is a better implementation method in most cases. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer floppy disk, a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the wheat moisture distribution THz image denoising method described in each embodiment of the present invention.

[0036] It should be noted that, although the above describes the various steps in a specific order, it does not mean that the various steps must be performed in the above specific order. In fact, some of these steps can be performed concurrently or even in a different order, as long as the required functions can be achieved. The number of devices and processing scales described here are used to simplify the description of the present invention, and the application, modification and variation of the present invention are obvious to those skilled in the art.

[0037] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation modes, and they can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the illustrations shown and described herein.

Claims

1. THz image denoising method for wheat moisture distribution, It is characterized in that The following steps are involved: The THz image data obtained by scanning wheat samples is input into the information multi-distillation network IMDN, and the image features are hierarchically extracted using the cascaded information multi-distillation block IMDB in IMDN and fused into the output feature map; The feature map output by IMDN is input into the convolutional neural network DRUnet. After convolution, the feature map is downsampled and upsampled using multiple consecutive residual blocks. After convolution, the output is the denoised THz image of wheat moisture distribution.

2. The wheat moisture distribution THz image denoising method according to claim 1, It is characterized in that The operation mode of the information multi-distillation network IMDN is as follows: The input THz image is first subjected to a convolution operation of size 3×3, and then the feature information is hierarchically extracted through a cascade of multiple information multi-distillation blocks IMDB. The extracted multi-layer feature information is concatenated and then subjected to two convolution operations of size 1×1 and 3×3 respectively. The results of the two convolution operations and the result of the first convolution operation are aggregated and processed by an upsampler to obtain the output feature map.

3. The wheat moisture distribution THz image denoising method according to claim 2, It is characterized in that The upsampler consists of a cascade of a 3×3 convolution and a sub-pixel convolution.

4. The wheat moisture distribution THz image denoising method according to claim 1, It is characterized in that The information multi-distillation block IMDB is composed of a progressive refinement module PRM, a contrast-aware channel attention mechanism CCA and a convolution of size 1×1, which are connected in sequence; wherein the progressive refinement module PRM includes multiple layers of convolution of size 3×3, each layer of convolution is followed by a Leaky ReLU activation function, and the result of each layer of convolution processing is subjected to channel segmentation operation, and the features of a part of the channels are retained as the refined features of the layer, and the features of the remaining channels are sent to the next layer, and the convolution processing result of the last layer and the refined features of other layers are subjected to channel connection aggregation, and the aggregation result is input into CCA for processing.

5. The wheat moisture distribution THz image denoising method according to claim 4, It is characterized in that The contrast-aware channel attention mechanism CCA operates as follows: For input X=[ , …, , …, ], there are C spaces of size The feature map is used to calculate the contrast information value using the following expression: Among them, among them, Indicates the output elements, Represents the global contrast information evaluation function.

6. The wheat moisture distribution THz image denoising method according to claim 1, It is characterized in that The convolutional neural network DRUnet is provided with an input-end convolution and an output-end convolution, and 7 residual blocks are connected between the two convolutions, wherein the first 4 residual blocks are connected with 3 separation convolutions SConv at intervals to form a downsampling part; the last 4 residual blocks are connected with 3 transposed convolutions TConv at intervals to form an upsampling part, and a unit jump connection is adopted between a pair of symmetrically distributed separation convolutions SConv and transposed convolutions TConv with the 4th residual block as the center.

7. The wheat moisture distribution THz image denoising method according to claim 6, It is characterized in that Each of the residual blocks contains a ReLU activation function, while no activation function is set after the separation convolution SConv and the transposed convolution TConv.

8. Electronic equipment, It is characterized in that include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the method described in any one of claims 1 to 7.

9. A storage medium having a computer program stored thereon, It is characterized in that When the program is executed by a processor, the method described in any one of claims 1 to 7 is implemented.

Citation Information

Patent Citations

  • Terahertz time domain signal noise reduction method, image reconstruction method and system

    CN114386456A

Cited By

  • Low-dose CT image denoising method based on deep learning

    CN121504760A