Quick judgment method for side-scan sonar image view field target doubt

Abstract

The invention discloses a side-scan sonar image view field target doubt rapid discrimination method, and relates to the technical field of image discrimination, and the method comprises the steps: S1, obtaining side-scan sonar image data, and generating an original image matrix; s2, based on the original image matrix, executing two-dimensional discrete cosine transform operation to generate a transformation coefficient matrix; s3, calculating logarithmic transformation of coefficient absolute values based on the transformation coefficient matrix, and generating an intensity characteristic matrix; s4, extracting a plurality of region feature matrixes according to a preset segmentation rule based on dimension information of the intensity feature matrix; s5, on the basis of the multiple region feature matrixes, feature statistical values of all regions are calculated respectively, and a region feature value sequence is generated; and S6, on the basis of the regional feature value sequence, calculating a ratio relationship between feature values, and generating a target existence judgment result according to a ratio judgment rule. According to the specific value of the characteristic values of different areas, whether the target object exists in the side-scan sonar image or not is quickly and effectively judged.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image discrimination, and particularly relates to a quick discrimination method for side scan sonar image field of view target suspicion. BACKGROUND

[0002] As an important underwater acoustic detection device, the side scan sonar plays an important role in underwater geological and topographic survey, target searching, obstacle detection and marine three-dimensional model inversion.

[0003] However, the existing side scan sonar image processing technology has obvious defects. The current image segmentation method mainly processes part of the region by intercepting from the whole image, which leads to extremely low image resolution and seriously affects the subsequent image segmentation and target recognition effect. The local image directly intercepted from the side scan sonar image generally has the problems of too concentrated pixel range and poor image resolution. Even after gain correction processing, the image quality improvement effect is still not ideal, which is difficult to meet the demand of fast and accurate target recognition.

[0004] Therefore, there is an urgent need for a quick discrimination method for side scan sonar image field of view target suspicion. SUMMARY

[0005] The present application provides a quick discrimination method for side scan sonar image field of view target suspicion to solve the above problems existing in the prior art.

[0006] In order to achieve the above purpose, the present application provides the following technical scheme: A quick discrimination method for side scan sonar image field of view target suspicion, comprising: S1: acquiring side scan sonar image data to generate an original image matrix; S2: performing two-dimensional discrete cosine transform operation based on the original image matrix to generate a transform coefficient matrix; S3: calculating the logarithmic transform of the absolute value of the coefficient based on the transform coefficient matrix to generate an intensity feature matrix; S4: extracting a plurality of regional feature matrices according to a preset segmentation rule based on the dimension information of the intensity feature matrix; S5: calculating the feature statistical value of each region based on the plurality of regional feature matrices to generate a regional feature value sequence; S6: calculating the ratio relationship between the feature values based on the regional feature value sequence, and generating a target existence judgment result according to the ratio judgment rule.

[0007] Among them, the step S2 comprises: S21: receiving pixel data of the original image matrix; S22: performing one-dimensional discrete cosine transform in the row direction on the pixel data to generate a row transform intermediate matrix; S23: Perform a one-dimensional discrete cosine transform in the column direction based on the row transform intermediate matrix to generate a transform coefficient matrix.

[0008] The S3 step includes: S31: Extract the coefficient values at each position in the transform coefficient matrix; S32: Perform an absolute value operation on each coefficient value to generate an absolute value matrix; S33: Perform a logarithmic operation on each element in the absolute value matrix to generate an intensity feature matrix.

[0009] The S4 step includes: S41: Obtain the number of rows and the number of columns of the intensity feature matrix; S42: Select the smaller value between the number of rows and the number of columns as a segmentation reference value; S43: Determine a region size parameter according to a preset ratio based on the segmentation reference value; S44: Extract region feature matrices of the same size from the top-left corner, the middle, and the bottom-right corner of the intensity feature matrix based on the region size parameter.

[0010] The S5 step includes: S51: Receive a plurality of region feature matrices; S52: Perform a summation operation on the elements within each region feature matrix to generate a region sum value; S53: Divide each region sum value by the number of elements in the corresponding region to generate a sequence of region feature values.

[0011] The S6 step includes: S61: Extract a first feature value f1, a second feature value f2, and a third feature value f3 from the sequence of region feature values; S62: Calculate the ratio R1 of the first feature value to the second feature value, where R1 = f1 / f2; S63: Calculate the ratio R2 of the first feature value to the third feature value, where R2 = f1 / f3; S64: Perform a decision operation based on the ratios R1 and R2 to generate a target existence decision result.

[0012] The decision operation of the S64 step includes: Operation condition one: when R1 is greater than or equal to a first threshold T1, the decision result is that a target object exists; Operation condition two: when R1 is greater than or equal to a second threshold T2 and less than the first threshold T1, and R2 is greater than or equal to a third threshold T3, the decision result is that a target object exists; When neither operation condition one nor operation condition two is satisfied, the decision result is that a target object does not exist. Among them, T1>T2>1, T3>1.

[0013] Step S44 includes: S441: Take the top left corner of the intensity feature matrix as the starting position of the first region; S442: Based on the number of rows M and columns N of the intensity feature matrix and the region size parameter d, the starting position of the central region is calculated as ((Md) / 2, (Nd) / 2); S443: Calculate the starting position of the lower right corner region so that the lower right boundary of the region is aligned with the lower right boundary of the intensity feature matrix.

[0014] Step S43 includes: S431: Receive segmentation reference value m; S432: Divide the segmentation reference value by the preset segmentation coefficient K to generate the region side length value d, where d=m / K, and the preset segmentation coefficient K is 3; S433: Use the region side length value d as the region size parameter for subsequent region extraction.

[0015] Compared with the prior art, the present invention has the following advantages: This invention proposes an eigenvalue ratio discrimination algorithm based on two-dimensional cosine transform. It uses the coefficient matrix of the two-dimensional cosine transform of side-scan sonar as the feature matrix, selects the average coefficient of a specific region as the eigenvalue, and quickly and effectively determines whether there is a target object in the side-scan sonar image based on the ratio of the eigenvalues ​​of different regions.

[0016] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention.

[0017] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a method for rapid identification of suspicious targets in the field of view of a side-scan sonar image, according to an embodiment of the present invention. Figure 2 These are side-scan sonar images of the sunken ship in an embodiment of the present invention; Figure 3 The F1, F2, and F3 feature matrices represent the regions in this embodiment of the invention; Figure 4This is a side-scan sonar image without a target in an embodiment of the present invention; Figure 5 This is the coefficient intensity image 1 in the embodiment of the present invention; Figure 6 These are side-scan sonar images of the sunken ship in an embodiment of the present invention; Figure 7 Image 2 showing the coefficient intensity in this embodiment of the invention; Figure 8 These are side-scan sonar images of the sunken ship in an embodiment of the present invention; Figure 9 Image 3 is the coefficient intensity image in this embodiment of the invention; Figure 10 This is a side-scan sonar image without a target in an embodiment of the present invention; Figure 11 Image 4 shows the coefficient intensity in this embodiment of the invention. Detailed Implementation

[0019] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0020] The embodiments of the present invention provide, as follows Figure 1 As shown, a rapid method for identifying suspicious targets in the field of view of a side-scan sonar image includes: S1: Acquire side-scan sonar image data and generate the original image matrix; S2: Based on the original image matrix, perform a two-dimensional discrete cosine transform operation to generate a transform coefficient matrix; S3: Based on the transformation coefficient matrix, calculate the logarithmic transformation of the absolute values ​​of the coefficients to generate the intensity feature matrix; S4: Based on the dimensional information of the intensity feature matrix, extract multiple region feature matrices according to preset segmentation rules; S5: Based on multiple regional feature matrices, calculate the feature statistics of each region and generate a sequence of regional feature values; S6: Based on the regional feature value sequence, calculate the ratio relationship between feature values, and generate the target existence determination result according to the ratio determination rule.

[0021] The working principle and beneficial effects of the above technical solution are as follows: In step S1, the system acquires the acoustic echo signal of the underwater target area through the side-scan sonar device, converts the received analog acoustic signal into digital pixel data through the analog-to-digital converter, and arranges these digital data into an M×N dimensional original image matrix according to the two-dimensional coordinate relationship of the scan line sequence and the range gate, where M represents the number of scan lines and N represents the number of sampling points on each scan line.

[0022] In step S2, a two-dimensional discrete cosine transform operation is performed on the original image matrix. First, a one-dimensional DCT calculation is carried out for each row of the matrix. By using cosine basis functions to decompose the spatial domain signal into the frequency domain, the pixel intensity information in the time domain is converted into coefficient representation in the frequency domain. Subsequently, this process is repeated in the column direction. Finally, a transform coefficient matrix Co with the same size as the original image is obtained, and this matrix reflects the energy distribution characteristics of the image at different frequency components.

[0023] In step S3, for each complex coefficient at each position in the transform coefficient matrix, first calculate its absolute value to obtain the amplitude information and eliminate the influence of the phase on subsequent analysis. Then, perform a natural logarithm transform on each absolute value, and generate an intensity feature matrix I using the formula I = log(abs(Co)). This logarithmic transform can compress the dynamic range and highlight the differences in low-frequency components.

[0024] In step S4, obtain the number of rows M and the number of columns N of the intensity feature matrix I, select min(M, N) as the segmentation reference value m, divide this reference value by the preset segmentation coefficient K = 3 to get the region side length d = m / 3. On the intensity feature matrix, extract sub-matrices with a size of d×d from the upper-left region (starting position (0, 0)), the middle region (starting position ((M - d) / 2, (N - d) / 2)), and the lower-right region (aligning its lower-right boundary with the matrix boundary) as the feature matrices F1, F2, F3, that is: take matrices with sizes of m / 3 in the upper-left, m / 3 in the middle, and m / 3 in the lower-right as the feature matrices F1, F2, and F3 respectively.

[0025] In step S5, perform an accumulation summation operation on all elements within each region feature matrix, divide the obtained region sum value by the total number d² of elements within the region, calculate the average coefficient values f1, f2, f3 of each region, and form a region eigenvalue sequence. These eigenvalues reflect the degree of frequency domain energy concentration of the image at different spatial positions. [[ID=ll]]

[0026] In step S6, calculate the ratio R1 = f1 / f2 of the first eigenvalue to the second eigenvalue, and the ratio R₂ = f1 / f3 of the first eigenvalue to the third eigenvalue. Conduct a target existence assessment according to the preset determination rules: directly determine that there is a target object when R1 ≥ T1 (the first threshold); also determine that there is a target object when T₂ ≤ R1 < T1 and R₂ ≥ T3; otherwise, determine that there is no target object, where the threshold relationship is T1 > T2 > 1, T3 > 1.

[0027] The image target identification is performed using side-scan sonar. A two-dimensional discrete cosine transform is applied to the image to obtain the image discrete cosine transform coefficient matrix Co. The coefficient intensity matrix I = log (abs(Co)) is taken, and the smaller value m is taken from the number of rows and columns of matrix I. The matrices of size m / 3 in the upper left, middle, and lower right are taken as feature matrices F1, F2, and F3, respectively. The average values ​​f1, f2, and f3 of feature matrices F1, F2, and F3 are taken as eigenvalues. like If so, then it is determined that a target object exists in the image; if If the target object is found, the image is determined to contain the target object; otherwise, the image is determined to contain the target object.

[0028] For example Figure 2 The side-scan sonar image of the shipwreck shown is subjected to a two-dimensional discrete cosine transform to obtain a discrete cosine transform coefficient matrix Co with the same size as the image; the coefficient intensity matrix I = log (abs(Co)) is then taken, and its intensity image is shown below. Figure 3 As shown, the size is 205×189, so the smaller value of the number of rows and columns in the matrix, 189, is taken as the segmentation m value; the matrices of size 63×63 in the upper left, middle, and lower right are taken as feature matrices F1, F2, and F3 respectively; the average values ​​of feature matrices F1, F2, and F3, f1=33.15, f2=8.30, and f3=2.04, are taken as eigenvalues; according to the comparison rules, 2 ≤ f1 / f2 = 3.99 ≤ 5, f1 / f3 = 16.5 ≥ 10, it is determined that there is a target object in the image.

[0029] For example Figure 4 The side-scan sonar image without a target shown is subjected to a two-dimensional discrete cosine transform to obtain a discrete cosine transform coefficient matrix Co with the same size as the image. Take the coefficient intensity matrix I = log (abs(Co)), and its intensity image is as follows. Figure 5 As shown, the size is 416×456, so the smaller value of 416 among the number of rows and columns of the matrix is ​​taken as the segmentation value m; The matrices of size 138×138 in the upper left, middle, and lower right parts are respectively taken as characteristic matrices F1, F2, and F3; The average values ​​of the feature matrices F1, F2, and F3, f1=22.40, f2=11.19, and f3=4.6, are taken as the eigenvalues. According to the comparison rule, 2 ≤ f1 / f2 = 2.00 ≤ 5, but f1 / f3 = 4.87 ≤ 10, so it is determined that there is no target object in the image.

[0030] For example Figure 6The side-scan sonar image of the shipwreck shown is subjected to a two-dimensional discrete cosine transform to obtain a discrete cosine transform coefficient matrix Co with the same size as the image; the coefficient intensity matrix I = log (abs(Co)) is then taken, and its intensity image is shown below. Figure 7 As shown, the size is 376×330, so the smaller value of the number of rows and columns in the matrix, 330, is taken as the segmentation m value; the matrices of size 110×110 in the upper left, middle and lower right are taken as feature matrices F1, F2 and F3 respectively; the average values ​​of feature matrices F1, F2 and F3, f1=47.14, f2=9.40 and f3=3.56, are taken as eigenvalues; according to the comparison rule, f1 / f2 = 5.01≥ 5, it is determined that there is a target object in the image.

[0031] For example Figure 8 The side-scan sonar image of the shipwreck shown is subjected to a two-dimensional discrete cosine transform to obtain a discrete cosine transform coefficient matrix Co with the same size as the image; the coefficient intensity matrix I = log (abs(Co)) is then taken, and its intensity image is shown below. Figure 9 As shown, the size is 342×338, so the smaller value of the number of rows and columns in the matrix, 338, is taken as the segmentation m value; the matrices of size 332×332 in the upper left, middle and lower right are taken as feature matrices F1, F2 and F3 respectively; the average values ​​of feature matrices F1, F2 and F3, f1=53.85, f2=8.87 and f3=2.66, are taken as eigenvalues; according to the comparison rule, f1 / f2 = 6.07≥ 5, it is determined that there is a target object in the image.

[0032] For example Figure 10 The side-scan sonar image of the shipwreck shown is subjected to a two-dimensional discrete cosine transform to obtain a discrete cosine transform coefficient matrix Co with the same size as the image; the coefficient intensity matrix I = log (abs(Co)) is then taken, and its intensity image is shown below. Figure 11 As shown, the size is 414×377, so the smaller value of the number of rows and columns in the matrix, 377, is taken as the segmentation m value; the matrices of size 125×125 in the upper left, middle and lower right are taken as feature matrices F1, F2 and F3 respectively; the average values ​​of feature matrices F1, F2 and F3, f1=23.36, f2=11.46 and f3=4.50, are taken as eigenvalues; according to the comparison rules, 2 ≤ f1 / f2 = 2.04 ≤ 5, f1 / f3 = 5.19 ≤ 10, it is determined that there is no target object in the image.

[0033] In another embodiment, step S2 includes: S21: Receive pixel data of the original image matrix; S22: Perform a one-dimensional discrete cosine transform on the pixel data in the row direction to generate an intermediate matrix of the row transform; S23: Based on the row transformation intermediate matrix, perform a one-dimensional discrete cosine transformation in the column direction to generate a transformation coefficient matrix.

[0034] The working principle and beneficial effects of the above technical solution are as follows: In step S21, the system receives the pixel intensity values ​​arranged in row and column order in the original image matrix, and each pixel value represents the acoustic reflection intensity at the corresponding spatial location.

[0035] In step S22, a one-dimensional discrete cosine transform is performed independently on each row of pixel data in the image matrix. The spatial domain signal in the row direction is decomposed in the frequency domain using a DCT basis function sequence. The original spatial domain pixel values ​​are converted into coefficient representations in the frequency domain to generate an intermediate matrix for row transformation. This process keeps the number of rows of the matrix unchanged while changing the way the data is represented.

[0036] In step S23, the row transformation intermediate matrix is ​​used as input, and the one-dimensional discrete cosine transform operation is repeatedly performed on each column of data. The calculation process of two-dimensional DCT is completed through the frequency domain transformation in the column direction, and finally a complete transform coefficient matrix is ​​generated. This matrix contains the complete representation information of the original image in the two-dimensional frequency space.

[0037] In another embodiment, step S3 includes: S31: Extract the coefficient values ​​at each position in the transformation coefficient matrix; S32: Perform absolute value operation on each coefficient value to generate an absolute value matrix; S33: Based on the absolute value matrix, perform logarithmic operations on each element to generate an intensity feature matrix.

[0038] The working principle and beneficial effects of the above technical solution are as follows: In step S31, the system reads the complex coefficient values ​​corresponding to each spatial position (i,j) in the transformation coefficient matrix Co one by one. These coefficients contain real and imaginary part information.

[0039] In step S32, the modulus of each complex coefficient is calculated to obtain the absolute value. The absolute values ​​of all positions are arranged in the original space to form an absolute value matrix to eliminate the influence of phase information.

[0040] In step S33, the natural logarithm operation is performed on each element in the absolute value matrix to generate the intensity feature matrix. The logarithmic transformation can compress the dynamic range of the data, amplify the differences of small values, and facilitate subsequent feature analysis processing.

[0041] In another embodiment, step S4 includes: S41: Obtain the number of rows and columns of the intensity feature matrix; S42: Select the smaller value between the number of rows and the number of columns as the splitting reference value; S43: Determine the region size parameters according to a preset ratio based on the segmentation reference value; S44: Based on the region size parameter, extract region feature matrices of the same size from the upper left, middle and lower right corners of the intensity feature matrix.

[0042] The working principle and beneficial effects of the above technical solution are as follows: In step S41, the number of rows M and the number of columns N of the intensity feature matrix I are obtained through the matrix dimension query function. These two parameters determine the spatial range that can be used for feature extraction.

[0043] In step S42, the number of rows M and the number of columns N are compared, and the smaller value is selected as the segmentation reference value m=min(M,N) to ensure that the extracted region matrix is ​​within the image boundary range.

[0044] In step S43, the segmentation reference value m is divided by the preset segmentation coefficient K=3 to calculate the region side length d=m / 3. This parameter determines the spatial size of each feature region.

[0045] In step S44, spatial localization and extraction of three regions are performed on the intensity feature matrix: the upper left region extracts a d×d submatrix F1 starting from position (0,0); the middle region extracts a d×d submatrix F2 starting from position ((Md) / 2,(Nd) / 2); and the lower right region extracts a d×d submatrix F3 starting from position (Md,Nd).

[0046] In another embodiment, step S5 includes: S51: Receives multiple region feature matrices; S52: Perform a summation operation on the elements within the feature matrix of each region to generate the region sum value; S53: Divide the sum of each region by the number of elements in the corresponding region to generate a sequence of region feature values.

[0047] The working principle and beneficial effects of the above technical solution are as follows: In step S51, the system simultaneously receives three regional feature matrices F1, F2, and F3, each containing d×d elements.

[0048] In step S52, all elements in the feature matrix of each region are summed to obtain the sum value of each region.

[0049] In step S53, the sum of each region is divided by the total number of elements in that region, d², to calculate the average characteristic value of each region and generate a sequence of region characteristic values ​​containing f1, f2, and f3. These values ​​reflect the average concentration of frequency domain energy in different regions.

[0050] In another embodiment, step S6 includes: S61: Extract the first feature value f1, the second feature value f2, and the third feature value f3 from the region feature value sequence; S62: Calculate the ratio R1 of the first eigenvalue to the second eigenvalue, where R1 = f1 / f2; S63: Calculate the ratio R2 of the first eigenvalue to the third eigenvalue, where R2 = f1 / f3; S64: Perform a decision operation based on the ratios R1 and R2 to generate a result determining the existence of the target.

[0051] The working principle and beneficial effects of the above technical solution are as follows: In step S61, three numerical parameters are extracted sequentially from the region feature value sequence: the upper left region feature value f1, the middle region feature value f2, and the lower right region feature value f3.

[0052] In step S62, the first ratio calculation is performed, and the ratio relationship between the feature values ​​of the upper left corner region and the middle region is obtained by using the formula R1=f1 / f2. This ratio reflects the frequency domain energy difference between the image edge and the center region.

[0053] In step S63, the second ratio calculation is performed, and the ratio of the feature values ​​of the upper left corner region and the lower right corner region is obtained by using the formula R2=f1 / f3. This ratio reflects the difference in energy distribution between the diagonal regions of the image.

[0054] In step S64, based on the calculated ratios R1 and R2, the existence of the target is analyzed according to the preset judgment logic rules. By comparing the relationship between these ratios and the preset threshold, a binary judgment result of "the target exists" or "the target does not exist" is finally output.

[0055] In another embodiment, the decision operation in step S64 includes: Operation condition 1: When R1 is greater than or equal to the first threshold T1, the result is that the target object exists; Operation condition 2: When R1 is greater than or equal to the second threshold T2 and less than the first threshold T1, and R2 is greater than or equal to the third threshold T3, the result is that the target object exists. When neither operation condition one nor operation condition two is satisfied, the result is that the target object does not exist. Among them, T1>T2>1, T3>1.

[0056] The working principle and beneficial effects of the above technical solution are as follows: The judgment operation adopts a hierarchical threshold comparison strategy: In the first-level judgment, when the ratio R1 is greater than or equal to the first threshold T1, it indicates that the frequency domain energy of the upper left region is significantly higher than that of the middle region, and the judgment result of "the target object exists" is directly output.

[0057] In the second-level determination, when R1 is between the second threshold T2 and the first threshold T1 (T2 ≤ R1 < T1), it is necessary to further check whether R2 is greater than or equal to the third threshold T3. If both of these conditions are met, it is determined that "the target object exists".

[0058] In the fallback determination, when none of the above condition combinations are satisfied, the determination result of "the target object does not exist" is output. The threshold settings follow the constraint relationship of T1 > T2 > 1 and T3 > 1 to ensure the rationality of the determination logic.

[0059] In another embodiment, step S44 includes: S441: Take the upper left corner position of the intensity feature matrix as the starting position of the first region; S442: Based on the number of rows M and the number of columns N of the intensity feature matrix and the region size parameter d, calculate the starting position of the middle region as ((M - d) / 2, (N - d) / 2); S443: Calculate the starting position of the lower right corner region so that the lower right boundary of this region is aligned with the lower right boundary of the intensity feature matrix.

[0060] The working principle and beneficial effects of the above technical solution are as follows: In step S441, the coordinate origin (0, 0) of the intensity feature matrix I is set as the starting extraction position of the first region F1, and a sub-matrix of size d×d is extracted in the lower right direction from this position.

[0061] In step S442, the starting position coordinates of the middle region F2 are determined as ((M - d) / 2, (N - d) / 2) through the calculation formula, where M and N are the number of rows and the number of columns of the intensity feature matrix respectively, and d is the side length of the region. This position makes the extracted region near the geometric center of the entire matrix.

[0062] In step S443, the starting position of the lower right corner region F3 is calculated as (M - d, N - d), ensuring that the right boundary and the lower boundary of the extracted d×d region are exactly aligned with the right boundary and the lower boundary of the intensity feature matrix respectively, realizing the accurate extraction of the lower right corner region of the matrix.

[0063] In another embodiment, step S43 includes: S431: Receive the segmentation reference value m; S432: Divide the segmentation reference value by the preset segmentation coefficient K to generate the region side length value d, where d = m / K, and the preset segmentation coefficient K is 3; S433: Use the region side length value d as the region size parameter for subsequent region extraction.

[0064] The working principle and beneficial effects of the above technical solution are as follows: In step S431, the system receives the segmentation reference value m determined in the previous step, which is equal to the smaller value in the number of rows and columns of the intensity feature matrix.

[0065] In step S432, a division operation is performed to divide the segmentation reference value m by the preset segmentation coefficient K=3, and the region side length value d is calculated using the formula d=m / K. The setting of segmentation coefficient K=3 ensures that the three feature regions can be reasonably distributed in the image space without overlapping.

[0066] In step S433, the calculated region side length value d is used as the size parameter for subsequent region extraction operations. This parameter uniformly controls the spatial size of the three feature regions, ensuring the fairness and effectiveness of feature comparison.

[0067] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from the spirit and scope of this invention.

Claims

1. A rapid method for identifying suspicious targets in the field of view of side-scan sonar images, characterized in that, include: S1: Acquire side-scan sonar image data and generate the original image matrix; S2: Based on the original image matrix, perform a two-dimensional discrete cosine transform operation to generate a transform coefficient matrix; S3: Based on the transformation coefficient matrix, calculate the logarithmic transformation of the absolute values ​​of the coefficients to generate the intensity feature matrix; S4: Based on the dimensional information of the intensity feature matrix, extract multiple region feature matrices according to preset segmentation rules; S5: Based on multiple regional feature matrices, calculate the feature statistics of each region and generate a sequence of regional feature values; S6: Based on the regional feature value sequence, calculate the ratio relationship between feature values, and generate the target existence determination result according to the ratio determination rule.

2. The rapid discrimination method for suspicious targets in the field of view of side-scan sonar images according to claim 1, characterized in that, Step S2 includes: S21: Receive pixel data of the original image matrix; S22: Perform a one-dimensional discrete cosine transform on the pixel data in the row direction to generate an intermediate matrix of the row transform; S23: Based on the row transformation intermediate matrix, perform a one-dimensional discrete cosine transformation in the column direction to generate a transformation coefficient matrix.

3. The rapid discrimination method for suspicious targets in the field of view of side-scan sonar images according to claim 1, characterized in that, Step S3 includes: S31: Extract the coefficient values ​​at each position in the transformation coefficient matrix; S32: Perform absolute value operation on each coefficient value to generate an absolute value matrix; S33: Based on the absolute value matrix, perform logarithmic operations on each element to generate an intensity feature matrix.

4. The rapid discrimination method for suspicious targets in the field of view of side-scan sonar images according to claim 1, characterized in that, Step S4 includes: S41: Obtain the number of rows and columns of the intensity feature matrix; S42: Select the smaller value between the number of rows and the number of columns as the splitting reference value; S43: Determine the region size parameters according to a preset ratio based on the segmentation reference value; S44: Based on the region size parameter, extract region feature matrices of the same size from the upper left, middle and lower right corners of the intensity feature matrix.

5. The rapid discrimination method for suspicious targets in the field of view of side-scan sonar images according to claim 1, characterized in that, The S5 steps include: S51: Receives multiple region feature matrices; S52: Perform a summation operation on the elements within the feature matrix of each region to generate the region sum value; S53: Divide the sum of each region by the number of elements in the corresponding region to generate a sequence of region feature values.

6. The rapid discrimination method for suspicious targets in the field of view of side-scan sonar images according to claim 1, characterized in that, Step S6 includes: S61: Extract the first feature value f1, the second feature value f2, and the third feature value f3 from the region feature value sequence; S62: Calculate the ratio R1 of the first eigenvalue to the second eigenvalue, where R1 = f1 / f2; S63: Calculate the ratio R2 of the first eigenvalue to the third eigenvalue, where R2 = f1 / f3; S64: Perform a decision operation based on the ratios R1 and R2 to generate a result determining the existence of the target.

7. The rapid discrimination method for suspicious targets in the field of view of side-scan sonar images according to claim 6, characterized in that, The decision operation in step S64 includes: Operation condition 1: When R1 is greater than or equal to the first threshold T1, the result is that the target object exists; Operation condition 2: When R1 is greater than or equal to the second threshold T2 and less than the first threshold T1, and R2 is greater than or equal to the third threshold T3, the result is that the target object exists. When neither operation condition one nor operation condition two is satisfied, the result is that the target object does not exist. Among them, T1>T2>1, T3>1.

8. The rapid discrimination method for suspicious targets in the field of view of side-scan sonar images according to claim 4, characterized in that, Step S44 includes: S441: Take the top left corner of the intensity feature matrix as the starting position of the first region; S442: Based on the number of rows M and columns N of the intensity feature matrix and the region size parameter d, the starting position of the central region is calculated as ((Md) / 2, (Nd) / 2); S443: Calculate the starting position of the lower right corner region so that the lower right boundary of the region is aligned with the lower right boundary of the intensity feature matrix.

9. The rapid discrimination method for suspicious targets in the field of view of side-scan sonar images according to claim 4, characterized in that, Step S43 includes: S431: Receive segmentation reference value m; S432: Divide the segmentation reference value by the preset segmentation coefficient K to generate the region side length value d, where d=m / K, and the preset segmentation coefficient K is 3; S433: Use the region side length value d as the region size parameter for subsequent region extraction.

Images