SAR image despeckle and noise reduction method based on DnCNN pre-training model
By constructing an image preprocessing model and using the DnCNN pretrained model to perform denoising of SAR images, the problem of high cost of SAR image training set is solved, and the rapid and batch denoising effect is achieved, reducing learning cost and time overhead.
Patent Information
- Application Number
- CN202411285062.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-13
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-09-13
AI Technical Summary
The cost of obtaining the training set of SAR images in the prior art is too high, especially when preprocessing using the MuLoG-DnCNN model, it takes a lot of time and effort, and preprocessing SAR images one by one is unrealistic, and a batch processing method is required. .
By constructing an image preprocessing model, obtaining the initial SAR image set and calculating the initial intensity map, adding speckle noise, performing logarithmic transformation and normalization processing, generating a transformed image set, and using the DnCNN pretrained model for denoising, reducing learning Cost and implement batch processing.
The learning cost required to use the MuLoG-DnCNN method is greatly reduced, and the preprocessed SAR image data is quickly obtained, and the denoising processing is convenient and quick to use the DnCNN model, which improves the cleanliness of SAR images.
Smart Images

Figure CN118941465A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of image processing, and in particular to a SAR image despeckle and denoising method based on a DnCNN pre-training model. Background Art
[0002] SAR (Synthetic Aperture Radar) images have very important applications in environmental monitoring, geological exploration, military reconnaissance and other fields. However, the speckle noise in SAR images greatly affects the clarity of the image and the accuracy of information processing, and brings many inconveniences to the subsequent links. Therefore, for a long time, how to eliminate speckle noise and improve the quality of SAR images has been an important research topic in related fields.
[0003] In recent years, deep learning-based methods have been widely used in the field of natural image denoising. In particular, convolutional neural networks (CNNs) have very deep architectures, large capacity, and extremely high flexibility to represent image features, making them very suitable for image restoration. By designing a specialized network structure and training a large amount of image data, CNN methods are able to adaptively learn complex noise patterns, thereby achieving efficient denoising effects that are difficult to achieve with other methods. Despeckle is a long-standing problem in synthetic aperture radar (SAR) imaging. Many different SAR image restoration schemes have been proposed. Among various possible strategies, methods based on convolutional neural networks (CNNs) have been shown to produce the best SAR image restoration results.
[0004] However, in order to learn an effective model, it is often necessary to collect a large number of speckle-free SAR images for training in advance, which undoubtedly consumes a lot of time and energy.
[0005] The DnCNN model and the MuLoG method are combined together to eliminate the multiplicative speckle noise in SAR images using the DnCNN pre-trained model, which achieves the expected goal and obtains excellent results.
[0006] However, learning to use the MuLoG-DnCNN model, especially how to preprocess SAR images, still requires a certain amount of time and effort, even for technicians in this field. In addition, when using remote sensing data for analysis and processing, a large number of SAR images are often required. It is very unrealistic to preprocess them one by one. A method is needed to batch process SAR images. Summary of the invention
[0007] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a SAR image despeckling and denoising method based on a DnCNN pre-training model, which solves the problem of high cost of obtaining SAR image training sets in the prior art.
[0008] To achieve the above object, the present invention provides the following solutions:
[0009] A SAR image despeckling and denoising method based on a DnCNN pre-training model, comprising:
[0010] Acquire a noisy SAR image to be processed, and input the SAR image to be processed into a constructed image preprocessing model to obtain a preprocessed image data set;
[0011] Inputting the preprocessed image data set into the DnCNN pre-trained model to obtain a clean SAR image;
[0012] The method for constructing the image preprocessing model is:
[0013] Obtain an initial SAR image set and calculate the corresponding initial intensity map;
[0014] Adding speckle noise to the initial intensity map according to the number of views to obtain a noise atlas;
[0015] Performing logarithmic transformation on the noise atlas to obtain a logarithmic intensity map;
[0016] Determine a normalization factor according to the number of views and add the normalization factor to the logarithmic intensity map to obtain a set of transformed images;
[0017] Based on the CNN network architecture, an image preprocessing model is constructed with the initial SAR image set as input and the transformed image set as input.
[0018] Preferably, the relationship between the image and noise of the initial SAR image set is expressed as:
[0019] Y=N×X,
[0020] Among them, Y is the image in the initial SAR image set, N is the speckle noise, and X is the clean image.
[0021] Preferably, the expression of the clean image in the logarithmic intensity map after logarithmic transformation is:
[0022]
[0023] Among them, f is the DnCNN pre-trained model, L is the number of views, and ψ is the digamma function.
[0024] Preferably, determining a normalization factor according to the number of views and adding the normalization factor to the logarithmic intensity map comprises:
[0025] Calculate the input variance;
[0026] Approximately process the input method to obtain an approximate variance;
[0027] A normalization factor is obtained according to the approximate variance.
[0028] Preferably, the calculation expression of the input variance is:
[0029]
[0030] Among them, σ * is the input variance, Y min is the 0.3% quantile of Y, Y max is the 99.7% quantile of Y.
[0031] Preferably, the expression of the approximate variance is:
[0032] σ=floor(σ * / 5)×5;
[0033] Here, σ is the approximate variance.
[0034] Preferably, the expression of the normalization factor is:
[0035]
[0036] Among them, NF is the normalization factor.
[0037] The present invention discloses the following technical effects:
[0038] The present invention provides a SAR image despeckle and denoising method based on a DnCNN pre-training model, comprising: obtaining a noisy SAR image to be processed, and inputting the SAR image to be processed into a constructed image preprocessing model to obtain a preprocessed image data set; inputting the preprocessed image data set into the DnCNN pre-training model to obtain a clean SAR image; the image preprocessing model is constructed by: obtaining an initial SAR image set and calculating a corresponding initial intensity map; adding speckle noise to the initial intensity map according to the number of views to obtain a noise map set; performing a logarithmic transformation on the noise map set to obtain a logarithmic intensity map; determining a normalization factor according to the number of views and adding it to the logarithmic intensity map to obtain a converted image set; based on a CNN network architecture, taking the initial SAR image set as input and taking the converted image set as input to construct an image preprocessing model. By constructing an image preprocessing model, the present invention greatly reduces the learning cost required for using the MuLoG-DnCNN method, quickly obtains preprocessed SAR image data, and conveniently and quickly uses the DnCNN model to denoise and obtain a relatively clean SAR image. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0040] Figure 1 A flow chart of a SAR image despeckle and denoising method based on a DnCNN pre-trained model provided in an embodiment of the present invention;
[0041] Figure 2 A detailed schematic diagram of a SAR image despeckling and denoising method based on a DnCNN pre-trained model provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0043] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0044] like Figure 1 As shown, the present invention provides a SAR image despeckle and denoising method based on a DnCNN pre-training model, comprising:
[0045] Step 100: obtaining a noisy SAR image to be processed, and inputting the SAR image to be processed into a constructed image preprocessing model to obtain a preprocessed image data set;
[0046] Step 200: input the pre-processed image data set into the DnCNN pre-trained model to obtain a clean SAR image;
[0047] The method for constructing the image preprocessing model is:
[0048] Step 101: Acquire an initial SAR image set and calculate a corresponding initial intensity map;
[0049] Step 102: adding speckle noise to the initial intensity map according to the number of views to obtain a noise atlas;
[0050] Step 103: performing logarithmic transformation on the noise atlas to obtain a logarithmic intensity map;
[0051] Step 104: determining a normalization factor according to the number of views and adding the normalization factor to the logarithmic intensity map to obtain a transformed image set;
[0052] Step 105: Based on the CNN network architecture, an image preprocessing model is constructed with the initial SAR image set as input and the transformed image set as input.
[0053] like Figure 2 As shown, the embodiment of the present invention also provides a detailed process of a SAR image despeckle and denoising method based on a DnCNN pre-training model:
[0054] Step 1: Prepare noisy images in the training set data
[0055] Obtain 400 clean SAR image files or ordinary grayscale images; calculate the intensity map of the image, and add speckle noise to the image with the number of views L being 1-5 in sequence; save the noisy images, a total of 400*5=2000 images;
[0056] Step 2: Convert data distribution to obtain converted image
[0057] Read the 2000 saved noise images; perform logarithmic transformation on the images to convert the multiplicative speckle noise into additive noise. This step is because the DnCNN pre-trained model is trained for additive noise, and the multiplicative noise must be converted into additive noise before the DnCNN pre-trained model can be applied; the original relationship between the image and the noise (the relationship between the image and the noise of the initial SAR image set) is as follows: Y = N × X, where Y is the noisy image, N is the speckle noise, and X is the clean image. After logarithmic transformation, the composition of the clean image X (the clean image in the logarithmic intensity map after logarithmic transformation) becomes: Where f is the DnCNN pre-trained model, L is the number of views, and ψ is the digamma function. Normalize the logarithmic transformed image. The purpose of this step is to reduce the heavy tail effect. The normalization formula is as follows: Input = (YY min ) / (Y max -Y min ), where Y min Take the 0.3% quantile of Y, Y max Take the 99.7% quantile of Y. Finally, add the normalization factor NF to get the transformed image. The steps to calculate NF are as follows: First, assume that the input variance of the homogeneous region is similar to the variance of the training dataset, and calculate the input variance σ * : Since the σ range of the pre-trained DnCNN model is 10-75 with a step size of 5, we need to use the floor function to approximate the calculated variance to make it fit the range of the pre-trained model: σ = floor (σ * / 5)×5, finally, the normalization factor NF is calculated and added to the image: Save the converted images as a test set, totaling 400*5=2000 images.
[0058] Step 3: Effect verification
[0059] The converted image is directly input into the DnCNN pre-trained model to obtain the denoised image; the denoised image is inversely normalized to obtain the original size; the image is debiased; the inverse logarithmic transformation is performed and the data is rounded to obtain the final denoised image; PSNR, SSIM, and ENL are calculated to verify the denoising effect.
[0060] Step 4: Train the CNN model
[0061] Use the noisy image in step 1 as the model input and the converted image in step 2 as the model output; build a CNN network architecture with convolutional layers and pooling layers; process the training set images (a total of 2000+2000=4000 images) to obtain the model file; save the model file trainedModel.mat.
[0062] Step 5: Effect verification
[0063] Input another part of the noise image into the CNN model, obtain the converted image and repeat step 3 to verify the effect.
[0064] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.
[0065] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of the present invention. At the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.
Claims
1. A SAR image despeckle and denoising method based on a DnCNN pre-trained model, characterized in that: include: Acquire a noisy SAR image to be processed, and input the SAR image to be processed into a constructed image preprocessing model to obtain a preprocessed image data set; Inputting the preprocessed image data set into the DnCNN pre-trained model to obtain a clean SAR image; The method for constructing the image preprocessing model is: Obtain an initial SAR image set and calculate the corresponding initial intensity map; Adding speckle noise to the initial intensity map according to the number of views to obtain a noise atlas; Performing logarithmic transformation on the noise atlas to obtain a logarithmic intensity map; Determine a normalization factor according to the number of views and add the normalization factor to the logarithmic intensity map to obtain a set of transformed images; Based on the CNN network architecture, an image preprocessing model is constructed with the initial SAR image set as input and the transformed image set as input.
2. The SAR image despeckling and denoising method based on the DnCNN pre-training model according to claim 1, characterized in that: The relationship between the image and noise of the initial SAR image set is expressed as: Y=N×X, Among them, Y is the image in the initial SAR image set, N is the speckle noise, and X is the clean image.
3. The SAR image despeckle and denoising method based on the DnCNN pre-training model according to claim 2, characterized in that: The expression of the clean image in the logarithmic intensity map after logarithmic transformation is: Among them, f is the DnCNN pre-trained model, L is the number of views, and ψ is the digamma function.
4. The SAR image despeckling and denoising method based on the DnCNN pre-training model according to claim 3 is characterized in that: The determining of the normalization factor according to the number of views and adding the normalization factor to the logarithmic intensity map comprises: Calculate the input variance; Approximately process the input method to obtain an approximate variance; A normalization factor is obtained according to the approximate variance.
5. The SAR image despeckle and denoising method based on the DnCNN pre-training model according to claim 4, characterized in that: The calculation expression of the input variance is: Among them, σ * is the input variance, Y min is the 0.3% quantile of Y, Y max is the 99.7% quantile of Y.
6. The SAR image despeckling and denoising method based on the DnCNN pre-training model according to claim 5, characterized in that: The expression of the approximate variance is: σ=floor(σ * / 5)×5; Here, σ is the approximate variance.
7. The SAR image despeckling and denoising method based on the DnCNN pre-training model according to claim 6, characterized in that: The expression of the normalization factor is: Among them, NF is the normalization factor.
Citation Information
Patent Citations
Incremental time sequence SAR image noise reduction method based on sparse expression
CN116883254A
Generative anti-SAR image speckle noise suppression method based on expansion residual error
CN118429208A
Image processing apparatus, image processing method, and computer-readable medium
US20230206404A1
Cited By
Neural network prediction method and device for two-dimensional quantization vortex dynamics
CN119720758A
Method and device for enhancing fatty liver image
CN120807327A
SAR (Synthetic Aperture Radar) image denoising method and device based on logarithm domain diffusion model
CN121280264A
A SAR image denoising method and device based on a logarithmic domain diffusion model
CN121280264B
Training system and training method of video noise reduction model, and related products
CN122265073A
