A NSST domain sonar image denoising method based on density clustering and gray scale transformation

Abstract

The application discloses a NSST domain sonar image denoising method based on density clustering and gray scale transformation, and belongs to the technical field of sonar image denoising.The application solves the problems of poor denoising effect, poor edge feature maintaining capability and low denoising efficiency of the existing sonar image denoising method.The application can improve the low contrast and serious noise interference of a sonar image, decomposes the noisy sonar image through NSST transformation, removes noise signals in high-frequency coefficients by using density clustering, retains detail signals, performs gray scale transformation on low-frequency coefficients, improves image contrast, and finally performs NSST inverse transformation to obtain a denoised sonar image.The method can be applied to denoising of a sonar image.

Description

Technical Field

[0001] This invention belongs to the field of sonar image denoising technology, specifically relating to a sonar image denoising method in the NSST domain based on density clustering and grayscale transformation. Background Technology

[0002] Due to the influence of sea breezes, ocean currents, water temperature, impurities, and imaging equipment, images obtained using sonar detection technology often contain various types of noise, mainly manifested as granular speckles. The presence of speckle noise severely degrades image quality and adversely affects subsequent image processing. Currently, commonly used denoising methods mainly include spatial domain denoising and transform domain denoising. Spatial domain denoising typically achieves denoising by directly modifying image pixel values ​​using a sliding window method, ignoring the inherent characteristics of the pixels themselves. This results in significant loss of image detail after denoising, leading to poor denoising performance. In transform domain denoising, methods such as wavelet transform can only represent horizontal, vertical, and diagonal directions, failing to optimally approximate the singularity locations of high-dimensional data; ridge transform can only represent linear singularities, failing to effectively describe the curved singularities in signals; and curve transform has a complex structure and high computational cost. These methods result in inaccurate capture of image edge features and low denoising efficiency during denoising.

[0003] In summary, existing sonar image denoising methods still suffer from poor denoising performance, poor edge feature preservation, and low denoising efficiency. Therefore, effectively removing noise from sonar images under severe noise interference, improving edge feature preservation, and enhancing denoising efficiency are urgent problems to be solved. Summary of the Invention

[0004] The purpose of this invention is to address the problems of poor denoising effect, poor edge feature preservation, and low denoising efficiency in existing sonar image denoising methods, and to propose a sonar image denoising method based on density clustering and grayscale transformation in the NSST domain.

[0005] The technical solution adopted by this invention to solve the above-mentioned technical problems is: a method for denoising NSST domain sonar images based on density clustering and grayscale transformation, the method specifically includes the following steps:

[0006] Step 1: Decompose the acquired underwater noisy nacelle image to obtain high-frequency subband image and low-frequency subband image, and calculate the NSST coefficients of the high-frequency subband image and low-frequency subband image respectively;

[0007] Step 2: Cluster the NSST coefficients of each high-frequency sub-band image, remove the NSST coefficients representing noise signals from the NSST coefficients of each high-frequency sub-band image, and obtain the NSST coefficients representing detail signals from the NSST coefficients of each high-frequency sub-band image.

[0008] Step 3: Process the NSST coefficients representing the detail signals corresponding to each high-frequency sub-band image to obtain the processed NSST coefficients corresponding to each high-frequency sub-band image.

[0009] Step 4: Perform grayscale stretching on the NSST coefficients of the low-frequency sub-band image using grayscale transformation to obtain the grayscale stretched NSST coefficients corresponding to the low-frequency sub-band image.

[0010] Step 5: Reconstruct the NSST coefficients processed in Step 3 and the NSST coefficients stretched in Step 4 using the inverse NSST transform to obtain the denoised sonar image.

[0011] Furthermore, in step one, the acquired underwater noisy nacelle image is decomposed to obtain a high-frequency subband image and a low-frequency subband image, and the NSST coefficients of the high-frequency subband image and the low-frequency subband image are calculated respectively; the method used is NSST transform.

[0012] Furthermore, the specific process of step one is as follows:

[0013] Step 1: Perform an upsampling operation on the Laplace pyramid filter bank to construct a two-channel non-downsampling Laplace pyramid filter bank;

[0014] Step 1 and 2: Based on the shear wave, the acquired underwater noisy nag image is decomposed into one low-frequency sub-band image and four high-frequency sub-band images using a two-channel non-subsampled Laplace pyramid filter bank, and discrete Fourier transform is performed on the low-frequency sub-band image and each high-frequency sub-band image respectively.

[0015] Step 13: Construct the shear filter W in the pseudo-polarized grid coordinate system;

[0016] Step 14: Map the shear filter W from the pseudo-polarized grid coordinate system to the Cartesian coordinate system to obtain the new shear filter W′;

[0017] Step 15: Apply the new shearing filter W′ to the four high-frequency sub-band images obtained from the discrete Fourier transform in Step 1 and 2 respectively, and obtain the directional sub-band of each high-frequency sub-band image.

[0018] Step 16: Perform inverse discrete Fourier transform on the directional sub-band corresponding to each high-frequency sub-band image to obtain the NSST coefficients corresponding to each high-frequency sub-band image.

[0019] Perform an inverse discrete Fourier transform on the low-frequency subband image to obtain the NSST coefficients corresponding to the low-frequency subband image.

[0020] Furthermore, the shear wave is a synthetic wavelet.

[0021] Furthermore, the synthesized wavelet is:

[0022]

[0023] Where, ψ j，k，l (x) represents the synthesized wavelet, x is the input signal, A and B are both two-dimensional invertible matrices, |det A| is the determinant of matrix A, j is the scaling parameter, l is the geometric parameter, k is the translation parameter, Z represents an integer, and A j Let A be the j-th power of B. l Let B be the l-th power of matrix B.

[0024] Furthermore, the synthesized wavelet ψ j，k，l (x) satisfies the condition: ψ j，k，l (x)∈L 2 (R2) and for All guaranteed Among them, L 2 (R 2 ) represents the finite energy condition of the function, ||·|| represents the norm, and <·> represents the inner product.

[0025] Furthermore, the matrix matrix

[0026] Furthermore, in step two, the clustering process uses the DBSCAN density clustering method.

[0027] Furthermore, in step three, the NSST coefficients representing detail signals corresponding to each high-frequency sub-band image are processed by thresholding, removing coefficients smaller than the threshold from the NSST coefficients representing detail signals; the grayscale transformation in step four uses a piecewise function.

[0028] Furthermore, the specific process of step five is as follows:

[0029] Step 51: Perform Discrete Fourier Transform on the processed NSST coefficients corresponding to the i-th high-frequency sub-band image to obtain the components of the i-th high-frequency sub-band image in each direction, i = 1, 2, 3, 4;

[0030] Step 5.2: Perform inverse bandpass filtering on the components of the i-th layer high-frequency sub-band image in each direction to obtain the inverse bandpass filtering result. i = 1, 2, 3, 4;

[0031] Step 53: Analyze the results of the inverse bandpass filter. Perform inverse discrete Fourier transform to obtain the high-pass image. i = 1, 2, 3, 4;

[0032] Step 54: Perform Discrete Fourier Transform on the gray-scale stretched NSST coefficients corresponding to the low-frequency sub-band image to obtain the low-frequency sub-band image components.

[0033] Step 55: Perform inverse discrete Fourier transform on the low-frequency subband image components to obtain the low-pass images of each layer. i = 1, 2, 3, 4;

[0034] Steps 5 and 6: Reconstruct the high-pass image using the non-subsampled Laplacian pyramid algorithm. and low-pass image Reconstructing the image using i = 1, 2, 3, 4 yields the denoised sonar image.

[0035] The beneficial effects of this invention are:

[0036] This invention combines DBSCAN density clustering, grayscale transformation, and NSST transformation to maximize the denoising effect, edge feature preservation, and denoising efficiency of sonar images. This invention addresses the problems of low contrast and severe noise interference in sonar images. It decomposes the noisy sonar image using NSST transformation, then applies density clustering to remove noise signals from high-frequency coefficients while preserving detail signals. Grayscale transformation is then applied to low-frequency coefficients to enhance image contrast. Finally, an inverse NSST transformation is performed to obtain the denoised sonar image. Attached Figure Description

[0037] Figure 1 This is a flowchart of an NSST domain sonar image denoising method based on density clustering and grayscale transformation according to the present invention.

[0038] Figure 2 This is a schematic diagram of the NSST decomposition.

[0039] Figure 3 This is a diagram illustrating the working principle of DBSCAN density clustering.

[0040] Figure 4 This is a schematic diagram of grayscale transformation;

[0041] Figure 5 Images of underwater shipwreck wreckage;

[0042] Figure 6 To use the method of the present invention to Figure 5 An image of the underwater shipwreck wreckage after noise reduction. Detailed Implementation

[0043] Specific implementation method one, such as Figure 1 As shown in this embodiment, a method for denoising NSST domain sonar images based on density clustering and grayscale transformation specifically includes the following steps:

[0044] Step 1: Decompose the acquired underwater noisy nacelle image to obtain high-frequency subband image and low-frequency subband image, and calculate the NSST coefficients of the high-frequency subband image and low-frequency subband image respectively;

[0045] Step 2: Cluster the NSST coefficients of each high-frequency sub-band image, remove the NSST coefficients representing noise signals from the NSST coefficients of each high-frequency sub-band image, and obtain the NSST coefficients representing detail signals from the NSST coefficients of each high-frequency sub-band image.

[0046] Step 3: Process the NSST coefficients representing the detail signals corresponding to each high-frequency sub-band image to obtain the processed NSST coefficients corresponding to each high-frequency sub-band image.

[0047] Step 4: Perform grayscale stretching on the NSST coefficients of the low-frequency sub-band image using grayscale transformation to obtain the grayscale stretched NSST coefficients corresponding to the low-frequency sub-band image.

[0048] Step 5: Reconstruct the NSST coefficients processed in Step 3 and the NSST coefficients stretched in Step 4 using the inverse NSST transform to obtain the denoised sonar image.

[0049] The denoising target of the present invention can be underwater sonar images. Moreover, the present invention can achieve good denoising effect and edge preservation capability even when the sonar image is heavily polluted by noise.

[0050] Specific Implementation Method Two: This implementation method differs from Specific Implementation Method One in that, in step one, the acquired underwater noisy nacelle image is decomposed to obtain a high-frequency sub-band image and a low-frequency sub-band image, and the NSST coefficients of the high-frequency sub-band image and the low-frequency sub-band image are calculated respectively; the method used is NSST transform (non-subsampled shear wave transform).

[0051] Based on sheared waves, a multi-scale decomposition is performed using a non-subsampled Laplace pyramid filter bank. In directional localization, the standard shear filter is mapped from the pseudo-pole to the grid coordinate system and then to the Cartesian coordinate system. A mathematical model is then used to complete the relevant operations through two-dimensional convolution using inverse Fourier transform.

[0052] The other steps and parameters are the same as in Specific Implementation Method 1.

[0053] Specific Implementation Method Three: This implementation method differs from Specific Implementation Method One or Two in that the specific process of step one is as follows:

[0054] Step 1: Perform an upsampling operation on the Laplace pyramid filter bank to construct a two-channel non-downsampling Laplace pyramid filter bank;

[0055] Step 1 and 2: Based on the shear wave, the acquired underwater noisy nag image is decomposed into one low-frequency sub-band image and four high-frequency sub-band images using a two-channel non-subsampled Laplace pyramid filter bank, and discrete Fourier transform is performed on the low-frequency sub-band image and each high-frequency sub-band image respectively.

[0056] Step 13: Construct a standard shear filter W in the pseudo-polarized grid coordinate system;

[0057] Step 14: Map the shear filter W from the pseudo-polarized grid coordinate system to the Cartesian coordinate system to obtain the new shear filter W′;

[0058] Step 15: Apply the new shearing filter W′ to the four high-frequency sub-band images obtained from the discrete Fourier transform in Step 1 and 2 respectively, and obtain the directional sub-band of each high-frequency sub-band image.

[0059] The number of directional sub-bands corresponding to each high-frequency sub-band image is 2. 3 2 3 2 4 and 2 4 indivual;

[0060] Step 16: Perform inverse discrete Fourier transform on the directional sub-band corresponding to each high-frequency sub-band image to obtain the NSST coefficients corresponding to each high-frequency sub-band image.

[0061] Perform an inverse discrete Fourier transform on the low-frequency subband image to obtain the NSST coefficients corresponding to the low-frequency subband image.

[0062] Other steps and parameters are the same as in specific implementation method one or two.

[0063] Specific Implementation Method Four: This implementation method differs from Specific Implementation Methods One to Three in that the shearing wave is a synthetic wavelet.

[0064] The other steps and parameters are the same as those in one of the specific implementation methods one to three.

[0065] Specific Implementation Method Five: This implementation method differs from Specific Implementation Methods One to Four in that the synthesized wavelet is:

[0066]

[0067] Where, ψ j，k，l (x) represents the synthesized wavelet, x is the input signal, A and B are both two-dimensional invertible matrices, |det A| is the determinant of matrix A, j is the scaling parameter, l is the geometric parameter, k is the translation parameter, Z represents an integer, and A j Let A be the j-th power of matrix A. jRepresents scaling transformation, B l Let Bl be the l-th power of matrix B, where Bl represents the geometric transformation.

[0068] The other steps and parameters are the same as those in one of the specific implementation methods one to four.

[0069] Specific Implementation Method Six: This implementation method differs from Specific Implementation Methods One to Five in that the synthesized wavelet ψ j，k，l (x) satisfies the condition: ψ j，k，l (x)∈L 2 (R 2 And for any function All guaranteed Among them, L 2 (R 2 ) represents the finite energy condition of the function, ||·|| represents the norm, and <·> represents the inner product.

[0070] The other steps and parameters are the same as those in one of the specific implementation methods one to five.

[0071] Specific Implementation Method Seven: This implementation method differs from Specific Implementation Methods One through Six in that the matrix... matrix

[0072] The other steps and parameters are the same as those in one of the specific implementation methods one to six.

[0073] Specific Implementation Method Eight: This implementation method differs from Specific Implementation Methods One to Seven in that, in step two, the clustering process uses the DBSCAN density clustering method.

[0074] The specific process of the DBSCAN density clustering method is as follows:

[0075] Step 2: Define a cluster as the largest set of all closely connected data points;

[0076] Step 22: Within a region centered on a certain data point and with Eps as the radius, DBSCAN uses a region query to determine whether the number of data points contained in the region is greater than the set threshold MinPts.

[0077] Steps 2 and 3: If the number of data points in the neighborhood of Eps is greater than the threshold MinPts, then create a new cluster;

[0078] Step 24: Merge related clusters;

[0079] Step 25: If an object is not in any cluster, treat that object as a noise point.

[0080] The other steps and parameters are the same as those in any of the specific implementation methods one to seven.

[0081] Specific Implementation Method Nine: This implementation method differs from Specific Implementation Methods One to Eight in that, in step three, the NSST coefficients representing detail signals corresponding to each high-frequency sub-band image are processed by thresholding, removing coefficients smaller than the threshold from the NSST coefficients representing detail signals; the grayscale transformation in step four uses a piecewise function, allowing for segmental adjustment of image grayscale values ​​as needed.

[0082] The other steps and parameters are the same as those in one of the specific implementation methods one to eight.

[0083] Specific Implementation Method Ten: This implementation method differs from Specific Implementation Methods One to Nine in that the specific process of step five is as follows:

[0084] Step 51: Perform Discrete Fourier Transform on the processed NSST coefficients corresponding to the i-th high-frequency sub-band image to obtain the components of the i-th high-frequency sub-band image in each direction, i = 1, 2, 3, 4.

[0085] Step 5.2: Perform inverse bandpass filtering on the components of the i-th layer high-frequency sub-band image in each direction to obtain the inverse bandpass filtering result. i = 1, 2, 3, 4;

[0086] Step 53: Analyze the results of the inverse bandpass filter. Perform inverse discrete Fourier transform to obtain the high-pass image. i = 1, 2, 3, 4;

[0087] Step 54: Perform Discrete Fourier Transform on the gray-scale stretched NSST coefficients corresponding to the low-frequency sub-band image to obtain the low-frequency sub-band image components.

[0088] Step 55: Perform inverse discrete Fourier transform on the low-frequency subband image components to obtain the low-pass images of each layer. i = 1, 2, 3, 4;

[0089] Steps 5 and 6: Reconstruct the high-pass image using the non-subsampled Laplacian pyramid algorithm. and low-pass image Reconstructing the image using i = 1, 2, 3, 4 yields the denoised sonar image.

[0090] The other steps and parameters are the same as those in any of the specific implementation methods one to nine.

[0091] Example

[0092] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0093] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the scope of the invention.

[0094] This embodiment of an underwater sonar image denoising method based on image processing includes: performing NSST transformation on the acquired sonar image to decompose it into high-frequency coefficients and low-frequency coefficients; processing the high-frequency coefficients using DBSCAN density clustering to remove noise signals and retain detail signals; performing grayscale transformation on the low-frequency coefficients to improve image contrast; and finally performing inverse NSST transformation to obtain the denoised sonar image.

[0095] A denoising experiment was conducted using images of shipwreck wreckage as an example. The image denoising method and process are as follows: Figure 1 As shown:

[0096] like Figure 2 The diagram shows the NSST decomposition process. In the diagram, NSPF is the non-subsampled Laplacian pyramid filter, and SF is the shearing filter. NSPF is used to perform multi-scale decomposition on the original image, and SF is used for directional localization.

[0097] like Figure 3 The diagram shows the working principle of DBSCAN density clustering. The area within the dashed circle is the Eps neighborhood, x1 is the core object, x2 is directly accessible from the x1 density, x3 is accessible from the x1 density, and x3 is connected to the x4 density.

[0098] like Figure 4 The diagram shown illustrates grayscale transformation, where the output grayscale level can be adjusted based on the input grayscale level.

[0099] like Figure 5 The image shown is of a noisy shipwreck. The image suffers from severe noise interference and low contrast, which affects the overall quality and visual appeal of the image.

[0100] Therefore, this embodiment uses the NSST algorithm to denoise the image.

[0101] Step 1: Decompose the noisy natron image using NSST to obtain high-frequency subband and low-frequency subband images. The specific steps are as follows:

[0102] Step 1: Define the two-dimensional affine system as follows:

[0103]

[0104] Where ψ∈L 2 (R 2 A and B are both two-dimensional invertible matrices, and |detB|=1, j and l are the scale parameter and geometric parameter, respectively, and k is the translation parameter.

[0105] Step 2: Let A in formula (1) AB (ψ) satisfies the condition of a tightly supported frame, i.e.:

[0106]

[0107] Then A in formula (1) AB The element ψ in the (ψ) system is called the synthetic wavelet, where the expansion matrix A j Represents scaling transformation, matrix B l Representing geometric transformations, the aforementioned affine transformation system can be used to construct tightly supported frames of various scales and directions.

[0108] Step 3: In formula (1) It constitutes a shear wave.

[0109] Step 4: Perform an upsampling operation on the Laplace pyramid filter bank to construct a two-channel non-downsampling pyramid filter bank;

[0110] Step 5: Decompose the noisy image into one low-frequency subband image and four high-frequency subband images using a non-subsampled Laplacian pyramid filter bank;

[0111] Step 6: Construct a shear filter W in the pseudo-polarized grid coordinate system;

[0112] Step 7: Remap the shear filter W from the pseudo-polarized grid coordinate system to the Cartesian coordinate system to obtain the new shear filter W′;

[0113] Step 8: Apply the shearing filter W′ to the 4-layer high-frequency sub-band images from steps one and two, and obtain 2 3 2 3 2 4 and 2 4 One directional sub-band;

[0114] Step 9: Perform inverse Fourier transform on the subbands obtained in each direction to generate NSST coefficients.

[0115] Step 2: Use DBSCAN density clustering to cluster the high-frequency subband image, dividing the signal into detail signals and noise signals. The specific steps are as follows:

[0116] Step 1: Define a cluster as the largest set of all closely connected data points;

[0117] Step 2: Within a region centered on a certain data point and with Eps as the radius, DBSCAN uses a region query to determine whether the number of data points contained in the region is greater than the set threshold MinPts.

[0118] Step 3: If the number of data points in the neighborhood of Eps is greater than the threshold MinPts, then create a new cluster;

[0119] Step 4: Merge related clusters;

[0120] Step 5: If an object is not in any cluster, treat it as a noise point.

[0121] Step 3: Based on the detail signal and noise signal obtained in Step 2, perform threshold processing and zero-value processing respectively.

[0122] Step 4: Use grayscale transformation to stretch the low-frequency sub-band image to improve image contrast. The grayscale transformation is a piecewise function, and the image grayscale values ​​can be adjusted segment by segment as needed.

[0123] Step 5: Reconstruct the high-frequency subband and low-frequency subband images processed in steps 2 to 4 using the inverse NSST transform, to obtain the following... Figure 6 The denoised image shown is produced through the following steps:

[0124] Step 1: Perform a discrete Fourier transform on the NSST coefficients of the i-th layer to obtain the components in each direction, i = 1, 2, 3, 4;

[0125] Step 2: Perform inverse bandpass filtering on the components in each direction to obtain the matrix.

[0126] Step 3: For the matrix Perform inverse discrete Fourier transform to obtain the high-pass image.

[0127] Step 4: Perform Discrete Fourier Transform on the grayscale stretched NSST coefficients corresponding to the low-frequency sub-band image to obtain the low-frequency sub-band image components; perform Inverse Discrete Fourier Transform on the low-frequency sub-band image components to obtain the low-pass images of each layer. i = 1, 2, 3, 4;

[0128] High-pass image reconstruction using non-subsampled Laplacian pyramid algorithm and low-pass image The image is then reconstructed to obtain the denoised image.

[0129] Experiments show that the present invention can improve the overall image quality and intuitive visual effect, enhance the image edge preservation ability, reduce the amount of computation, and improve the denoising efficiency after denoising images of underwater shipwrecks.

[0130] The above examples of the present invention are merely illustrative of the computational model and process of the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is impossible to exhaustively list all possible implementations here. Any obvious variations or modifications derived from the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for denoising sonar images in NSST domain based on density clustering and gray transformation, characterized in that, The method specifically comprises the following steps: Step one, decompose the obtained underwater noisy sonar image to obtain a high-frequency sub-band image and a low-frequency sub-band image, and calculate NSST coefficients of the high-frequency sub-band image and the low-frequency sub-band image respectively; The method used in the step one is NSST transformation; The specific process of the step one is as follows: Step one, up-sampling operation is performed on the Laplacian pyramid filter set to construct a two-channel non-subsampled Laplacian pyramid filter set; Step two, based on the shearlet, the two-channel non-subsampled Laplacian pyramid filter set is used to decompose the obtained underwater noisy sonar image into a 1-layer low-frequency sub-band image and 4-layer high-frequency sub-band images, and discrete Fourier transform is performed on the low-frequency sub-band image and each layer of high-frequency sub-band images respectively; The shearlet is a composite wavelet, and the composite wavelet is as follows: wherein ψ j，k，l (x) is a synthesis wavelet, x is an input signal, A and B are both two-dimensional invertible matrices, |detA| is the value of the determinant of matrix A, j is a scale parameter, 1 is a geometry parameter, k is a translation parameter, Z represents an integer, A j is the jth power of matrix A, B l is the lth power of matrix B; The matrix Matrix The synthesis wavelet ψ j，k，l (x) satisfies the condition: ψ j，k，l (x) ∈ L 2 (R 2 ) and for all guarantees where L 2 (R 2 ) represents the function energy limited condition, ||·|| represents the norm, and <·> represents the inner product. Step three, a shear filter W is constructed in a pseudo-polar grid coordinate system; Step four, the shear filter W is mapped from the pseudo-polar grid coordinate system to the Cartesian coordinate system to obtain a new shear filter W'; Step five, the new shear filter W' is respectively applied to the 4-layer high-frequency sub-band images obtained after the discrete Fourier transform in step two to obtain a directional sub-band of each layer of high-frequency sub-band images; Step six, inverse discrete Fourier transform is respectively performed on the directional sub-band corresponding to each layer of high-frequency sub-band images to obtain NSST coefficients corresponding to each layer of high-frequency sub-band images; Inverse discrete Fourier transform is performed on the low-frequency sub-band image to obtain NSST coefficients corresponding to the low-frequency sub-band image; Step two, clustering processing is respectively performed on the NSST coefficients of each layer of high-frequency sub-band images, the NSST coefficients representing noise signals in the NSST coefficients of each layer of high-frequency sub-band images are removed, and the NSST coefficients representing detail signals in the NSST coefficients of each layer of high-frequency sub-band images are respectively obtained; Step three, the NSST coefficients representing detail signals corresponding to each layer of high-frequency sub-band images are respectively processed to obtain processed NSST coefficients corresponding to each layer of high-frequency sub-band images; Step four, the NSST coefficients of the low-frequency sub-band image are subjected to gray scale stretching by using a gray scale transformation to obtain gray scale stretched NSST coefficients corresponding to the low-frequency sub-band image; Step five, the NSST coefficients processed in step three and the NSST coefficients subjected to gray scale stretching in step four are reconstructed by using NSST inverse transformation to obtain a denoised sonar image.

2. The NSST domain sonar image denoising method based on density clustering and gray scale transformation according to claim 1, characterized in that, In the step two, the clustering processing adopts a DBSCAN density clustering method.

3. The NSST domain sonar image denoising method based on density clustering and gray transformation according to claim 2, characterized in that, In the step three, the NSST coefficients representing detail signals corresponding to each layer of high-frequency sub-band images are respectively processed by threshold processing; and the gray scale transformation in the step four adopts a piecewise function.

4. The NSST domain sonar image denoising method based on density clustering and gray transformation of claim 3, characterized in that, The specific process of the step five is as follows: Step five, discrete Fourier transform is performed on the processed NSST coefficients corresponding to the i-layer high-frequency sub-band image to obtain components of the i-layer high-frequency sub-band image in each direction, i=1, 2, 3, 4; Step five two, inverse band-pass filtering operation is performed on the components of the i-th layer high frequency sub-band image in each direction to obtain inverse band-pass filtering result i = 1, 2, 3, 4; Step five three, inverse band-pass filtering result performing inverse discrete Fourier transform to obtain a high-pass image i = 1, 2, 3, 4; Step five, discrete Fourier transform is performed on the gray scale stretched NSST coefficients corresponding to the low-frequency sub-band image to obtain a low-frequency sub-band image component; Step five, inverse discrete Fourier transform is performed on the low frequency subband image components to obtain low pass images of each layer i = 1, 2, 3, 4; Steps 5 and 6: Reconstruct the high-pass image using the non-subsampled Laplacian pyramid algorithm. and low-pass image Reconstructing the image using i = 1, 2, 3, 4 yields the denoised sonar image.

Images