Radar image denoising method and device based on deep network, and electronic equipment

Abstract

The invention discloses a radar image denoising method and device based on a deep network, and electronic equipment. The method comprises the steps of inputting a to-be-denoised image into an image denoising network; wherein the image to be denoised is a dual-channel ISAR image, and the image denoising network is constructed according to wavelet transform, KAN and UNet; the image denoising network comprises an encoder, a WKAN module and a decoder; the encoder is used for performing hierarchical feature extraction and wavelet embedding on an image to be denoised in sequence to obtain multi-scale input; the WKAN module is used for sequentially carrying out noise suppression, wavelet transformation and nonlinear modeling on the multi-scale input so as to obtain optimized multi-scale input; the decoder is used for processing the optimized multi-scale input through up-sampling and inverse wavelet transform so as to reconstruct a de-noised image; and the de-noised image is obtained. The method provided by the invention has good de-noising capability, and gives consideration to both calculation efficiency and network flexibility.

Description

Technical Field

[0001] This invention belongs to the field of radar image processing technology, specifically relating to a radar image denoising method, apparatus, and electronic device based on deep networks. Background Technology

[0002] Inverse Synthetic Aperture Radar (ISAR) possesses all-weather, all-day, long-range, and high-resolution imaging capabilities, making it valuable in situationsal awareness, air defense and early warning, and civil aviation control. However, during signal acquisition and processing, the quality of ISAR images is often affected by various types of noise. This noise may originate from environmental interference, system hardware limitations, and the inherent randomness of electromagnetic wave propagation. Notably, the noise characteristics in ISAR images vary with the distance between the radar and the target, resulting in a spatially non-uniform noise distribution. This variability in noise significantly degrades image quality. Summary of the Invention

[0003] To address the aforementioned problems in the prior art, this invention provides a radar image denoising method, apparatus, and electronic device based on deep networks.

[0004] The technical problem to be solved by this invention is achieved through the following technical solution: In a first aspect, the present invention provides a radar image denoising method based on deep networks, the method comprising: The image to be denoised is input into an image denoising network; wherein, the image to be denoised is a dual-channel ISAR image, and the image denoising network is constructed based on wavelet transform, KAN, and UNet; the image denoising network includes an encoder, a WKAN module, and a decoder; the encoder is used to perform hierarchical feature extraction and wavelet embedding on the image to be denoised to obtain a multi-scale input; the WKAN module is used to sequentially perform noise suppression, wavelet transform, and nonlinear modeling on the multi-scale input to obtain an optimized multi-scale input; the decoder is used to process the optimized multi-scale input through upsampling and inverse wavelet transform to reconstruct the denoised image; The denoised image is obtained.

[0005] Optionally, the encoder includes multiple cyclically connected ConvBlock and AvgPooling layers.

[0006] Optionally, the WKAN module includes two WKAN_down modules and one WKAN_up module.

[0007] Optionally, the WKAN_down module includes a Conv layer, a DWT layer, a Flatten layer, a Transpose layer, a LayerNorm layer, a KANBlock layer, and a DeConv layer connected in sequence.

[0008] Optionally, the WKAN_up module includes Conv&IWT and KANBlock connected in sequence.

[0009] Optionally, the decoder includes an UpBlock layer, a SkipConnection layer, a ConvBlock layer, and multiple UpBlock layers connected in sequence.

[0010] Secondly, the present invention provides a radar image denoising device based on deep networks, the device comprising: An image denoising module is used to input an image to be denoised into an image denoising network. The image to be denoised is a dual-channel ISAR image, and the image denoising network is constructed based on wavelet transform, KAN, and UNet. The image denoising network includes an encoder, a WKAN module, and a decoder. The encoder performs hierarchical feature extraction and wavelet embedding on the image to be denoised to obtain a multi-scale input. The WKAN module sequentially performs noise suppression, wavelet transform, and nonlinear modeling on the multi-scale input to obtain an optimized multi-scale input. The decoder processes the optimized multi-scale input through upsampling and inverse wavelet transform to reconstruct the denoised image. The denoised image is obtained.

[0011] Thirdly, the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When the processor executes the program stored in the memory, it implements the steps of the radar image denoising method based on deep networks described in the first aspect above.

[0012] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: In the above technical solution, this invention avoids relying on prior assumptions about noise statistical characteristics and image structural features. By combining KAN and wavelet transform, it achieves linear computational complexity while maintaining the ability to model long-range dependencies, making it suitable for dense prediction tasks. It also solves the problem that the secondary computational complexity of self-attention mechanisms limits their scalability and efficiency. The UNet architecture enables a coarse-to-fine feature extraction and reconstruction process, avoiding the increased system complexity and reduced flexibility caused by manual parameter tuning. This combination not only improves the denoising capability of ISAR images in low signal-to-noise ratio environments but also balances computational efficiency and network flexibility.

[0013] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the structure of an image denoising network provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a satellite simulation image and an optical image provided in an embodiment of the present invention; Figure 3 This is a comparison chart of simulation data imaging results provided in an embodiment of the present invention; Figure 4 This is a block diagram of a radar image denoising device based on a deep network provided in an embodiment of the present invention; Figure 5 This is a block diagram of an electronic device for a radar image denoising method based on deep networks, provided by an embodiment of the present invention. Detailed Implementation

[0015] To facilitate understanding of the present invention, a brief description of the prior art and the inventive concept of the present invention will be provided first.

[0016] In recent years, significant progress has been made in the field of image denoising, with methods evolving from traditional model-driven approaches to data-driven deep learning techniques. Traditional image denoising methods such as mean filtering, Gaussian filtering, and wavelet transform are widely used. Mean filtering smooths noise by calculating the average value of local pixels. Gaussian filtering uses a Gaussian function as weights for weighted averaging, which can preserve edge information to some extent. Wavelet transform separates the signal into different frequency bands through multi-scale decomposition, effectively removing high-frequency noise. With the development of deep learning technology, methods based on convolutional neural networks have gradually become the mainstream in the field of image denoising. Among them, DnCNN (Denoising Convolutional Neural Network), as a classic deep learning denoising model, adopts a residual learning mechanism to directly predict noise components rather than the clean image itself, thereby simplifying the training process and improving denoising performance. The core of DnCNN lies in its deep network structure and residual block design, which can capture complex features in the image and effectively remove noise. To overcome the limitations of DnCNN, Cwin-UNet (Cross-Window Attention Enhanced U-Net) was developed. Building upon the UNet architecture, it introduces a window attention mechanism, enhancing the model's understanding of local context and improving edge and detail recovery. Cwin-UNet divides the input image into multiple windows and performs self-attention computation within each window, enabling it to better capture local features in the image.

[0017] Traditional image denoising methods, such as mean filtering, smooth noise by calculating the average value of local pixels, but this can easily lead to blurred image edges. Gaussian filtering uses a Gaussian function as weights for weighted averaging, which can preserve edge information to some extent, but its effectiveness is limited in complex noisy environments. Wavelet transform separates the signal into different frequency bands through multi-scale decomposition, effectively removing high-frequency noise, but may introduce artifacts when processing non-stationary signals. DnCNN has limited adaptability to different types of noise, especially when processing ISAR images with complex textures and edges, where its performance degrades. Furthermore, its relatively fixed model structure makes it difficult to flexibly adjust to adapt to varying noise distributions. Therefore, while DnCNN improves image quality to some extent in practical applications, further optimization is needed to handle more complex noisy environments. Cwin-UNet divides the input image into multiple windows, but this method may lead to detail loss or artifacts when processing textures with strong directionality due to insufficient window fineness. Additionally, Cwin-UNet relies on predefined window sizes, which may not be optimal when dealing with images of different resolutions or content. Therefore, this invention proposes a radar image denoising method based on deep networks to solve this technical problem.

[0018] Figure 1 This is a schematic diagram of the structure of an image denoising network provided in an embodiment of the present invention. The radar image denoising method based on deep networks provided in this embodiment of the present invention may include the following steps: S101. Input the image to be denoised into an image denoising network; wherein, the image to be denoised is a dual-channel ISAR image, and the image denoising network is constructed based on wavelet transform, KAN, and UNet; the image denoising network includes an encoder, a WKAN module, and a decoder; the encoder is used to sequentially perform hierarchical feature extraction and wavelet embedding on the image to be denoised to obtain a multi-scale input; the WKAN module is used to sequentially perform noise suppression, wavelet transform, and nonlinear modeling on the multi-scale input to obtain an optimized multi-scale input; the decoder is used to process the optimized multi-scale input through upsampling and inverse wavelet transform to reconstruct the denoised image; The denoised image is obtained.

[0019] It is understandable that, such as Figure 1 The core architecture of this image denoising network integrates the hierarchical feature extraction mechanism of UNet, the nonlinear modeling capability of the KAN module, and the advantages of wavelet transform's multi-scale decomposition and physical prior constraints. The network takes a dual-channel ISAR image (real and imaginary parts) as input. The separation of the real and imaginary parts preserves the complete phase information of the original signal, which is crucial for radar image reconstruction. The denoising network needs to simultaneously optimize the real and imaginary channels to maintain signal consistency. The size of the dual-channel ISAR image is... ,in, Image height, Image width, Given the number of input channels, the output is a dual-channel enhanced image of the same size. The goal is to achieve noise suppression and image quality improvement through multi-scale feature extraction and nonlinear optimization.

[0020] Optionally, such as Figure 1 The encoder consists of multiple cyclically connected ConvBlock and AvgPooling layers.

[0021] Understandably, the encoder achieves multi-scale input through hierarchical feature extraction and wavelet embedding, laying the foundation for subsequent nonlinear modeling. The input image to be denoised first passes through two... A ConvBlock convolutional block consisting of a convolutional layer and a ReLU activation function has a channel count starting from... Expand to The output size is The image; then through AvgPooling layer (average pooling layer) downsampled to This forms intermediate shape features; the third layer is downsampled again through ConvBlock convolutional blocks and AvgPooling layers to... The number of channels reached Finally, shallow encoded features are generated through ConvBlock convolutional blocks. This leads to multi-scale input. The encoder enhances local texture extraction capabilities by stacking convolutional layers, while gradually abstracting edges, shapes, and semantic features.

[0022] Optionally, such as Figure 1 The WKAN module includes two WKAN_down modules and one WKAN_up module.

[0023] Understandably, the WKAN part achieves noise suppression and multi-scale modeling through two WKAN_down modules and one WKAN_up module. It combines the physical prior constraints of wavelet transform with the adaptive nonlinear modeling capability of the KAN module to jointly optimize feature representation from the frequency domain and spatial domain.

[0024] Optionally, the WKAN_down module includes a Conv layer, a DWT layer, a Flatten layer, a Transpose layer, a LayerNorm layer, a KANBlock layer, and a DeConv layer connected in sequence.

[0025] Understandably, the WKAN_down module first processes the multi-scale input from the encoder output, i.e. The feature map is processed through the Conv layer. After extracting local texture through convolution, a Discrete Wavelet Transform (DWT) is performed using the db4 wavelet basis of the DWT layer, decomposing it into approximate coefficients (LL) and horizontal / vertical / diagonal details (LH / HL / HH). This design is based on the frequency domain characteristics of ISAR images—the four vanishing moments of the db4 basis can effectively separate noise and target signals, while multi-scale decomposition can capture texture details in different directions. The decomposed sub-band features are then transformed through a Flatten layer and a Transpose layer. token sequence ( For the number of patches, The embedding dimension is normalized using LayerNorm (layer normalization layer) to provide stable input for subsequent attention mechanisms. Then, KANBlock (Kolmogorov-Arnold Networks Block, a modular neural network structure based on the Kolmogorov-Arnold theorem) uses three attention branches ( This approach models local details (e.g., aircraft outline), mesoscale textures (e.g., aircraft wing continuity), and global context (e.g., background noise distribution) separately. By combining this with a B-spline parameterized activation function, noise is suppressed while preserving the sharpness of target edges, making the nonlinear modeling more closely match the physical characteristics of ISAR images. Finally, a spatial feature map is restored through a DeConv layer (deconvolutional layer). After passing through two WKAN_down modules, the dimension is obtained as follows: The output space feature map.

[0026] Optionally, the WKAN_up module includes Conv&IWT and KANBlock connected in sequence.

[0027] Understandably, the WKAN_up module consists of two parts: Conv&IWT and KANBlock. Conv&IWT is accessed through... Convolution adjusts the number of channels in the output spatial feature map, and inverse wavelet transform (IWT) is performed on the frequency domain features based on the db4 wavelet basis to reconstruct the LL / LH / HL / HH subband features into a spatial representation, restoring the compressed local details of the output spatial feature map. KANBlock then performs tokenization and B-spline activation function-driven nonlinear modeling on the reconstructed output spatial feature map. Attention mechanisms and depthwise separable convolution (DwConv) enhance cross-scale feature fusion, resulting in an optimized multi-scale input. This design, through the synergy of frequency domain reconstruction and spatial modeling, further optimizes the separation of target contours and noise distribution.

[0028] It's worth noting that image denoising can also be performed by replacing other networks besides WKAN-UNet, such as DnCNN (Deep Convolutional Neural Network), UNet++ (Improved Encoder-Decoder Structure), and CBDNet (Blind Denoising Network Based on Convolutional Neural Network). Local detail enhancement during denoising can be achieved by replacing modules other than wavelet transform, such as the Spatial Attention Module, Non-local Means Module, Gated Recurrent Unit (GRU), or Local Feature Enhancement Module. Loss functions can be designed using metrics other than normalized root mean square error (such as mean difference, sharpness difference, and image entropy difference) for network training.

[0029] Optionally, the decoder includes an UpBlock layer, a SkipConnection layer, a ConvBlock layer, and multiple UpBlock layers connected in sequence.

[0030] It is understandable that, such as Figure 1 The decoder reconstructs the image through stepwise upsampling and inverse wavelet transform, ultimately restoring the target contour and local details. Combining the inverse transform properties of wavelet transform with the local perceptual capability of convolution, it achieves spatial detail restoration under physical prior constraints. The output of the WKAN_up module is the optimized multi-scale input, i.e. The feature map is processed via UpBlock (upsampling module), which includes a ConvTranspose (convolutional transpose) operation to resize the feature map from... Upsampling to The number of channels is adjusted through design parameters. Based on this, the SkipConnection layer concatenates the feature maps of the corresponding encoder layers with the upsampled feature maps, forming... Channel fusion features. This design improves image reconstruction quality by preserving edge and detail information extracted by the encoder, compensating for local details that the decoder may lose during upsampling. Then, a ConvBlock (two...) is used... The convolutional layer (using the ReLU activation function) fuses the upsampled features and enhances local representation. Finally, two UpBlocks are used to progressively restore the feature map size, outputting a denoised image with the same resolution as the input.

[0031] In one embodiment, the invention is verified by using simulated images of each satellite and their optical images obtained through electromagnetic data simulation. Figure 2 This is a schematic diagram of a satellite simulation image and optical image provided in an embodiment of the present invention. Figure 3 This is a comparison image of simulation data imaging results provided in an embodiment of the present invention. Figure 3This paper visually demonstrates the outstanding performance of the image denoising network WKAN-UNet of this invention. By comparing aircraft images processed by different methods, the advantages of WKAN-UNet in detail reconstruction are clearly visible. For example, when processing aircraft such as the F-15C, F-16, and F-117, WKAN-UNet can more completely restore the wing details, avoiding edge blurring and information loss. For the nose section of the F-18, WKAN-UNet retains more texture information, making the image richer and more realistic. When processing the fuselage scattering points of the MiG-29, the results of WKAN-UNet are closer to the labeled image, accurately capturing key features. These visual effects further confirm WKAN-UNet's powerful denoising capabilities and high-fidelity reconstruction capabilities in complex noisy environments.

[0032] Figure 4 This is a block diagram of a radar image denoising device based on deep networks provided in an embodiment of the present invention, such as... Figure 4 The device 400 may include: Image denoising module 401 is used to input the image to be denoised into an image denoising network; wherein, the image to be denoised is a dual-channel ISAR image, and the image denoising network is constructed based on wavelet transform, KAN, and UNet; the image denoising network includes an encoder, a WKAN module, and a decoder; the encoder is used to sequentially perform hierarchical feature extraction and wavelet embedding on the image to be denoised to obtain a multi-scale input; the WKAN module is used to sequentially perform noise suppression, wavelet transform, and nonlinear modeling on the multi-scale input to obtain an optimized multi-scale input; the decoder is used to process the optimized multi-scale input through upsampling and inverse wavelet transform to reconstruct the denoised image; The denoised image is obtained.

[0033] This invention avoids the prior assumptions about noise statistics and image structural features relied upon by traditional methods. By combining KAN and wavelet transform, it achieves linear computational complexity while maintaining the ability to model long-range dependencies, making it particularly suitable for dense prediction tasks. It solves the problem that the secondary computational complexity of self-attention mechanisms limits their scalability and efficiency. Through the UNet architecture, it realizes the coarse-to-fine feature extraction and reconstruction process, avoiding the increased system complexity and reduced flexibility caused by manual parameter tuning. This combination not only improves the denoising capability of ISAR images in low signal-to-noise ratio environments but also balances computational efficiency and model flexibility.

[0034] The method provided in this invention can be applied to electronic devices. Specifically, the electronic device can be a desktop computer, a portable computer, a smart mobile terminal, a server, etc. No limitation is made herein; any electronic device that can implement this invention falls within the protection scope of this invention.

[0035] Figure 5 This is a block diagram of an electronic device for a radar image denoising method based on deep networks, provided by an embodiment of the present invention. The present invention also provides an electronic device, such as... Figure 5 As shown, it includes a processor 501, a communication interface 502, a memory 503, and a communication bus 504, wherein the processor 501, the communication interface 502, and the memory 503 communicate with each other through the communication bus 504. Memory 503 is used to store computer programs; When the processor 501 executes the program stored in the memory 503, it implements the steps of the radar image denoising method based on deep networks described in any of the above embodiments.

[0036] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus.

[0037] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0038] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0039] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0040] It should be noted that the terms "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention.

[0041] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0042] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings and the disclosure in carrying out the claimed invention. In the description of the invention, the word "comprising" does not exclude other components or steps, "a" or "an" does not exclude a plurality, and "a plurality" means two or more, unless otherwise explicitly specified. Furthermore, while different embodiments may describe certain measures, this does not mean that these measures cannot be combined to produce good results.

[0043] The method provided in this invention can be applied to electronic devices. Specifically, the electronic device can be a desktop computer, a portable computer, a smart mobile terminal, a server, etc. No limitation is made herein; any electronic device that can implement this invention falls within the protection scope of this invention.

[0044] For the device / electronic device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and relevant parts can be referred to in the description of the method embodiments.

[0045] It should be noted that the device and electronic device in the embodiments of the present invention are respectively devices and electronic devices that apply the above-mentioned radar image denoising method based on deep networks. Therefore, all embodiments of the above-mentioned radar image denoising method based on deep networks are applicable to the device, electronic device and storage medium, and can achieve the same or similar beneficial effects.

[0046] Those skilled in the art will understand that embodiments of the present invention can be provided as methods or apparatus (devices). Therefore, the present invention may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects, all of which are collectively referred to herein as "modules" or "systems". This invention is described with reference to flowchart illustrations and / or block diagrams of methods and apparatus (devices) according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0047] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A radar image denoising method based on deep networks, characterized in that, The method includes: The image to be denoised is input into an image denoising network; wherein the image to be denoised is a dual-channel ISAR image, and the image denoising network is constructed based on wavelet transform, KAN, and UNet; the image denoising network includes an encoder, a WKAN module, and a decoder; the encoder is used to sequentially perform hierarchical feature extraction and wavelet embedding on the image to be denoised to obtain a multi-scale input; the WKAN module is used to sequentially perform noise suppression, wavelet transform, and nonlinear modeling on the multi-scale input to obtain an optimized multi-scale input; the decoder is used to process the optimized multi-scale input through upsampling and inverse wavelet transform to reconstruct the denoised image; The denoised image is obtained.

2. The radar image denoising method based on deep networks according to claim 1, characterized in that, The encoder includes multiple cyclically connected ConvBlock and AvgPooling layers.

3. The radar image denoising method based on deep networks according to claim 1, characterized in that, The WKAN module includes two WKAN_down modules and one WKAN_up module.

4. The radar image denoising method based on deep networks according to claim 3, characterized in that, The WKAN_down module includes the Conv layer, DWT layer, Flatten layer, Transpose layer, LayerNorm layer, KANBlock and DeConv layer connected in sequence.

5. The radar image denoising method based on deep networks according to claim 3, characterized in that, The WKAN_up module includes Conv&IWT and KANBlock connected in sequence.

6. The radar image denoising method based on deep networks according to claim 3, characterized in that, The decoder includes an UpBlock layer, a SkipConnection layer, a ConvBlock layer, and multiple UpBlock layers connected in sequence.

7. A radar image denoising device based on deep networks, characterized in that, The device includes: An image denoising module is used to input an image to be denoised into an image denoising network. The image to be denoised is a dual-channel ISAR image, and the image denoising network is constructed based on wavelet transform, KAN, and UNet. The image denoising network includes an encoder, a WKAN module, and a decoder. The encoder is used to sequentially perform hierarchical feature extraction and wavelet embedding on the image to be denoised to obtain a multi-scale input. The WKAN module is used to sequentially perform noise suppression, wavelet transform, and nonlinear modeling on the multi-scale input to obtain an optimized multi-scale input. The decoder is used to process the optimized multi-scale input through upsampling and inverse wavelet transform to reconstruct the denoised image. The denoised image is obtained.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to execute the executable instructions to implement the step of the radar image denoising method based on deep networks according to any one of claims 1 to 6.

Images