An online target recognition method and system for forward-looking sonar in an extremely shallow water environment

Abstract

The application discloses a kind of extremely shallow water environment under forward-looking sonar target online identification method and system, it is related to underwater target identification field.To solve the problem that the detail feature of existing underwater sonar detection shallow beach nearshore target is less, outline is fuzzy, imaging background false target is much and with target similarity is high.Problems include: S1, based on forward-looking sonar collection shallow water environment under sonar image, and image is denoised in combination with median filter and wavelet transform;S2, fuzzy enhancement, segmentation and morphological processing are used to optimize the image after filtering, and the outline of the target is extracted;S3, according to the position of suspected target in the current frame sonar image, it is assumed that the motion of platform between two frames is uniform linear motion, and the position of suspected target in next frame is calculated in combination with the attitude and motion information of platform;By nearest neighbor data association algorithm, the position of suspected target in next frame sonar image is associated with the position data of suspected target calculated, and the real target is obtained.

Description

Technical Field

[0001] This invention relates to the field of underwater target identification technology, and more specifically, to an online identification method and system for forward-looking sonar targets in extremely shallow water environments. Background Technology

[0002] The vast ocean holds abundant resources and possesses immense development potential. With global development, marine resources have attracted the attention and importance of various countries, leading to vigorous development of marine resources and the marine economy. In marine development, underwater exploration can provide richer and more accurate data about the ocean, accelerating our understanding and development of it. Therefore, with the increasing demand for underwater exploration, underwater target identification has become a hot topic in recent years. Currently, underwater target identification research focuses on deep-sea areas, while research on underwater target identification in shallow, nearshore areas is relatively limited. Compared to deep-sea areas, sound waves have a strong scattering effect in shallow, nearshore areas. Sonar images suffer from problems such as high noise, non-uniform imaging backgrounds, low contrast, color shifts, and image distortion. Target features in sonar images are fewer, outlines are blurred, and they are easily submerged by noise. Furthermore, shallow lakebeds often contain many rocks, and the imaging background contains numerous false targets with high similarity to the actual targets, making it difficult to extract target features and identify targets from sonar images.

[0003] Therefore, in response to the aforementioned problems and difficulties in sonar image recognition of nearshore targets in shallow waters, there is an urgent need for an online recognition technology for forward-looking sonar images of small targets nearshore in shallow waters. Summary of the Invention

[0004] The technical problem to be solved by this invention is:

[0005] Existing underwater sonar images suffer from problems such as high noise, non-uniform imaging background, low contrast, easy color shift, and image distortion. They also suffer from limited detailed features of nearshore targets in shallow waters, blurred outlines, and numerous false targets in the imaging background that are highly similar to the actual targets.

[0006] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0007] This invention provides an online identification method for forward-looking sonar targets in extremely shallow water environments, comprising the following steps:

[0008] S1. The forward-looking sonar is mounted on an amphibious unmanned vehicle platform. Sonar images in shallow water environments are collected based on the forward-looking sonar, and the images are denoised by median filtering and wavelet transform.

[0009] S2. The filtered image is optimized using blur enhancement, segmentation, and morphological processing to extract the contour of the target.

[0010] S3. Based on the location of the suspected target in the current frame of the sonar image, assuming that the platform moves at a constant speed in a straight line between the two frames, and combining the platform's attitude and motion information, calculate the location of the suspected target in the next frame.

[0011] For the location of a suspected target in the next frame of sonar image, the nearest neighbor data association algorithm is used to associate it with the calculated location data of the suspected target. The associated suspected targets in the next frame are retained. False targets are gradually removed during the platform's movement, and real targets are selected.

[0012] Furthermore, the specific process in S2 includes: first, performing blur enhancement, image segmentation, and image dilation on the filtered image to enhance the contour features of the target; then, performing secondary segmentation, contour detection, and contour area filtering on the image to remove false areas and further enhance the contour features; finally, performing secondary dilation and contour detection on the image to extract the contour of the target.

[0013] Furthermore, the blur enhancement process in S2 is as follows:

[0014] The filtered M× image R is represented as a blur matrix as follows:

[0015]

[0016] in Represents a certain pixel x ij The degree to which a certain characteristic is possessed, μ ij The fuzzy membership function defined for this feature is as follows:

[0017]

[0018] in e F is the exponential fuzzy factor. d x is the reciprocal fuzzy factor. max For the maximum gray value of a pixel in image R, the membership transformation is:

[0019]

[0020] μ ij ′=T r (μ ij )=T1(T r-1 (μ ij r = 1, 2, ...

[0021] x ij ′=G -1 (μ ij ′),x ij ′∈′

[0022] After r membership transformations, the enhanced image′ is obtained from the inverse membership function to improve the contrast between the target and the background.

[0023] Furthermore, the image segmentation process in S2 is specifically as follows: based on the 3σ criterion, each row of data in the sonar image is binarized, that is: each row of data representing distance information in the sonar image is divided into a group of data to be detected, and the mean and standard deviation of each group of data to be detected are calculated; the pixel values ​​with values ​​distributed in (μ-3σ,μ+3σ) are set to 255, and the pixel values ​​with values ​​outside (μ-3σ,μ+3σ) are set to 0.

[0024] Furthermore, the contour detection in S2 specifically involves using the findContours function for contour detection.

[0025] Furthermore, the contour area filtering in S2 specifically involves: filtering suspected targets by calculating the area within the contour, and removing smaller and larger false targets.

[0026] Furthermore, S3 includes the following steps:

[0027] S31: Determine the azimuth and distance of the suspected target in the sonar coordinate system. The calculation formula is as follows:

[0028] θ=(x osp *b r -f / 2)*r d

[0029] r = (y osp *r r )

[0030] Where x osp y osp These are the x and y coordinates of the suspected target in the sonar image coordinate system, respectively, b r r r These represent the azimuth resolution and range resolution, respectively, where f is the horizontal opening angle of the sonar, and r... d Let θ be a constant for converting degrees to radians, and let θ and r be the azimuth and distance of the suspected target in the sonar coordinate system, respectively.

[0031] The coordinates of the suspected target in the sonar coordinate system are calculated from its azimuth and range in the sonar coordinate system. The specific calculation formula is as follows:

[0032] x os = r*cos(θ)

[0033] y os = r*sin(θ)

[0034] z 0s =Du

[0035] Where x os y os z 0s These are the x, y, and z coordinates of the suspected target in the sonar coordinate system, D. u For the depth of multi-platform;

[0036] S32: For the suspected target location in the sonar coordinate system in S31, calculate the target's position in the platform's center of gravity coordinate system. The calculation formula is as follows:

[0037] x ou =x os +x c

[0038] y ou =y os +y c

[0039] z ou =z 0s +z c

[0040] Where x ou y ou z ou These are the x, y, and z coordinates of the suspected target in the platform's center-of-gravity coordinate system, respectively. c y c z c These are the x, y, and z position compensations from the sonar to the platform's center of gravity, respectively.

[0041] S33: Based on the position of the suspected target obtained in S32 in the platform's center-of-gravity coordinate system, calculate the target's position in the northeast coordinate system. The calculation formula is as follows:

[0042] p oNE =R*p ou +p uNE

[0043] Where p oNE p is the northeast depth location coordinate of the target. ou Let p be the position coordinates of the target in the platform's center of gravity coordinate system. uNE The coordinates of the platform's northeast depth location in the northeast coordinate system;

[0044] R is the six-degree-of-freedom rotation matrix of the platform:

[0045]

[0046] Where φ, θ, and ψ are the platform's roll angle, pitch angle, and heading angle, respectively;

[0047] S34: Assuming the platform's motion between two frames is uniform linear motion, and combining the platform's roll angle, pitch angle, heading angle, and velocity information, based on the coordinates of the suspected target in the previous frame in the NE coordinate system obtained in S33, the kinematic model of the underwater vehicle is used to calculate the coordinates of the suspected target in the NE coordinate system in the next frame.

[0048] p oNE ′=R*p v +p nNE

[0049] p v =v*t

[0050] Where p oNE ′ represents the northeast-deep coordinates of the suspected target in the next frame, v represents the three-dimensional velocity of the platform in x, y, and z, and t represents the time interval between two frames;

[0051] S35: For suspected targets in the next frame of sonar image, perform calculations S31-S33 to obtain the coordinates of the suspected targets in the northeast coordinate system. Then, use the nearest neighbor data association algorithm to associate the coordinates with the estimated suspected target coordinates in S34, retain the associated suspected targets in the next frame, and gradually filter out the real targets during the platform's movement.

[0052] Furthermore, the removal of false targets in S6 includes the following process:

[0053] Establish a container `current_vector` to store current feature information, a container `infer_vector` to store current feature position inference information, and a container `target_floww_vector` to store continuous frame matching feature information; calculate the centroid of the current frame features and store it sequentially into `current_vector`, store the position inference information of the current frame centroid into `infer_vector`, since there is no previous frame information to match it initially, store the centroid of the current frame features sequentially into `target_floww_vector`, and assign a corresponding feature ID number to each feature; then perform calculations for the next frame.

[0054] Calculate the centroid of the current frame and store it in the container current_vector; sequentially retrieve the feature information in the container current_vector and use the nearest neighbor association algorithm to associate and match it with the inferred information of the previous frame; if the corresponding association conditions are met, store the feature in the container target_floww_vector at the position of the corresponding feature of the previous frame and mark it with the same ID; if the association conditions are not met, store it as a new feature in the container target_floww_vector and assign a new feature ID;

[0055] Based on the platform's motion information, the position of the centroid of the current frame features in the container current_vector is calculated sequentially and stored in the container infer_vector; features that have not been updated for three consecutive frames in the container target_floww_vector are considered false features and are deleted from the container target_floww_vector.

[0056] A forward-looking sonar target online identification system for extremely shallow water environments, the system having a program module corresponding to the steps of the method described in any of the above technical solutions, and executing the steps in the above-described forward-looking sonar target online identification method for extremely shallow water environments during runtime.

[0057] A computer-readable storage medium storing a computer program configured to, when invoked by a processor, implement the steps of the online identification method for forward-looking sonar targets in extremely shallow water environments as described in any of the above technical solutions.

[0058] Compared with the prior art, the beneficial effects of the present invention are:

[0059] This invention designs a noise removal method combining median filtering and wavelet transform, effectively removing most of the salt-and-pepper noise and high-frequency noise from sonar images. Blur enhancement is used to improve the contrast between the target and the background. Image segmentation separates the shallow water background from suspected targets. Addressing the issue of online target identification, this invention performs target recognition based on platform motion information. This invention dynamically and online filters suspected targets and identifies genuine targets during platform movement, improving the online identification accuracy of nearshore targets in shallow waters.

[0060] This invention is applicable to the identification of small targets near the shore in shallow waters of lakes and seabeds. It has a wide detection range and high recognition rate. When applied to multi-purpose unmanned transport platforms equipped with the corresponding equipment, it can undertake tasks such as obstacle avoidance in water, underwater target search and rescue, landing and exploration of unmanned islands. Attached Figure Description

[0061] Figure 1 This is a flowchart of the sonar image optimization processing in an embodiment of the present invention;

[0062] Figure 2 This is a flowchart illustrating the main process of target recognition based on platform motion information data association in this embodiment of the invention.

[0063] Figure 3 This is a diagram illustrating the optimized processing of nearshore sonar images of shallow waters and the recognition of small targets in an embodiment of the present invention. Detailed Implementation

[0064] To enable those skilled in the art to better understand the present invention, exemplary embodiments or examples of the present invention will be described below in conjunction with the accompanying drawings. Obviously, the described embodiments or examples are merely some, not all, of the embodiments or examples of the present invention. All other embodiments or examples obtained by those skilled in the art based on the embodiments or examples of the present invention without inventive effort should fall within the scope of protection of the present invention.

[0065] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0066] Example 1

[0067] Combination Figures 1 to 3 As shown, this invention provides an online identification method for forward-looking sonar targets in extremely shallow water environments, comprising the following steps:

[0068] S1. The forward-looking sonar is mounted on an amphibious unmanned vehicle platform. The forward-looking sonar collects sonar images in shallow water environments and combines median filtering and wavelet transform to denoise the images.

[0069] Median filtering replaces the barrier pixel value with the median of the current pixel's neighborhood, effectively removing isolated points and salt-and-pepper noise while preserving edge features in the image. Wavelet transform improves the image blurring caused by median filtering, while also preserving image details and filtering out high-frequency noise. Table 1 shows the results of median filtering, wavelet transform filtering, combined median and wavelet transform filtering, mean filtering, Gaussian filtering, and bilateral filtering applied to the same sonar image dataset, with the PSNR (Peak Signal-to-Noise Ratio) of the filtered results calculated.

[0070] Table 1

[0071]

[0072]

[0073] As can be seen from the PSNR results of the above filtering algorithms, the filtering algorithm combining median filtering and wavelet transform of this invention has the best noise removal effect.

[0074] S2, such as Figure 1 and Figure 3 As shown, blur enhancement, segmentation, and morphological processing are used to optimize the filtered image and extract the target contour.

[0075] First, for the filtered sonar image, blur enhancement is applied to edge contour features and detail resolution, i.e.:

[0076] The filtered M× image R is represented as a blur matrix as follows:

[0077]

[0078] in Represents a certain pixel x ij The degree to which a certain characteristic is possessed, μ ij The fuzzy membership function defined for this feature is as follows:

[0079]

[0080] Where F e F is the exponential fuzzy factor. d x is the reciprocal fuzzy factor. max For the maximum gray value of a pixel in image R, the membership transformation is:

[0081]

[0082] μ ij ′=T r (μ ij )=T1(T r-1 (μ ij r = 1, 2, ...

[0083] x ij ′=G -1 (μ ij ′),x ij '∈R'

[0084] After r membership transformations, the enhanced image R′ is obtained from the inverse membership function to improve the contrast between the target and the background.

[0085] Image segmentation is performed specifically: threshold segmentation of sonar images based on the 3σ criterion of grayscale. The 3σ criterion is based on repeated measurements with equal precision following a normal distribution, but interference or noise from singular data makes it difficult to maintain a normal distribution. The 3σ criterion is applied when the mean is μ and the variance is σ. 2In a normal distribution, the probability of a value falling within (μ-σ, μ+σ) is 0.6826, within (μ-2σ, μ+2σ) is 0.9544, and within (μ-3σ, μ+3σ) is 0.9974. For a random error in a normal distribution, the probability of falling outside ±3σ is only 0.27%. Therefore, if the absolute value of the residual error νi of a measurement in a set of data is greater than 3σ, that measurement is considered an outdated value and should be discarded. An error equal to ±3σ is usually considered the limiting error. Each row of sonar image data is filtered using the 3σ criterion, retaining the highlighted portions returned by hard objects such as targets and removing most of the noise and interference.

[0086] The sonar image data is binarized based on the 3σ criterion, that is, each row of data representing distance information in the sonar image is divided into a group of data to be detected, and the mean and standard deviation of each group of data to be detected are calculated.

[0087]

[0088] Where μ i Let σ be the mean of the data to be measured in the i-th row of the sonar image. i Let x be the standard deviation of the data to be measured in the i-th row of the sonar image. ij Let be the pixel grayscale value in the i-th row and j-th column of the sonar image.

[0089] Pixel values ​​within the range (μ-3σ, μ+3σ) are set to 255, and pixel values ​​outside this range are set to 0. This retains the highlights returned by hard materials such as the target, while removing most of the background noise and interference.

[0090] Image dilation is performed by using a structuring element, specifically calculating the maximum value of the overlapping area between the structuring element and the image. This expands the boundaries of objects, increasing their size. The specific calculation formula is as follows:

[0091] (A⊕B)(x,y)=max (i,j)∈B {A(x+i,y+j)}

[0092] Where A is the original image and B is the structuring element.

[0093] The processing utilizes a cross-shaped structuring element, with its value set to 1. A dilation operation is performed by sliding the cross-shaped structuring element across the image, setting the grayscale value of the image pixel at the structuring element's center to the maximum value of the corresponding image region's pixels where the structuring element value is 1. Based on this processing, the true target contour features in the sonar image are further enhanced, and minor noise and false targets are removed.

[0094] Then, the image is segmented a second time based on the 3σ criterion of gray level.

[0095] The `findContours` function in OpenCV is used to perform contour detection on the image, extract the target contour, and filter suspected targets by calculating the area within the contour, removing smaller and larger false targets.

[0096] Finally, the image is also subjected to a second dilation process.

[0097] Similarly, the findContours function in OpenCV is used to perform contour detection on the image, and finally the contour of the target is extracted.

[0098] S3. Identify the target by combining the platform's pose and motion information.

[0099] Even after optimization, sonar images may still contain suspected targets of varying sizes. In order to identify the real targets from these suspected targets, it is necessary to filter these suspected targets in the sonar images.

[0100] False targets within the suspected targets identified by contour detection in sonar images vary randomly in each frame, while the motion of real targets in each frame is related to the platform's motion. Therefore, false targets in sonar images can be effectively removed based on the platform's motion information. Figure 2 As shown, the specific steps include the following:

[0101] S31: Determine the azimuth and distance of the suspected target in the sonar coordinate system. The calculation formula is as follows:

[0102] θ=(x osp *b r -f / 2)*r d

[0103] r = (y osp *r r )

[0104] Where x osp y osp These are the x and y coordinates of the suspected target in the sonar image coordinate system, respectively, b r r r These represent the azimuth resolution and range resolution, respectively, where f is the horizontal opening angle of the sonar, and r... d Let θ be a constant for converting degrees to radians, and let θ and r be the azimuth and distance of the suspected target in the sonar coordinate system, respectively.

[0105] The coordinates of the suspected target in the sonar coordinate system are calculated from its azimuth and range in the sonar coordinate system. The specific calculation formula is as follows:

[0106] xos = r*cos(θ)

[0107] y os = r*sin(θ)

[0108] z 0s =D u

[0109] Where x os y os z 0s These are the x, y, and z coordinates of the suspected target in the sonar coordinate system, and D. u For the depth of multi-platform;

[0110] S32: For the suspected target location in the sonar coordinate system in S31, calculate the target's position in the platform's center of gravity coordinate system. The calculation formula is as follows:

[0111] x ou =x os +x c

[0112] y ou =y os +y c

[0113] z ou =z 0s +z c

[0114] Where x ou y ou z ou These are the x, y, and z coordinates of the suspected target in the platform's center-of-gravity coordinate system, respectively. c y c z c These are the x, y, and z position compensations from the sonar to the platform's center of gravity, respectively.

[0115] S33: Based on the position of the suspected target obtained in S32 in the platform's center-of-gravity coordinate system, calculate the target's position in the northeast coordinate system. The calculation formula is as follows:

[0116] p oNE =R*p ou +p uNE

[0117] Where p oNE p is the northeast depth location coordinate of the target. ou Let p be the position coordinates of the target in the platform's center of gravity coordinate system. uBE The coordinates of the platform's northeast depth location in the northeast coordinate system;

[0118] R is the six-degree-of-freedom rotation matrix of the platform:

[0119]

[0120] Where φ, θ, and ψ are the platform's roll angle, pitch angle, and heading angle, respectively;

[0121] S34: Assuming the platform's motion between two frames is uniform linear motion, and combining the platform's roll angle, pitch angle, heading angle, and velocity information, based on the coordinates of the suspected target in the previous frame in the NE coordinate system obtained in S33, the kinematic model of the underwater vehicle is used to calculate the coordinates of the suspected target in the NE coordinate system in the next frame.

[0122] p oNE ′=R*p v +p oNE

[0123] p v =v*t

[0124] Where p oNE ′ represents the northeast-deep coordinates of the suspected target in the next frame, v represents the three-dimensional velocity of the platform in x, y, and z, and t represents the time interval between two frames;

[0125] S35: For suspected targets in the next frame of sonar image, perform calculations S31-S33 to obtain the coordinates of the suspected targets in the northeast coordinate system. Then, use the nearest neighbor data association algorithm to associate the coordinates with the estimated suspected target coordinates in S34, retain the associated suspected targets in the next frame, and gradually filter out the real targets during the platform's movement.

[0126] A container `current_vector` is created to store current feature information, a container `infer_vector` to store inferred current feature positions, and a container `target_floww_vector` to store matching feature information from consecutive frames. Initially, the centroids of the features in the current frame are calculated and stored sequentially in `current_vector`. The inferred positions of the centroids in the current frame are stored in `infer_vector`. Since there is no information from the previous frame to match initially, the centroids of the features in the current frame are stored sequentially in `target_floww_vector`, and a corresponding feature ID number is assigned to each feature. Then, calculations are performed for the next frame.

[0127] Calculate the centroid of the current frame and store it in the container `current_vector`. Sequentially retrieve feature information from `current_vector` and use a nearest neighbor algorithm to match it with the inferred information from the previous frame. If the matching condition is met, store the feature in the container `target_floww_vector` at the corresponding position of the feature from the previous frame, marking it with the same ID. If the matching condition is not met, store it as a new feature in the container `target_floww_vector` and assign it a new feature ID.

[0128] Based on the platform's motion information, the centroids of the current frame features in the container `current_vector` are sequentially calculated and stored in the container `infer_vector`. Features in the container `target_floww_vector` that have not been updated for three consecutive frames (i.e., no matching feature is found in the next frame for three consecutive frames) are considered false features and are removed from the container `target_floww_vector`.

[0129] After the fifth frame, the feature with the longest matching length in the container target_floww_vector is considered the real target each time.

[0130] As shown in Table 2, seven sets of sonar images and data obtained in the experiment were selected. The aforementioned online target recognition method for forward-looking sonar in extremely shallow water environments was used to optimize the sonar images and identify the targets. The feature extraction rate and target recognition rate of the method were then statistically analyzed. The results are shown in Table 2.

[0131] Table 2

[0132]

[0133]

[0134] The high feature extraction rate in Table 2 indicates that the aforementioned sonar image processing method can effectively handle noise in nearshore sonar images of shallow waters. The recognition rate shows that the target recognition technology based on platform motion information data association can effectively identify small targets nearshore in shallow waters.

[0135] The online identification method (algorithm) for forward-looking sonar targets in extremely shallow water environments proposed in this invention is the underlying technical core of this invention, and various products can be derived based on the algorithm.

[0136] Based on the method proposed in this invention, an online identification system for forward-looking sonar targets in extremely shallow water environments is developed using a programming language. This system has program modules corresponding to the steps of the above-mentioned technical solution, and executes the steps in the above-mentioned online identification method for forward-looking sonar targets in extremely shallow water environments when running.

[0137] The developed system (software) computer program is stored on a computer-readable storage medium. This computer program is configured to implement the steps of the above-described online identification method for forward-looking sonar targets in extremely shallow water environments when invoked by a processor. In other words, the invention is materialized on a carrier, becoming a computer program product.

[0138] Various implementations of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, application-specific integrated circuits (ASICs), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include: implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0139] The computational programs (also referred to as programs, software, software applications, or code) of this invention include machine instructions of a programmable processor and can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or apparatus (e.g., disk, optical disk, memory, programmable logic device PLD) for providing machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0140] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A method for online identification of forward-looking sonar targets in extremely shallow water environments, characterized in that, Includes the following steps: S1. The forward-looking sonar is mounted on an amphibious unmanned vehicle platform. Sonar images in shallow water environments are collected based on the forward-looking sonar, and the images are denoised by median filtering and wavelet transform. S2. The filtered image is optimized using blur enhancement, segmentation, and morphological processing to extract the contour of the target. S3. Based on the location of the suspected target in the current frame of the sonar image, assuming that the platform moves at a constant speed in a straight line between the two frames, and combining the platform's attitude and motion information, calculate the location of the suspected target in the next frame. For the location of a suspected target in the next frame of sonar image, the nearest neighbor data association algorithm is used to associate it with the calculated location data of the suspected target, and the associated suspected target in the next frame is retained. False targets are gradually removed during the platform's movement, and real targets are selected. S3 includes the following steps: S31: Determine the azimuth and distance of the suspected target in the sonar coordinate system. The calculation formula is as follows: in , These are the x and y coordinates of the suspected target in the sonar image coordinate system, respectively. , These are the azimuth resolution and the range resolution, respectively. For the horizontal opening angle of the sonar, The constant for converting degrees to radians. , These represent the azimuth and distance of the suspected target in the sonar coordinate system. The coordinates of the suspected target in the sonar coordinate system are calculated from its azimuth and range in the sonar coordinate system. The specific calculation formula is as follows: in , , These represent the x, y, and z coordinates of the suspected target in the sonar coordinate system. For the depth of multi-platform; S32: For the suspected target location in the sonar coordinate system in S31, calculate the target's position in the platform's center of gravity coordinate system. The calculation formula is as follows: in , , These are the x, y, and z coordinates of the suspected target in the platform's center-of-gravity coordinate system. , , These are the x, y, and z position compensations from the sonar to the platform's center of gravity, respectively. S33: Based on the position of the suspected target obtained in S32 in the platform's center-of-gravity coordinate system, calculate the target's position in the northeast coordinate system. The calculation formula is as follows: in The target's northeast-deep location coordinates. Let the target's position coordinates be in the platform's center of gravity coordinate system. The coordinates of the platform's northeast depth location in the northeast coordinate system; For the platform's six-degree-of-freedom rotation matrix: in , , These are the platform's roll angle, pitch angle, and heading angle, respectively. S34: Assuming the platform's motion between two frames is uniform linear motion, and combining the platform's roll angle, pitch angle, heading angle, and velocity information, based on the coordinates of the suspected target in the previous frame in the NE coordinate system obtained in S33, the kinematic model of the underwater vehicle is used to calculate the coordinates of the suspected target in the NE coordinate system in the next frame. in Let v be the northeast-east depth coordinates of the suspected target in the next frame, v be the three-dimensional velocity of the platform in x, y, and z, and t be the time interval between two frames. S35: For suspected targets in the next frame of sonar image, perform calculations S31-S33 to obtain the coordinates of the suspected targets in the northeast coordinate system. Then, use the nearest neighbor data association algorithm to associate the coordinates with the estimated suspected target coordinates in S34, retain the associated suspected targets in the next frame, and gradually filter out the real targets during the platform's movement.

2. The online identification method for forward-looking sonar targets in extremely shallow water environments according to claim 1, characterized in that, The specific process in S2 includes: First, the filtered image is subjected to blur enhancement, image segmentation and image dilation to enhance the contour features of the target. Then, the image is subjected to secondary segmentation, contour detection and contour area screening to remove false areas and further enhance the contour features. Finally, the image is subjected to secondary dilation and contour detection to extract the contour of the target.

3. The online identification method for forward-looking sonar targets in extremely shallow water environments according to claim 1, characterized in that, The blur enhancement process in S2 is as follows: After filtering The image R is represented as a blur matrix as follows: in Represents a certain pixel The degree to which a certain characteristic is possessed. The fuzzy membership function defined for this feature is as follows: in For the exponential fuzzy factor, The reciprocal fuzzy factor, For the maximum gray value of a pixel in image R, the membership transformation is: After r membership transformations, the enhanced image is obtained from the inverse membership function. This enhances the contrast between the target and the background.

4. The online identification method for forward-looking sonar targets in extremely shallow water environments according to claim 1, characterized in that, The image segmentation process in S2 is specifically based on 3 The criterion binarizes each row of data in the sonar image, that is: each row of data representing distance information in the sonar image is divided into a group of data to be detected, and the mean and standard deviation of each group of data to be detected are calculated; the pixel values ​​with values ​​distributed in (μ-3σ,μ+3σ) are set to 255, and the pixel values ​​with values ​​outside (μ-3σ,μ+3σ) are set to 0.

5. The online identification method for forward-looking sonar targets in extremely shallow water environments according to claim 1, characterized in that, Contour detection in S2 specifically involves using the findContours function.

6. The online identification method for forward-looking sonar targets in extremely shallow water environments according to claim 1, characterized in that, The contour area filtering in S2 specifically involves: filtering suspected targets by calculating the area within the contour, and removing smaller and larger false targets.

7. The online identification method for forward-looking sonar targets in extremely shallow water environments according to claim 6, characterized in that, Removing false targets in S6 includes the following process: Establish a container `current_vector` to store current feature information, a container `infer_vector` to store current feature position inference information, and a container `target_floww_vector` to store continuous frame matching feature information; calculate the centroid of the current frame features and store it into `current_vector` in sequence; store the position inference information of the current frame centroid into `infer_vector`; since there is no previous frame information to match it initially, store the centroid of the current frame features into `target_floww_vector` in sequence, and assign a corresponding feature ID number to each feature; Then proceed with the calculation for the next frame; Calculate the centroid of the current frame and store it in the container current_vector; sequentially retrieve the feature information in the container current_vector and use the nearest neighbor association algorithm to associate and match it with the inferred information of the previous frame; if the corresponding association conditions are met, store the feature in the container target_floww_vector at the position of the corresponding feature of the previous frame and mark it with the same ID; if the association conditions are not met, store it as a new feature in the container target_floww_vector and assign a new feature ID; Based on the platform's motion information, the position of the centroid of the current frame features in the container current_vector is calculated sequentially and stored in the container infer_vector; features that have not been updated for three consecutive frames in the container target_floww_vector are considered false features and are deleted from the container target_floww_vector.

8. An online identification system for forward-looking sonar targets in extremely shallow water environments, characterized in that, The system has a program module corresponding to the steps of the method described in any one of claims 1 to 7, and executes the steps in the above-described online identification method for forward-looking sonar targets in extremely shallow water environments when it is run.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program configured to, when invoked by a processor, implement the steps of the online identification method for forward-looking sonar targets in extremely shallow water environments as described in any one of claims 1 to 7.

Images