Artificial fish reef posture acoustic evaluation method

Abstract

This invention discloses an acoustic assessment method for the attitude of artificial reefs. Based on side-scan sonar images, noise reduction is performed using two-dimensional empirical mode decomposition. Targets and their shadow contours in the images are extracted using fuzzy C-means and horizontal clustering algorithms. The image shadow contours are matched with shadow contours from a constructed 3D reef model library to estimate the reef's tilt angle. The height and subsidence of the reef are estimated based on the geometric relationship between the towed fish, the target length, and the shadow length, thus achieving the assessment of the artificial reef's attitude. This invention leverages the fast computation speed of fuzzy C-means clustering and the high segmentation accuracy of horizontal clustering to achieve rapid and accurate target segmentation in side-scan sonar images. Matching the target shadow contours in the side-scan sonar images with the projected contours of the constructed 3D artificial reef model enables the assessment of the artificial reef's attitude.

Description

Technical Field

[0001] This invention relates to the field of acoustic evaluation technology, and in particular to an acoustic evaluation method for the attitude of artificial reefs. Background Technology

[0002] Artificial reefs are an effective means of maintaining fishery resources and restoring the ecological environment of coastal waters, and have been constructed extensively along my country's coast. However, limited understanding of the quality of artificial reef construction hinders effective fisheries and aquaculture management.

[0003] Side-scan sonar is an advanced acoustic instrument that can generate large-scale maps of the seabed environment at high resolution. It is widely used for detecting underwater targets such as seabed reefs, shipwrecks, pipelines, and cables.

[0004] Existing technologies mainly utilize side-scan sonar to assess the height and volume of artificial reefs, with research focusing on the post-construction quality assessment stage, such as benthic habitat assessment and reef and reef mass state estimation. However, there is no suitable assessment method for using side-scan sonar to assess the attitude of artificial reefs, resulting in an incomplete understanding of artificial reefs and hindering fisheries and aquaculture management. Summary of the Invention

[0005] To overcome the aforementioned problems in the existing technology, this invention proposes a method for acoustic evaluation of the attitude of artificial reefs.

[0006] The technical solution adopted by this invention to solve its technical problem is: an acoustic evaluation method for the attitude of artificial reefs, comprising the following steps:

[0007] Step 1, Data Acquisition: Acquire seabed acoustic data using side-scan sonar;

[0008] Step 2, Target Extraction: Extract the image of the area to be tested from the data obtained in Step 1, perform image processing on the extracted image including image filtering, coarse target extraction, fine target extraction and target shadow matching, and store the position information of the target and the target shadow;

[0009] Step 3, Feature parameter calculation: Calculate the maximum height of each target and the seven moment vectors of the corresponding shadow based on the target and shadow position coordinates. Extract the tow height, target position and shape of the artificial reef from the image processing results. Construct a three-dimensional reef model. Calculate the artificial reef height and shadow outline when the model is rotated 360° around the x, y and z axes respectively. At the same time, calculate the seven moment vectors of the shadow and generate a database.

[0010] Step 4, Reef attitude estimation: Calculate the distance d between moment vectors: Where I w I represents the moment vector of the target shadow in a side-scan sonar image. sThe moment vector represents the target shadow in the model library. The model attitude corresponding to the minimum distance is the target spatial attitude in the side-scan sonar image. Combined with the evaluated target height information, the depth of the artificial reef buried on the seabed is obtained. The output image shows the height, tilt angle and burial depth of the artificial reef detected in the image.

[0011] The above-mentioned acoustic evaluation method for the attitude of artificial reefs, specifically the image filtering process in step 2, involves filtering the image using a two-dimensional empirical mode decomposition method, resulting in the following image representation:

[0012]

[0013] Where u(x,y) represents the sonar image, k0 represents the weighting coefficient of the margin, and k i The weighting coefficients of the i-th intrinsic mode function are represented by r(x,y), where r(x,y) is the margin, n is the iteration number, and IMF is the weighting coefficient. i Let be the i-th intrinsic mode function.

[0014] The above-mentioned method for acoustic evaluation of artificial reef attitude, specifically step 2, involves coarse target extraction as follows: The filtered image is processed using the C-means clustering algorithm to achieve coarse target extraction. A termination threshold and a maximum number of iterations are set. The Euclidean distance from all pixel values ​​to all initial centers is calculated and saved. Cluster centers and membership functions are calculated, and the membership matrix is ​​updated. A clustering loss function based on the membership function is calculated. The calculation is iterated until the increment of the clustering loss function is less than the termination threshold or the number of iterations reaches the maximum number of iterations, at which point the calculation is terminated, resulting in the coarse target extraction result.

[0015] The above-mentioned method for acoustic evaluation of artificial reef attitude, wherein the cluster center v t The calculation formula is:

[0016]

[0017] Where ii and jj are the pixel indices in the horizontal and vertical directions of the image, respectively, and u(x ii y jj ) represents the image after denoising using two-dimensional empirical mode decomposition, a t (u(x ii y jj )) m It is the pixel value u(x) ii y jj ) belongs to a t The membership degree of the class, M and N represent the number of pixels in the horizontal and vertical directions of the image, and m represents an additional parameter, which is usually 2.

[0018] The above-mentioned method for acoustic evaluation of the attitude of artificial reefs, wherein the membership function a t The calculation formula is:

[0019]

[0020] Among them, v t The cluster centers are represented by m, which is an additional parameter (usually 2), and c represents the number of subsets. ii y jj ) represents the image after noise reduction by two-dimensional empirical mode decomposition.

[0021] The above-mentioned method for acoustic evaluation of artificial reef attitude, wherein the clustering loss function based on the membership function is calculated using the following formula:

[0022]

[0023] Among them, a t V represents the membership function. t The cluster centers are represented by M and N, which represent the number of pixels in the horizontal and vertical directions of the image, respectively. m is an additional parameter, typically set to 2. c represents the number of subsets. u(x) ii y jj ) represents the image after noise reduction by two-dimensional empirical mode decomposition.

[0024] In the aforementioned method for acoustic evaluation of artificial reef attitude, the specific process of fine target extraction in step 2 is as follows: The mean value v of the target area from the coarse target extraction result is... tg Mean value of shaded area v sw , background area mean v bk The target contour φ1 and shadow contour φ2 are used as initialization parameters for the level set algorithm, respectively, and the mean value v of the background area is... bk Target area mean v tg Mean value of shaded area v sw The iterative calculation is performed using the following formula:

[0025]

[0026]

[0027]

[0028] Wherein, H(Φ) i ) is the Heaviside function;

[0029] During the iteration process, the iteration increments φ1 and φ2 are calculated separately. The calculation terminates when the maximum number of iterations is reached, and the target fine-grained structure is output. The specific formula for calculating the iteration increment is as follows:

[0030]

[0031]

[0032] Where v represents a constant coefficient, v≥0; IMF i Let λ be the i-th intrinsic mode function; NN is the total number of images, where N represents the number of pixels in the vertical direction of the image, and n represents the number of iterations; n λ represents the optimization margin weight. n =χ n ·α n α n Represents the residual weight; η n Indicates optimizing IMF weights, η n =χ n ·β n ,β n Indicates IMF weights; χ n Represents image weights; r n (x, y) represents the BEMD residual image, δ represents the Dirac function; μ represents the penalty term parameter, μ > 0; p represents the cluster center; Δφ2 represents the iteration increment of the shadow contour, and Δφ1 represents the iteration increment of the target contour.

[0033] In the aforementioned method for acoustic evaluation of the attitude of an artificial reef, the target height H in step 3... t The calculation formula is:

[0034]

[0035] Among them, L s H is the length of the shadow. s For the height of the towed fish, L t R is the length of the target object. s Slope distance;

[0036] The seven moment vectors corresponding to the shadow of the target are as follows:

[0037] I1 = h 20 +h 02

[0038]

[0039] I3=(h 30 -3h 12 ) 2 +(3h 21 -h 03 ) 2

[0040] I4=(h 30 +h 12 ) 2 +(h 21 +h 03 ) 2

[0041] I5=(h 30 -3h 12 (h) 30 +h 12 )[(h 30 +h 12 ) 2 -3(h 21 +h 03 ) 2 ]+(3h 21 -h 03 (h) 21 +h 03 )[3(h 30 +h 12 ) 2 -(h 21 +h 03 ) 2 ]

[0042] I6=(h 20 -h 02 )[(h 30 +h 12 ) 2 -(h 21 +h 03 ) 2 +4h 11 (h 30 +h 12 (h) 21 +h 03 )

[0043] I7=(3h 21 -h 03 (h) 30 +h 12 )[(h 30 +h 12 ) 2 -3(h 21 +h 03 ) 2 ]-(h 30 -3h 12 (h) 21 +h 03 )[3(h 30 +h 12 ) 2 -(h 21 +h 03 ) 2 ]

[0044] Wherein, the normalized central distance h pq : Center of the square (x0, y0): (p+q) order invariant moment m pq: p, q = 0, 1, 2, ..., M, N represent the number of pixels in the horizontal and vertical directions of the image, and u(x, y) represents the sonar image.

[0045] The beneficial effects of the present invention are: (1) by utilizing the fast operation speed of the fuzzy C-means clustering method and the high segmentation accuracy of the horizontal clustering method, the target in the side-scan sonar image is quickly and accurately segmented; (2) by matching the target shadow contour of the side-scan sonar image with the projection contour of the constructed three-dimensional artificial reef model, the attitude of the artificial reef is evaluated. Attached Figure Description

[0046] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0047] Figure 1 This is the acoustic evaluation process for the attitude of artificial reefs in this embodiment;

[0048] Figure 2 This is a schematic diagram illustrating the principle of calculating the height of the artificial reef using side-scan sonar in this embodiment;

[0049] Figure 3 This is the result of the artificial reef model constructed in this embodiment, where a) is the artificial reef model, and b) is an example of the shadow outline (39°, 46°, 20°).

[0050] Figure 4 This is the original side-scan sonar image in this embodiment;

[0051] Figure 5 This is the result of target and shadow extraction from the side-scan sonar image in this embodiment;

[0052] Figure 6 This is the artificial reef attitude evaluation result of this embodiment, where a) is the reef edge height estimation, b) is the simulation shadow contour with the highest matching degree, and c) is the estimated reef attitude (the tilt angles in the x, y, and z directions are 81°, 1°, and 81°, respectively). Detailed Implementation

[0053] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] This embodiment discloses an acoustic evaluation method for the attitude of artificial reefs, the basic principle of which is:

[0055] Sound waves propagating to the seabed generate backscattered echoes. Side-scan sonar generates sonar images of varying brightness based on the intensity of these backscattered echoes, with targets appearing brighter and shadowed areas darker. The shadow contours of targets differ significantly under different attitudes. The translation, rotation, and scale invariance of shadows can be used to match target attitudes. The target height can be calculated from the geometric relationship between the towed fish height, target position, and shadow position. Combining the target's attitude and height information allows for the estimation of the reef's burial depth. The proposed acoustic assessment method for reef attitude includes three main steps: ① Target extraction: Denoising the side-scan sonar images and extracting the target and shadow contours; ② Feature parameter calculation: Calculating the target height and shadow contour features based on geometric relationships; ③ Reef attitude estimation: Matching the reef tilt angle using shadow features and calculating the subsidence based on reef height information. The specific steps are as follows: Figure 1 As shown.

[0056] (1) Target extraction: The targets extracted from the side-scan sonar images are mainly the outlines of the targets and their shadows. The image processing flow includes image filtering, coarse target extraction, fine target extraction, and target shadow matching.

[0057] ① Side-scan sonar signals are characterized by low-frequency signals in the target and shadow areas, and high-frequency noise signals in the background area. Lower weights are assigned to low-frequency signals to retain useful image information, while lower weights are assigned to high-frequency signals to suppress noise and interference. BEMD adaptively decomposes side-scan sonar image signals into high-frequency components (multiple Intrinsic Mode Function (IMF) components) and low-frequency components (residuals) at different scales. Different IMFs represent detailed information at various scales of the original image, and a single residual component represents the overall contour information of the image.

[0058] Assume that the sonar image u(x,y) data plane, after taking its first or multiple derivatives, contains at least one maximum and one minimum point. The image filtered using the BEMD method can be represented as:

[0059]

[0060] In the formula, k represents the weighting coefficient, r(x,y) is the residual, n is the number of iterations, and IMF is the weighting factor. i Let be the i-th intrinsic mode function.

[0061] ② The Fuzzy C-means Clustering (FCM) algorithm obtains the membership degree of each sample point to all class centers by optimizing the objective function, and determines the class of the sample point based on the membership function. The FCM algorithm is easy to implement and runs quickly, but its segmentation accuracy is not high. In this invention, the FCM algorithm is used to process side-scan sonar images to achieve coarse segmentation.

[0062] Suppose the fuzzy set of the image is A(x,y), divide the fuzzy set into c subsets, and the cluster center is v. t , 1 << t << c. The membership function can be expressed as a t (x ii ,y jj ), representing a pixel (x) ii ,y jj The value belonging to the cluster center is v. t The degree of, where ii and jj are the pixel numbers in the horizontal and vertical directions of the image, respectively.

[0063] The clustering loss function based on membership function can be written as:

[0064]

[0065] Where M and N are the number of pixels in the horizontal and vertical directions of the image, and m is an additional parameter, usually taken as 2.

[0066] Let J be a t The partial derivatives of and u are 0. The necessary condition for obtaining the minimum value in Equation 2 is:

[0067]

[0068]

[0069] The cluster centers and membership functions are solved iteratively until the convergence condition is met, and the coarse image segmentation result can be obtained.

[0070] ③ Level set models typically construct functions based on constraints such as contour curvature and target region area, which are effective for images where the average pixel values ​​of the target and background differ significantly. However, the continuous initialization during the calculation process leads to high computational cost and long processing time. In this invention, the FCM coarse segmentation result is used as the initialization of the level set function, which ensures segmentation accuracy while reducing the computational cost of data processing.

[0071] The mean of the four regions after coarse segmentation of the sonar image is {v1, v2, v3, v4}. The four level set functions of BEMD denoising can be expressed as:

[0072]

[0073] in Threshold for the original image region H(Φ) represents the threshold of the image region after BEMD processing, χ is the image weight, μ1≥0 and μ2≥0 are fixed parameters, and H(Φ) is the image weight. i Let C1 and C2 be the contours of curves 1 and 2, respectively. Introducing the Heaviside function, the energy function can be written as:

[0074]

[0075] Where λ n =χ n ·α n η n =χ n ·β n , NN represents the total number of images. The Lagrange increment formulas for curves 1 and 2 are:

[0076]

[0077]

[0078] Where μ>0, v≥0, λ, η, and χ are parameters, φ1 and φ2 are two evolution curves, and H and δ are normalized values. The coarse segmentation result is used as the initial set of the level set, and the two contour curves are iterated according to Equations 7 and 8 respectively until the convergence condition is met, thus obtaining the fine segmentation result of the image.

[0079] ④ Remove smaller objects from the target and shadow parts of the fine segmentation results. Considering the characteristics of side-scan sonar imaging, for each shadow object in the image, match a target object that is closest to the target characteristics within a specific range near the sonar direction, and store the position information of the target and shadow in pairs.

[0080] (2) Calculation of feature parameters: The main parameters of the target are the height of the target and the contour features of the shadow.

[0081] ① Side-scan sonar images: The height of a target can be calculated using parameters such as the length of its shadow and the height of the side-scan sonar. The principle of artificial reef height measurement is as follows: Figure 2 As shown, the target height can be obtained according to the principle of similar triangles:

[0082]

[0083] Where Ls is the shadow length, Hs is the towed fish height, Lt is the target object length, and Rs is the slant distance.

[0084] The Hu moment is used to characterize shadow contour features. The Hu moment constructs seven invariant moments using second- and third-order center moments, which remain invariant to translation, scaling, and rotation under continuous image conditions. The seven invariant moments are defined as follows:

[0085] I1 = h 20 +h 02

[0086]

[0087] I3=(h 30 -3h 12 )2 +(3h 21 -h 03 ) 2

[0088] I4=(h 30 +h 12 ) 2 +(h 21 +h 03 ) 2

[0089] I5=(h 30 -3h 12 )(h 30 +h 12 )[(h 30 +h 12 ) 2 -3(h 21 +h 03 ) 2 ]

[0090] +(3h 21 -h 03 )(h 21 +h 03 )[3(h 30 +h 12 ) 2 -(h 21 +h 03 ) 2 ]

[0091] I6=(h 20 -h 02 )[(h 30 +h 12 ) 2 -(h 21 +h 03 ) 2 ]+4h 11 (h 30 +h 12 )(h 21 +h 03 )

[0092] I7=(3h 21 -h 03 )(h 30 +h 12 )[( 30 +h 12 ) 2 -3(h 21 +h 03 ) 2 ]

[0093] -(h 30 -3h12 (h) 21 +h 03 )[3(h 30 +h 12 ) 2 -(h 21 +h 03 ) 2 ]

[0094] in

[0095] p, q = 0, 1, 2, ...

[0096] ② Model: Extract the towed fish height, target position, and artificial reef shape from the image processing results to construct a 3D artificial reef model. Calculate the artificial reef height and shadow contour when the model is rotated 360° around the x, y, and z axes (360×360×360 cases in 1° intervals), and simultaneously calculate the seven moments of the shadow. Examples of the artificial reef model and shadow contour are shown below. Figure 3 As shown, where Figure 3 a) is an artificial reef model, and b) is an example of the shaded outline (39°, 46°, 20°).

[0097] (3) Artificial reef attitude estimation: Calculate the distance between the moment vector of the side-scan sonar image and the moment vector in the model library. The model attitude corresponding to the smallest distance is the target spatial attitude in the side-scan sonar image. Combine the target height information of the evaluation to obtain the depth of the artificial reef buried on the seabed.

[0098] This embodiment provides a method for acoustic evaluation of the attitude of artificial reefs in shallow waters, the specific steps of which are as follows:

[0099] 1. Data Acquisition: The side-scan sonar is fixed to the side of the ship to collect seabed acoustic data in the area to be measured. The artificial reef deployed in this area is known to be 2×2×2 in size. The original side-scan sonar image is shown below. Figure 4 As shown.

[0100] 2. Target Extraction:

[0101] (1) Select the side-scan sonar image u(x,y) of the area to be tested, and convert the color image to a grayscale image. Extract the IMF, obtain the extreme value envelope surface by finding local extreme points, update the residual, and then extract the remaining IMFs in sequence. Finally, decompose the sonar image into 4 IMF images and 1 residual image, with weighting coefficients of 0.2, 0.6, 1.2, 1.8, and 1, respectively. After weighting by IMF and residual, the reconstructed image is the filtered result.

[0102] (2) Import the filtered image and calculate the background area. Target area v tk={x max}、Shadow area v sw ={x min Five initial cluster centers were reclassified into three classes after iteration.

[0103] Set the termination threshold ε to 0.01 and the maximum number of iterations to 2000. Initialize the membership matrix, calculate and save the Euclidean distance from all pixel values ​​in the image to the 5 initial centers, and then proceed according to... Calculate cluster centers based on Calculate the membership function and update the membership matrix according to Calculate the clustering loss function based on the membership function. Iterate until the increment of the clustering loss function is less than 0.01 or the number of iterations reaches 2000, then terminate the calculation and output the clustering results.

[0104] (3) The mean value of the target region of the coarse image segmentation result v tg Mean value of shaded area v sw , background area mean v bk The target contour φ1 and the shadow contour φ2 are used as initialization parameters for the level set algorithm. The mean values ​​of the three regions are calculated iteratively according to the formula. During the iteration process, the iteration increments of φ1 and φ2 are calculated respectively.

[0105]

[0106] The calculation terminates after 100 iterations, and the clustering results are output.

[0107] (4) Remove objects with an area smaller than 30 from the target and shadow portions of the fine segmentation results. Considering the characteristics of side-scan sonar imaging, extract each shadow object sequentially. Find the largest target object within a range of 1.5 times its length and 1.3 times its width in the direction closest to the sonar, and store the position information of the target and shadow in pairs. The target and shadow extraction results of the side-scan sonar image are as follows: Figure 5 As shown.

[0108] 3. Calculation of characteristic parameters:

[0109] (1) Side scan sonar image parameters: Calculate the maximum height of each target based on the target and shadow position coordinates. At the same time, calculate the seven moments corresponding to the shadow of the target:

[0110] I1 = h 20 +h 02

[0111]

[0112] I3=(h 30 -3h12 ) 2 +(3h 21 -h 03 ) 2

[0113] I4=(h 30 +h 12 ) 2 +(h 21 +h 03 ) 2

[0114] I5=(h 30 -3h 12 )(h 30 +h 12 )[(h 30 +h 12 ) 2 -3(h 21 +h 03 ) 2 ]+(3h 21 -

[0115] h 03 )(h 21 +h 03 )[3(h 30 +h 12 ) 2 -(h 21 +h 03 ) 2 ]

[0116] I6=(h 20 -h 02 )[(h 30 +h 12 ) 2 -(h 21 +h 03 ) 2 ]+4h 11 (h 30 +h 12 )(h 21 +h 03 )

[0117] I7=(3h 21 -h 03 )(h 30 +h 12 )[(h 30 +h 12 ) 2 -3(h 21 +h 03 ) 2 ]-(h 30 -

[0118] 3h 12 (h) 21 +h 03 )[3(h 30 +h 12 ) 2 -(h 21 +h 03 ) 2 ]

[0119] The seven moments of the target shadow in the area to be measured are 0.27, 0.04, 0.00018, and 7.12 × 10⁻⁶. -5 -4.2×10 -9 1.4×10 -5 -6.3×10 -10 .

[0120] (2) Extract the height of the towed fish, the target position, and the shape of the artificial reef from the image processing results to construct a three-dimensional artificial reef model. The model is rotated 360° around the x, y, and z axes respectively (1° interval, for a total of 360×360×360 cases), and the seven moments of the shadow under different poses are calculated to generate a database.

[0121] 4. Artificial reef attitude estimation:

[0122] Calculate the target shadow moment vector I in the side-scan sonar image w With the target shadow moment vector I in the model library s Distance The model attitude corresponding to the minimum distance is the target spatial attitude in the side-scan sonar image. Combined with the assessed target height information, the depth of the artificial reef's burial on the seabed can be determined. The artificial reef attitude assessment results are as follows: Figure 6 As shown, from Figure 6 From c), it can be seen that the minimum distance corresponds to the rotation angles (81°, 1°, 81°). Figure 6 From a) and b), we can see that the maximum height of the attitude model is 2.333 meters, the maximum height of the target in the side-scan sonar image is 2.252 meters, and the estimated settlement height is 0.081 meters.

[0123] The above embodiments are merely exemplary embodiments of the present invention and are not intended to limit the present invention. The scope of protection of the present invention is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to the present invention within its spirit and scope of protection, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of the present invention.

Claims

1. A method for assessing the posture of an artificial reef by using acoustic signals, characterized in that: It comprises the following steps: Step 1, data acquisition: using side scan sonar to collect seabed acoustic data; Step 2, target extraction: cutting the image of the region to be measured from the data obtained in step 1, performing image processing including image filtering, target rough extraction, target fine extraction and target shadow matching on the cut image, and storing the position information of the target and the target shadow; Step 3, characteristic parameter calculation: calculating the maximum height of each target and the seven moment vectors of the target corresponding shadow according to the target and shadow position coordinates, extracting the height of the image processing result, the target position and the shape of the artificial fish reef, constructing a three-dimensional fish reef model, calculating the height of the artificial fish reef and the shadow contour when the model rotates 360° around the x, y and z axes respectively, and calculating the seven moment vectors of the shadow, and producing a database; Step 4, fish reef pose estimation: calculate the distance between the moment vectors d : wherein represents the moment vector of the target shadow in the side-scan sonar image, represents the moment vector of the target shadow in the database, and the distance minimum corresponds to the model pose, i.e. the spatial pose of the target in the side-scan sonar image, combined with the evaluated target height information to obtain the burying depth of the fish reef on the seabed, and output the height, tilt angle and burying depth of the artificial fish reef detected in the image; The height of the target in step 3 The formula for calculating the height of the target in step 3 is: wherein, L s is the shadow length, H s is the height of the trailing fish, L t is the target length, R s is the slant range; The seven moment vectors of the target corresponding shadow are respectively: wherein the normalized center distance h pq : , the centroid ( x 0, y 0): , , the ( p + q ) order invariant moment m pq : , p, q = 0, 1, 2, …, M , N represent the number of pixels in the horizontal and vertical directions of the image, u ( x , y ) represents the sonar image.​ 2. The method according to claim 1, wherein, The image filtering process in step 2 is specifically: after filtering the image using the two-dimensional empirical mode decomposition method, the image is represented as: wherein, u x , y represents a sonar image, k 0 represents a weighting coefficient for the residual, k i represents a weighting coefficient for the i th eigenmode function ，r x , y is a residual, n is an iteration number, IMF i is the i th eigenmode function.​​ 3. The method according to claim 1, wherein, The target rough extraction in step 2 is specifically: the filtered image is processed using the fuzzy C-means clustering algorithm to achieve the purpose of target rough extraction, the termination threshold and the maximum iteration number are set, the Euclidean distance of all pixel values of the image to all initial centers is calculated and saved, the clustering center and the membership function are calculated, and the membership matrix is updated, the clustering loss function based on the membership function is calculated, and the iteration calculation is terminated when the increment of the clustering loss function is less than the termination threshold or the iteration number reaches the maximum iteration number, and the target rough extraction result is obtained.

4. The method according to claim 3, wherein, The cluster centers The calculation formula is: wherein, ii , jj are the image horizontal and vertical pixel number indexes respectively, denotes the image after denoising by two-dimensional empirical mode decomposition, is the pixel value is the membership degree of a t class, M , N denote the image horizontal and vertical pixel number, m denotes the additional parameter.

5. The method according to claim 3, wherein, The membership function a t The calculation formula is: wherein, denotes a cluster center, m denotes an additional parameter, denotes a number of subsets, denotes a denoised image after two-dimensional empirical mode decomposition.

6. The method according to claim 3, wherein, The calculation formula of the clustering loss function based on the membership function is: wherein a t denotes a membership function, denotes a cluster center, M , N denotes the number of pixels in horizontal and vertical directions of the image, m denotes an additional parameter, denotes the number of subsets, denotes an image after denoising by two-dimensional empirical mode decomposition.

7. The method according to claim 1, wherein, The specific process of the target extraction in step 2 is: calculating the target region mean value of the target rough extraction result v tg , the shadow region mean value v sw , the background region mean value v bk , the target contour , the shadow contour As the initialization parameters of the level set algorithm, the background region mean value v bk , the target region mean value v tg , the shadow region mean value v sw is iteratively calculated, and the calculation formula is as follows: wherein is the Heaviside function; The iterative increments of the following formulas are calculated respectively in the iteration process , The calculation is terminated when the iteration number reaches the maximum iteration number, and the target fine extraction structure is output. The specific calculation formula of the iterative increment is as follows: wherein, represents a constant coefficient, IMF i is the first i intrinsic mode function; NN is the total number of images; N represents the number of pixels in the vertical direction of the image; n represents the number of iterations; represents the optimization residual weight, , represents the residual weight; represents the optimization IMF weight, , represents the IMF weight; represents the image weight; represents the two-dimensional empirical mode decomposition residual image, represents the Dirac function; represents the penalty term parameter, ; represents the cluster center; represents the iterative increment of the shadow contour, represents the iterative increment of the target contour.​​

Images