Underwater target recognition method and system

Abstract

The present invention application provides an underwater target recognition method and system, including: obtaining an underwater image to be recognized and an image of the regional environment of the target area; recognizing the regional environment image to obtain regional information; dividing the regional environment image into a plurality of sub-images and performing enhancement processing; obtaining flow field data of the target area and simulating the flow field through fluid mechanics to obtain a predicted movement trajectory; performing comparative analysis based on each enhanced sub-image and the underwater image to be recognized to obtain a preliminary comparison result; and correcting the preliminary comparison result based on the predicted movement trajectory and a static water pattern feature map to obtain a target recognition result. Compared with the prior art, which considers technical solutions other than water bodies, the present invention application takes into account the disturbance factors of the underwater flow field by obtaining the flow field data of the target area and using fluid mechanics simulation to predict the movement trajectory and then correct the comparison result, thereby effectively improving the accuracy of target recognition.

Description

Technical Field

[0001] The present invention relates to the field of image recognition, and in particular to an underwater target recognition method and system. Background Art

[0002] Underwater target recognition is of great significance for ocean exploration and underwater resource development. Currently, existing underwater target recognition methods often struggle to adapt to complex environments, especially when it comes to adapting to dynamic changes and coping with multiple interferences. They generally lack comprehensive consideration of environmental factors, resulting in recognition accuracy and stability that are difficult to meet practical requirements.

[0003] In some existing technologies, in order to reduce the impact of environmental factors on recognition accuracy, they mainly collect lighting data in the environment directly through a sensor network, and then use the collection results as reference data to process the collected images to be recognized. Although this processing method takes environmental factors into consideration to a certain extent, it considers influencing factors other than water bodies, and lacks sufficient consideration of the water body itself. Therefore, there is a problem of insufficient recognition accuracy. Summary of the Invention

[0004] The present invention provides an underwater target recognition method and system to solve the technical problem of how to improve recognition accuracy.

[0005] In order to solve the above technical problems, the present invention provides an underwater target recognition method, comprising:

[0006] Acquire in real time an underwater image of the target area to be identified and an image of the regional environment; wherein the regional environment image is collected by a multispectral camera and a laser radar, and the multispectral camera uses bands including visible light, infrared spectrum, and ultraviolet spectrum;

[0007] Identify the regional environment image to obtain regional information; segment the regional environment image into a plurality of sub-images, and enhance each sub-image according to the regional information; acquire flow field data of the target area in real time, and obtain a predicted movement trajectory of the object to be identified through fluid dynamics simulation; and extract static water ripple features based on the regional environment image using a target detection algorithm to generate a static water ripple feature map;

[0008] Performing comparative analysis on each enhanced sub-image and the underwater image to be identified to obtain a preliminary comparison result;

[0009] A comprehensive analysis is performed based on the predicted movement trajectory, the preset fluid mechanics conditions and the static water ripple feature map to obtain a water ripple category label; the water ripple category label and the predicted movement trajectory are used to correct the preliminary comparison result to obtain a target recognition result of the object to be identified.

[0010] As a preferred solution, the comparative analysis of each enhanced sub-image and the underwater image to be identified to obtain a preliminary comparison result includes:

[0011] Segmenting the underwater image to be identified based on the enhanced sub-images and the underwater image to be identified, separating the target signal and background noise;

[0012] Performing feature recognition on the target signal to obtain a signal feature set;

[0013] Using a dynamic mapping framework, the signal feature set is updated in real time to obtain a target feature description; wherein the dynamic mapping framework is used to describe the association mapping between environmental parameters and signal features;

[0014] Based on the target feature description, the preliminary comparison result is obtained.

[0015] As a preferred solution, segmenting the underwater image to be identified to separate the target signal and background noise includes:

[0016] Constructing a gray-level co-occurrence matrix of the underwater image to be identified, wherein the gray-level co-occurrence matrix is ​​used to reflect the texture features of the underwater image to be identified;

[0017] Decomposing the gray-level co-occurrence matrix into multiple signal components using independent component analysis to obtain a decomposition result;

[0018] Reconstructing the texture features of the underwater image to be identified according to the energy proportion of each signal component in the decomposition result to obtain a reconstruction result;

[0019] Determining the signal strength of each pixel in the reconstruction result; filtering out pixels having a signal strength greater than a first preset strength threshold from the reconstruction result to obtain a valid pixel combination;

[0020] A target signal is determined according to the effective pixel combination, and the remaining pixel points of the underwater image to be identified are determined as the background noise.

[0021] As a preferred solution, before using the dynamic mapping framework to update the signal feature set in real time, the method includes:

[0022] Acquire a sample image, perform feature recognition on the sample image to obtain sample features; and acquire environmental parameters of an area corresponding to the sample image; wherein the environmental parameters include suspended particle information and illumination change information;

[0023] Screening out significant features from the sample features based on the signal intensity of the sample image; and constructing a dynamic mapping relationship between the significant features and the suspended particle information and the illumination change information using a data table;

[0024] The dynamic mapping framework is obtained according to the dynamic mapping relationship.

[0025] As a preferred solution, the salient features include image contrast, and the suspended particle information includes water turbidity; and the use of a data table to construct a dynamic mapping relationship between the salient features and the suspended particle information and illumination change information includes:

[0026] Based on the water turbidity and light change information, a linear regression model was constructed;

[0027] Obtaining a correlation coefficient based on the linear regression model and the image contrast;

[0028] Using the data table to record the model parameters of the linear regression model, the image contrast and the correlation coefficient, to obtain a recorded result;

[0029] The recording result is recorded as the dynamic mapping relationship.

[0030] As a preferred solution, the water ripple category label is obtained by performing a comprehensive analysis based on the predicted movement trajectory, the preset fluid dynamics conditions and the static water ripple feature map, including:

[0031] According to the time information of the flow field data of the target area, a water ripple dynamic video within the same time period is obtained, and the water ripple dynamic video is split into a water ripple image sequence;

[0032] Setting a feature condition based on feature similarity according to the static watermark feature map, and screening a plurality of watermark images from the watermark image sequence according to the feature condition to obtain a screening result;

[0033] Deduction is performed based on the screening results, the predicted movement trajectory, and preset fluid mechanics conditions to obtain several water pattern category labels.

[0034] As a preferred solution, the water ripple image of the screening result includes interference objects; the deduction based on the screening result, the predicted movement trajectory and the preset fluid mechanics conditions is performed to obtain several water ripple category labels, including:

[0035] Extract the contour points of each interference object in the watermark image of the screening result;

[0036] Taking the contour points of the interference object as a reference, truncating the water ripple lines in each water ripple image containing the interference object in the screening results, extracting the feature points of the water ripple lines, and connecting the feature points to obtain the water ripple motion trajectory;

[0037] Based on the water ripple motion trajectory, calling a pre-trained water ripple prediction model to obtain a water ripple prediction trajectory;

[0038] According to the water ripple predicted trajectory and the predicted movement trajectory, deduction is performed on the basis of preset fluid dynamics conditions to obtain a number of water ripple motion patterns, and the water ripple motion patterns are determined as water ripple category labels.

[0039] As a preferred solution, the real-time acquisition of flow field data of the target area and the obtained predicted movement trajectory of the object to be identified through fluid dynamics simulation include:

[0040] Acquiring flow field data of the target area through an underwater sensor network, and determining an initial trajectory based on the flow field data; wherein the flow field data includes flow velocity information and flow direction information of the flow field;

[0041] Performing real-time analysis on the flow velocity information and flow direction information to obtain the disturbance frequency and disturbance amplitude of the flow field;

[0042] The disturbance frequency and disturbance amplitude of the flow field are decomposed by Fourier transform to determine the characteristic frequency and amplitude fluctuation range of the disturbance;

[0043] Determining trajectory adjustment parameters according to the disturbance characteristic frequency and amplitude fluctuation range;

[0044] A fluid mechanics simulation is performed based on the Navier-Stokes equations, and a fluid model is constructed according to the flow field data. Based on the fluid model, the trajectory adjustment parameters are used to predict the next moment of the initial trajectory to obtain the predicted movement trajectory of the object to be identified.

[0045] As a preferred solution, the use of watermark category labels and predicted movement trajectories to correct the preliminary comparison results to obtain the target recognition result of the object to be identified includes:

[0046] Obtaining a plurality of distinguishing pixel features of each enhanced sub-image and the underwater image to be identified from the preliminary comparison result; each distinguishing pixel feature includes a plurality of pixel points;

[0047] Determining a reference trajectory corresponding to the water ripple motion pattern according to the water ripple category label; analyzing the predicted motion trajectory to obtain a deviation value between the predicted motion trajectory and the reference trajectory;

[0048] Filtering out trajectory points whose deviation values ​​are greater than a preset deviation threshold from the predicted movement trajectory;

[0049] According to the coincidence of the trajectory points and the distinguishing pixel features, the preliminary comparison result is corrected to obtain the target recognition result.

[0050] Accordingly, the present invention also provides an underwater target recognition system, comprising an image acquisition module, a region information processing module, a comparison module and a target recognition module; wherein,

[0051] The image acquisition module is used to acquire in real time the underwater image to be identified and the regional environment image of the target area; wherein the regional environment image is collected by a multispectral camera and a laser radar, and the bands used by the multispectral camera include visible light, infrared spectrum and ultraviolet spectrum;

[0052] The regional information processing module is used to identify the regional environment image and obtain regional information; segment the regional environment image into a plurality of sub-images and perform enhancement processing on each sub-image according to the regional information; obtain flow field data of the target area in real time and obtain the predicted movement trajectory of the object to be identified through fluid dynamics simulation; and use the target detection algorithm to extract static water ripple features based on the regional environment image to generate a static water ripple feature map;

[0053] The processing module is configured to perform comparative analysis on each enhanced sub-image and the underwater image to be identified to obtain a preliminary comparison result;

[0054] The target recognition module is used to perform a comprehensive analysis based on the predicted movement trajectory, the preset fluid dynamics conditions and the static water ripple feature map to obtain a water ripple category label; and use the water ripple category label and the predicted movement trajectory to correct the preliminary comparison result to obtain a target recognition result of the object to be identified.

[0055] As a preferred solution, the target recognition module performs comparative analysis on each enhanced sub-image and the underwater image to be recognized to obtain preliminary comparison results, including:

[0056] The target recognition module segments the underwater image to be identified based on the enhanced sub-images and the underwater image to be identified, and separates the target signal and background noise;

[0057] Performing feature recognition on the target signal to obtain a signal feature set;

[0058] Using a dynamic mapping framework, the signal feature set is updated in real time to obtain a target feature description; wherein the dynamic mapping framework is used to describe the association mapping between environmental parameters and signal features;

[0059] Based on the target feature description, the preliminary comparison result is obtained.

[0060] As a preferred solution, the target recognition module segments the underwater image to be recognized to separate the target signal and background noise, including:

[0061] The target recognition module constructs a gray level co-occurrence matrix of the underwater image to be recognized, and the gray level co-occurrence matrix is ​​used to reflect the texture characteristics of the underwater image to be recognized;

[0062] Decomposing the gray-level co-occurrence matrix into multiple signal components using independent component analysis to obtain a decomposition result;

[0063] Reconstructing the texture features of the underwater image to be identified according to the energy proportion of each signal component in the decomposition result to obtain a reconstruction result;

[0064] Determining the signal strength of each pixel in the reconstruction result; filtering out pixels having a signal strength greater than a first preset strength threshold from the reconstruction result to obtain a valid pixel combination;

[0065] A target signal is determined according to the effective pixel combination, and the remaining pixel points of the underwater image to be identified are determined as the background noise.

[0066] As a preferred solution, the underwater target recognition system further includes a mapping framework construction module, which is configured to:

[0067] Acquire a sample image, perform feature recognition on the sample image to obtain sample features; and acquire environmental parameters of an area corresponding to the sample image; wherein the environmental parameters include suspended particle information and illumination change information;

[0068] Screening out significant features from the sample features based on the signal intensity of the sample image; and constructing a dynamic mapping relationship between the significant features and the suspended particle information and the illumination change information using a data table;

[0069] The dynamic mapping framework is obtained according to the dynamic mapping relationship.

[0070] As a preferred solution, the salient features include image contrast, and the suspended particle information includes water turbidity; the mapping framework construction module uses a data table to construct a dynamic mapping relationship between the salient features and the suspended particle information and illumination change information, including:

[0071] The mapping framework construction module constructs a linear regression model based on water turbidity and light change information;

[0072] Obtaining a correlation coefficient based on the linear regression model and the image contrast;

[0073] Using the data table to record the model parameters of the linear regression model, the image contrast and the correlation coefficient, to obtain a recorded result;

[0074] The recording result is recorded as the dynamic mapping relationship.

[0075] As a preferred solution, the target recognition module performs a comprehensive analysis based on the predicted movement trajectory, the preset fluid dynamics conditions and the static water ripple feature map to obtain a water ripple category label, including:

[0076] According to the time information of the flow field data of the target area, a water ripple dynamic video within the same time period is obtained, and the water ripple dynamic video is split into a water ripple image sequence;

[0077] Setting a feature condition based on feature similarity according to the static watermark feature map, and screening a plurality of watermark images from the watermark image sequence according to the feature condition to obtain a screening result;

[0078] Deduction is performed based on the screening results, the predicted movement trajectory, and preset fluid mechanics conditions to obtain several water pattern category labels.

[0079] As a preferred solution, the watermark image of the screening result includes interference objects; the target recognition module deduces based on the screening result, the predicted movement trajectory and the preset fluid dynamics conditions to obtain several watermark category labels, including:

[0080] The target recognition module extracts the contour points of each interference object in the watermark image of the screening result;

[0081] Taking the contour points of the interference object as a reference, truncating the water ripple lines in each water ripple image containing the interference object in the screening results, extracting the feature points of the water ripple lines, and connecting the feature points to obtain the water ripple motion trajectory;

[0082] Based on the water ripple motion trajectory, calling a pre-trained water ripple prediction model to obtain a water ripple prediction trajectory;

[0083] According to the water ripple predicted trajectory and the predicted movement trajectory, deduction is performed on the basis of preset fluid dynamics conditions to obtain a number of water ripple motion patterns, and the water ripple motion patterns are determined as water ripple category labels.

[0084] As a preferred solution, the regional information processing module acquires the flow field data of the target area in real time and obtains the predicted movement trajectory of the object to be identified through fluid mechanics simulation, including:

[0085] The regional information processing module obtains flow field data of the target area through an underwater sensor network and determines an initial trajectory based on the flow field data; wherein the flow field data includes flow velocity information and flow direction information of the flow field;

[0086] Performing real-time analysis on the flow velocity information and flow direction information to obtain the disturbance frequency and disturbance amplitude of the flow field;

[0087] The disturbance frequency and disturbance amplitude of the flow field are decomposed by Fourier transform to determine the characteristic frequency and amplitude fluctuation range of the disturbance;

[0088] Determining trajectory adjustment parameters according to the disturbance characteristic frequency and amplitude fluctuation range;

[0089] A fluid mechanics simulation is performed based on the Navier-Stokes equations, and a fluid model is constructed according to the flow field data. Based on the fluid model, the trajectory adjustment parameters are used to predict the next moment of the initial trajectory to obtain the predicted movement trajectory of the object to be identified.

[0090] As a preferred solution, the target recognition module uses the watermark category label and the predicted movement trajectory to correct the preliminary comparison result to obtain the target recognition result of the object to be recognized, including:

[0091] The target recognition module obtains a plurality of distinguishing pixel features of each enhanced sub-image and the underwater image to be recognized from the preliminary comparison result; each distinguishing pixel feature includes a plurality of pixel points;

[0092] Determining a reference trajectory corresponding to the water ripple motion pattern according to the water ripple category label; analyzing the predicted motion trajectory to obtain a deviation value between the predicted motion trajectory and the reference trajectory;

[0093] Filtering out trajectory points whose deviation values ​​are greater than a preset deviation threshold from the predicted movement trajectory;

[0094] According to the coincidence of the trajectory points and the distinguishing pixel features, the preliminary comparison result is corrected to obtain the target recognition result.

[0095] Compared with the prior art, the present invention has the following beneficial effects:

[0096] The present invention provides an underwater target recognition method and system, the method comprising: acquiring in real time an underwater image to be recognized and an area environment image of a target area; wherein the area environment image is collected by a multispectral camera and a laser radar, and the multispectral camera uses bands including visible light, infrared spectrum and ultraviolet spectrum; recognizing the area environment image to obtain area information; dividing the area environment image into a plurality of sub-images, and enhancing each sub-image according to the area information; acquiring flow field data of the target area in real time, and obtaining a predicted movement trajectory of the object to be recognized through fluid mechanics simulation; using a target detection algorithm, extracting static water ripple features based on the area environment image to generate a static water ripple feature map; performing comparative analysis between each enhanced sub-image and the underwater image to be recognized to obtain a preliminary comparison result; performing a comprehensive analysis based on the predicted movement trajectory, preset fluid mechanics conditions and the static water ripple feature map to obtain a water ripple category label; and correcting the preliminary comparison result using the water ripple category label and the predicted movement trajectory to obtain a target recognition result of the object to be recognized. Compared with the prior art, the present invention application considers technical solutions other than water bodies. By obtaining the flow field data of the target area and using fluid mechanics simulation to predict the movement trajectory and then correct the comparison result, the disturbance factors of the underwater flow field are taken into account, thereby effectively improving the accuracy of target recognition. In addition, the regional environmental image is divided into several sub-images, and each sub-image is enhanced to different degrees or the same degree according to the regional information. This can improve the quality of the regional environmental image, thereby improving the accuracy of the comparison analysis, obtaining higher quality preliminary comparison results, and further improving the accuracy of the target recognition results. Furthermore, in addition to considering the predicted movement trajectory and the preset fluid mechanics conditions, the static water ripple characteristics are also taken into account, that is, the relatively microscopic and relatively macroscopic movements are integrated to construct water ripple category labels for correcting the preliminary comparison results, further improving the accuracy of the target recognition results. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] Figure 1 : A flow chart of an embodiment of the underwater target recognition method provided by the present invention.

[0098] Figure 2 : A flow chart of a preferred implementation of an embodiment of the underwater target recognition method provided by the present invention.

[0099] Figure 3 : A flow chart of another preferred implementation of an embodiment of the underwater target recognition method provided by the present application.

[0100] Figure 4 : A structural diagram of an embodiment of the underwater target recognition system provided by the present invention. DETAILED DESCRIPTION

[0101] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0102] Example 1

[0103] Please refer to Figure 1 , Figure 1 The present invention provides an underwater target recognition method, comprising steps S101 to S104; wherein,

[0104] Step S101: acquiring an underwater image to be identified and an image of the regional environment of a target area in real time.

[0105] In this step, the regional environment image is collected by a multispectral camera and a lidar, and the bands used by the multispectral camera include visible light, infrared spectrum and ultraviolet spectrum.

[0106] Multispectral cameras and lidar can collect different band data and precise depth information of the target area. Different band data include visible light data, infrared spectrum data and ultraviolet spectrum data. Multispectral images can help capture water surface details, while lidar data helps to obtain precise water surface contours and height information.

[0107] In this embodiment, the underwater image of the target area to be identified can be acquired by one or more image acquisition devices. The type of the image acquisition device can be a camera, and the camera can be a monocular, binocular, or multi-camera camera.

[0108] In some embodiments, the coordinate data of the target area can be acquired using a high-precision positioning system aboard an underwater robot. For example, the longitude and latitude coordinates (116.32, 39.96) at a depth of 50 meters can be set. Using sonar positioning technology combined with an inertial navigation system, the positioning error can be controlled to within 0.5 meters, ensuring the accuracy of image acquisition. Next, to acquire underwater image data, an underwater high-definition camera (e.g., an industrial-grade camera with a resolution of 1920x1080) is used to continuously capture the target area at a rate of 5 frames per second for 10 minutes, capturing approximately 3,000 images. These 3,000 images can be used as images of the regional environment or as sample images for some pre-processing.

[0109] The regional environment image refers to an image of the target area when there is no target to be identified and tracked.

[0110] This embodiment obtains the target recognition result by processing the underwater image to be recognized and the regional environment image separately, and performing comparative analysis.

[0111] It is understood that when there are no targets to be identified in the target area, the comparative analysis results between the underwater image to be identified and the regional environment image are "consistent" or "highly similar." This "high similarity" occurs because the regional environment image undergoes a degree of processing, such as enhancement, in subsequent steps. The processed image will differ from the original image to a certain extent. The purpose of this processing is to improve recognition and enhance the image quality of the target recognition result.

[0112] Step S102: Identify the regional environment image to obtain regional information; divide the regional environment image into several sub-images, and enhance each sub-image according to the regional information; obtain the flow field data of the target area in real time, and obtain the predicted movement trajectory of the object to be identified through fluid mechanics simulation; adopt a target detection algorithm to extract static water ripple features based on the regional environment image and generate a static water ripple feature map.

[0113] In some embodiments, the region information may include information such as regional illumination and regional depth. The illumination and depth may vary between regions. Therefore, this embodiment considers differences in illumination and depth to perform enhancement processing on each sub-image to varying or equal degrees in subsequent steps.

[0114] In this embodiment, the regional environment image can be evenly divided into several sub-images, or unevenly divided into several sub-images depending on the situation. For example, in some application scenarios, the lighting of the external environment of the water body may be uneven. In this case, the regional environment image can be unevenly divided according to the distribution of the lighting.

[0115] Furthermore, in an implementation method in which the regional environment image is evenly divided into several sub-images, it is possible to consider performing different degrees of enhancement processing on each sub-image based on the regional information; and in an implementation method in which the regional environment image is unevenly divided into several sub-images, it is possible to consider performing the same degree of enhancement processing on each sub-image based on the regional information.

[0116] The purpose of the enhancement process in this embodiment is to optimize the image details, improve the image clarity and highlight the characteristic information of the target area, so as to facilitate better and more accurate recognition in step S104. In addition, noise filtering, sharpening and image contrast adjustment can also be performed.

[0117] In some preferred embodiments, Figure 2As shown, step S102 acquires the flow field data of the target area in real time, and obtains the predicted movement trajectory of the object to be identified through fluid dynamics simulation, including steps S201 to S205. Each step is described in detail as follows:

[0118] Step S201, acquiring flow field data of the target area through an underwater sensor network, and determining an initial trajectory based on the flow field data; wherein the flow field data includes flow velocity information and flow direction information of the flow field;

[0119] Step S202: Analyze the flow velocity information and flow direction information in real time to obtain the disturbance frequency and disturbance amplitude of the flow field;

[0120] Step S203, decomposing the disturbance frequency and disturbance amplitude of the flow field by Fourier transform to determine the disturbance characteristic frequency and amplitude fluctuation range;

[0121] Step S204, determining trajectory adjustment parameters according to the disturbance characteristic frequency and amplitude fluctuation range;

[0122] Step S205 , performing fluid mechanics simulation based on the Navier-Stokes equations and constructing a fluid model according to the flow field data; based on the fluid model, using the trajectory adjustment parameters, predicting the next moment of the initial trajectory to obtain a predicted movement trajectory of the object to be identified.

[0123] For example, an underwater sensor network can collect flow field data for a target area once per second, for example, if the flow velocity in the area is 0.5 m / s and the flow direction is 45 degrees northeast. This flow field data can then be uploaded to a cloud server in real time, where it can be processed, analyzed, or calculated using the computing power of the cloud server.

[0124] Preferably, assuming that the parameters of the initial trajectory are a speed of 5.2 m / s and a direction angle of 30 degrees, these parameters are input into the system through the built-in trajectory planning algorithm, and the expected initial trajectory is calculated using the quadratic Bezier curve algorithm.

[0125] For example, some existing path planning formulas:

[0126] B(t)= (1-t) 2 ×P0 + 2×(1-t) ×t×P1 + t 2 ×P2; where P0 is the starting point coordinate, P1 is the control point between the starting point and the end point, P2 is the end point coordinate, and t is the time parameter (or time coefficient) and can vary from 0 to 1, so that the initial trajectory can be obtained.

[0127] Furthermore, the disturbance frequency and amplitude of the flow field were decomposed through Fourier transform, and it was determined that the characteristic frequency of the disturbance was mainly concentrated at 0.1 Hz, and the amplitude fluctuation range was 0.2 m / s, indicating that the impact of the ocean current on the target trajectory has a periodic characteristic. It can be further inferred that the offset range is ±1.5 m in the horizontal direction and ±0.8 m in the vertical direction (the above values ​​are for illustration only).

[0128] A fluid model is constructed based on the flow field data by using fluid mechanics and Navier-Stokes equations and numerical simulation using finite element analysis software.

[0129] By using the simulated fluid model and the calculated trajectory adjustment parameters, the next moment of the initial trajectory can be predicted and updated to obtain the predicted movement trajectory.

[0130] Step S103 : performing comparative analysis on each enhanced sub-image and the underwater image to be identified to obtain a preliminary comparison result.

[0131] In some preferred embodiments, Figure 3 As shown, the comparison analysis between each enhanced sub-image and the underwater image to be identified is performed to obtain a preliminary comparison result, including steps S301 to S304; each step is described in detail as follows:

[0132] Step S301, segmenting the underwater image to be identified based on the enhanced sub-images and the underwater image to be identified, separating the target signal and background noise;

[0133] Step S302: performing feature recognition on the target signal to obtain a signal feature set;

[0134] Step S303: using a dynamic mapping framework to update the signal feature set in real time to obtain a target feature description; wherein the dynamic mapping framework is used to describe the association mapping between environmental parameters and signal features;

[0135] Step S304: obtaining the preliminary comparison result based on the target feature description.

[0136] In this embodiment, by comparing and analyzing each enhanced sub-image with the underwater image to be identified, the target signal and background noise can be roughly determined and separated. Furthermore, the target signal is feature identified and extracted to obtain a signal feature set. Then, a dynamic mapping framework is used to simulate environmental parameters in real time, and the correlation mapping between environmental parameters and signal features is used in real time to obtain the latest target feature description, thereby obtaining a real-time updated preliminary comparison result with high image quality.

[0137] Furthermore, the segmenting of the underwater image to be identified to separate the target signal and background noise includes:

[0138] Constructing a gray-level co-occurrence matrix of the underwater image to be identified, wherein the gray-level co-occurrence matrix is ​​used to reflect the texture features of the underwater image to be identified;

[0139] Using independent component analysis, the gray-level co-occurrence matrix is ​​decomposed into multiple signal components, and the decomposition results are obtained. For example, the signal components are A1, A2 and A3, and the energy proportions can be 70%, 20% and 10% respectively);

[0140] Reconstructing the texture features of the underwater image to be identified according to the energy proportion of each signal component in the decomposition result to obtain a reconstruction result;

[0141] Determine the signal strength of each pixel in the reconstruction result; filter out pixels with signal strength greater than a first preset intensity threshold from the reconstruction result to obtain a valid pixel combination (for example, 2500 valid pixels, accounting for 25%, are filtered out from a 100x100 image, thereby effectively eliminating low-intensity noise);

[0142] A target signal is determined according to the effective pixel combination, and the remaining pixel points of the underwater image to be identified are determined as the background noise.

[0143] In some preferred schemes, before using the dynamic mapping framework to update the signal feature set in real time in step S303, it also includes: obtaining a sample image, performing feature recognition on the sample image to obtain sample features; and obtaining environmental parameters of the area corresponding to the sample image; wherein the environmental parameters include suspended particle information and illumination change information; based on the signal intensity of the sample image, screening out significant features from the sample features (for example, the signal intensity greater than the second preset intensity threshold is a significant feature); and using a data table to construct a dynamic mapping relationship between the significant features and the suspended particle information and illumination change information; based on the dynamic mapping relationship, obtaining the dynamic mapping framework.

[0144] Furthermore, the suspended particle information includes water turbidity (in NTU).

[0145] The method of using a data table to construct a dynamic mapping relationship between the significant features and the suspended particle information and the illumination change information includes:

[0146] Based on the water turbidity and light change information, a linear regression model was constructed;

[0147] Based on the linear regression model and the image contrast, a correlation coefficient is obtained (for example, a positive correlation may be obtained, with a correlation coefficient value of 0.85, and a correlation coefficient value is generally between 0 and 1);

[0148] Using the data table to record the model parameters of the linear regression model, the image contrast and the correlation coefficient, to obtain a recorded result;

[0149] The recording result is recorded as the dynamic mapping relationship.

[0150] In some further preferred embodiments, when constructing a dynamic mapping relationship, factors such as light intensity (in lux), image signal-to-noise ratio, and grayscale information when the image is converted into a grayscale image can be further considered to further improve the accuracy and refinement of the dynamic mapping relationship.

[0151] Step S104, performing a comprehensive analysis based on the predicted movement trajectory, the preset fluid mechanics conditions and the static water ripple feature map to obtain a water ripple category label; using the water ripple category label and the predicted movement trajectory, correcting the preliminary comparison result to obtain a target recognition result of the object to be identified.

[0152] In this embodiment, a comprehensive analysis is performed based on the predicted movement trajectory, the preset fluid dynamics conditions and the static water ripple feature map to obtain a water ripple category label, including: obtaining a water ripple dynamic video within the same time period based on the time information of the flow field data of the target area, and splitting the water ripple dynamic video into a water ripple image sequence; setting feature conditions based on feature similarity based on the static water ripple feature map, and screening out a number of water ripple images from the water ripple image sequence according to the feature conditions to obtain a screening result; and deducing based on the screening result, the predicted movement trajectory and the preset fluid dynamics conditions to obtain a number of water ripple category labels.

[0153] This preferred embodiment is based on the characteristic conditions of the static water ripple feature map and the feature similarity setting, and selects a number of water ripple images from the water ripple image sequence obtained by splitting the water ripple dynamic video to obtain the screening results. It can ensure that the screening results that meet the water ripple feature requirements are selected, ensure the quality of the screened images, and then ensure the accurate setting of the water ripple category labels in the deduction of subsequent steps.

[0154] Furthermore, the water ripple image of the screening result includes interferences; the deduction based on the screening result, the predicted movement trajectory and the preset fluid mechanics conditions to obtain several water ripple category labels includes: extracting the contour points of each interference in the water ripple image of the screening result; taking the contour points of the interference as a reference, cutting off the water ripple lines in each water ripple image containing interferences in the screening result, and extracting the feature points of the water ripple lines, connecting the feature points to obtain the water ripple motion trajectory; based on the water ripple motion trajectory, calling a pre-trained water ripple prediction model to obtain a water ripple prediction trajectory; according to the water ripple prediction trajectory and the predicted movement trajectory, deduction is performed on the basis of the preset fluid mechanics conditions to obtain several water ripple motion patterns, and the water ripple motion patterns are determined as water ripple category labels.

[0155] This preferred embodiment can use interference objects to analyze the motion trajectory of water ripples. Specifically, based on the contour points of the interference objects, the water ripple lines in each water ripple image containing the interference objects in the screening results are cut off and the feature points are extracted. The water ripple motion trajectory is obtained by connecting them, and then the water ripple prediction trajectory is obtained. Subsequently, several water ripple motion patterns can be deduced, clustered and / or classified to provide a reference for determining the reference trajectory in the subsequent steps, so as to accurately obtain the deviation value between the moving trajectory and the reference trajectory.

[0156] In some preferred embodiments, the use of the watermark category label and the predicted movement trajectory to correct the preliminary comparison result to obtain the target recognition result of the object to be identified includes:

[0157] Obtaining a plurality of distinguishing pixel features of each enhanced sub-image and the underwater image to be identified from the preliminary comparison result; each distinguishing pixel feature includes a plurality of pixel points;

[0158] Determining a reference trajectory corresponding to the water ripple motion pattern according to the water ripple category label; analyzing the predicted motion trajectory to obtain a deviation value between the predicted motion trajectory and the reference trajectory;

[0159] Filtering out trajectory points whose deviation values ​​are greater than a preset deviation threshold from the predicted movement trajectory;

[0160] According to the coincidence of the trajectory points and the distinguishing pixel features, the preliminary comparison result is corrected to obtain the target recognition result.

[0161] This embodiment uses one or more of the above-described embodiments to identify underwater images, obtain one or more target objects, and generate target recognition results. These target objects can be underwater fish or plants, for example. This embodiment takes into account factors such as disturbances in the underwater flow field and further factors in the magnitude of the disturbance using a deviation threshold. Compared to existing solutions that consider methods other than water bodies, this embodiment effectively improves target recognition accuracy.

[0162] Accordingly, if Figure 4 As shown, the present invention also provides an underwater target recognition system 400, including an image acquisition module 401, a region information processing module 402, a comparison module 403 and a target recognition module 404; wherein,

[0163] The image acquisition module 401 is used to acquire in real time an underwater image of the target area to be identified and an image of the regional environment; wherein the regional environment image is collected by a multispectral camera and a laser radar, and the multispectral camera uses a wavelength band including visible light, infrared spectrum and ultraviolet spectrum;

[0164] The regional information processing module 402 is configured to identify the regional environment image and obtain regional information; segment the regional environment image into a plurality of sub-images and perform enhancement processing on each sub-image based on the regional information; acquire flow field data of the target area in real time and obtain a predicted movement trajectory of the object to be identified through fluid dynamics simulation; and extract static water ripple features based on the regional environment image using a target detection algorithm to generate a static water ripple feature map.

[0165] The comparison module 403 is configured to perform comparison analysis between each enhanced sub-image and the underwater image to be identified to obtain a preliminary comparison result;

[0166] The target recognition module 404 is used to perform a comprehensive analysis based on the predicted movement trajectory, the preset fluid dynamics conditions and the static water ripple feature map to obtain a water ripple category label; and use the water ripple category label and the predicted movement trajectory to correct the preliminary comparison result to obtain a target recognition result of the object to be identified.

[0167] As a preferred solution, the target recognition module 404 performs comparative analysis on each enhanced sub-image and the underwater image to be recognized to obtain preliminary comparison results, including:

[0168] The target recognition module 404 segments the underwater image to be recognized based on the enhanced sub-images and the underwater image to be recognized, and separates the target signal and background noise;

[0169] Performing feature recognition on the target signal to obtain a signal feature set;

[0170] Using a dynamic mapping framework, the signal feature set is updated in real time to obtain a target feature description; wherein the dynamic mapping framework is used to describe the association mapping between environmental parameters and signal features;

[0171] Based on the target feature description, the preliminary comparison result is obtained.

[0172] As a preferred solution, the target recognition module 404 segments the underwater image to be recognized to separate the target signal and background noise, including:

[0173] The target recognition module 404 constructs a gray level co-occurrence matrix of the underwater image to be recognized, where the gray level co-occurrence matrix is ​​used to reflect the texture features of the underwater image to be recognized;

[0174] Decomposing the gray-level co-occurrence matrix into multiple signal components using independent component analysis to obtain a decomposition result;

[0175] Reconstructing the texture features of the underwater image to be identified according to the energy proportion of each signal component in the decomposition result to obtain a reconstruction result;

[0176] Determining the signal strength of each pixel in the reconstruction result; filtering out pixels having a signal strength greater than a first preset strength threshold from the reconstruction result to obtain a valid pixel combination;

[0177] A target signal is determined according to the effective pixel combination, and the remaining pixel points of the underwater image to be identified are determined as the background noise.

[0178] As a preferred solution, the underwater target recognition system 400 further includes a mapping framework construction module, which is configured to:

[0179] Acquire a sample image, perform feature recognition on the sample image to obtain sample features; and acquire environmental parameters of an area corresponding to the sample image; wherein the environmental parameters include suspended particle information and illumination change information;

[0180] Screening out significant features from the sample features based on the signal intensity of the sample image; and constructing a dynamic mapping relationship between the significant features and the suspended particle information and the illumination change information using a data table;

[0181] The dynamic mapping framework is obtained according to the dynamic mapping relationship.

[0182] As a preferred solution, the salient features include image contrast, and the suspended particle information includes water turbidity; the mapping framework construction module uses a data table to construct a dynamic mapping relationship between the salient features and the suspended particle information and illumination change information, including:

[0183] The mapping framework construction module constructs a linear regression model based on water turbidity and light change information;

[0184] Obtaining a correlation coefficient based on the linear regression model and the image contrast;

[0185] Using the data table to record the model parameters of the linear regression model, the image contrast and the correlation coefficient, to obtain a recorded result;

[0186] The recording result is recorded as the dynamic mapping relationship.

[0187] As a preferred solution, the target recognition module performs a comprehensive analysis based on the predicted movement trajectory, the preset fluid dynamics conditions and the static water ripple feature map to obtain a water ripple category label, including:

[0188] According to the time information of the flow field data of the target area, a water ripple dynamic video within the same time period is obtained, and the water ripple dynamic video is split into a water ripple image sequence;

[0189] Setting a feature condition based on feature similarity according to the static watermark feature map, and screening a plurality of watermark images from the watermark image sequence according to the feature condition to obtain a screening result;

[0190] Deduction is performed based on the screening results, the predicted movement trajectory, and preset fluid mechanics conditions to obtain several water pattern category labels.

[0191] As a preferred solution, the watermark image of the screening result includes interference objects; the target recognition module deduces based on the screening result, the predicted movement trajectory and the preset fluid dynamics conditions to obtain several watermark category labels, including:

[0192] The target recognition module extracts the contour points of each interference object in the watermark image of the screening result;

[0193] Taking the contour points of the interference object as a reference, truncating the water ripple lines in each water ripple image containing the interference object in the screening results, extracting the feature points of the water ripple lines, and connecting the feature points to obtain the water ripple motion trajectory;

[0194] Based on the water ripple motion trajectory, calling a pre-trained water ripple prediction model to obtain a water ripple prediction trajectory;

[0195] According to the water ripple predicted trajectory and the predicted movement trajectory, deduction is performed on the basis of preset fluid dynamics conditions to obtain a number of water ripple motion patterns, and the water ripple motion patterns are determined as water ripple category labels.

[0196] As a preferred solution, the region information processing module 402 acquires the flow field data of the target region in real time and obtains the predicted movement trajectory of the object to be identified through fluid dynamics simulation, including:

[0197] The regional information processing module 402 obtains flow field data of the target area through the underwater sensor network and determines the initial trajectory according to the flow field data; wherein the flow field data includes flow velocity information and flow direction information of the flow field;

[0198] Performing real-time analysis on the flow velocity information and flow direction information to obtain the disturbance frequency and disturbance amplitude of the flow field;

[0199] The disturbance frequency and disturbance amplitude of the flow field are decomposed by Fourier transform to determine the characteristic frequency and amplitude fluctuation range of the disturbance;

[0200] Determining trajectory adjustment parameters according to the disturbance characteristic frequency and amplitude fluctuation range;

[0201] A fluid mechanics simulation is performed based on the Navier-Stokes equations, and a fluid model is constructed according to the flow field data. Based on the fluid model, the trajectory adjustment parameters are used to predict the next moment of the initial trajectory to obtain the predicted movement trajectory of the object to be identified.

[0202] As a preferred solution, the target recognition module 404 uses the watermark category label and the predicted movement trajectory to correct the preliminary comparison result to obtain the target recognition result of the object to be recognized, including:

[0203] The target recognition module 404 obtains a plurality of distinguishing pixel features of each enhanced sub-image and the underwater image to be recognized from the preliminary comparison result; each distinguishing pixel feature includes a plurality of pixel points;

[0204] Determining a reference trajectory corresponding to the water ripple motion pattern according to the water ripple category label; analyzing the predicted motion trajectory to obtain a deviation value between the predicted motion trajectory and the reference trajectory;

[0205] Filtering out trajectory points whose deviation values ​​are greater than a preset deviation threshold from the predicted movement trajectory;

[0206] According to the coincidence of the trajectory points and the distinguishing pixel features, the preliminary comparison result is corrected to obtain the target recognition result.

[0207] Compared with the prior art, the present invention has the following beneficial effects:

[0208] The present invention provides an underwater target recognition method and system, the method comprising: acquiring in real time an underwater image to be recognized and an area environment image of a target area; wherein the area environment image is collected by a multispectral camera and a laser radar, and the multispectral camera uses bands including visible light, infrared spectrum and ultraviolet spectrum; recognizing the area environment image to obtain area information; dividing the area environment image into a plurality of sub-images, and enhancing each sub-image according to the area information; acquiring flow field data of the target area in real time, and obtaining a predicted movement trajectory of the object to be recognized through fluid mechanics simulation; using a target detection algorithm, extracting static water ripple features based on the area environment image to generate a static water ripple feature map; performing comparative analysis between each enhanced sub-image and the underwater image to be recognized to obtain a preliminary comparison result; performing a comprehensive analysis based on the predicted movement trajectory, preset fluid mechanics conditions and the static water ripple feature map to obtain a water ripple category label; and correcting the preliminary comparison result using the water ripple category label and the predicted movement trajectory to obtain a target recognition result of the object to be recognized. Compared with the prior art, the present invention application considers technical solutions other than water bodies. By obtaining the flow field data of the target area and using fluid mechanics simulation to predict the movement trajectory and then correct the comparison result, the disturbance factors of the underwater flow field are taken into account, thereby effectively improving the accuracy of target recognition. In addition, the regional environmental image is divided into several sub-images, and each sub-image is enhanced to different degrees or the same degree according to the regional information. This can improve the quality of the regional environmental image, thereby improving the accuracy of the comparison analysis, obtaining higher quality preliminary comparison results, and further improving the accuracy of the target recognition results. Furthermore, in addition to considering the predicted movement trajectory and the preset fluid mechanics conditions, the static water ripple characteristics are also taken into account, that is, the relatively microscopic and relatively macroscopic movements are integrated to construct water ripple category labels for correcting the preliminary comparison results, further improving the accuracy of the target recognition results.

[0209] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for underwater target recognition, characterized in that: include: Acquire in real time an underwater image of the target area to be identified and an image of the regional environment; wherein the regional environment image is collected by a multispectral camera and a laser radar, and the multispectral camera uses bands including visible light, infrared spectrum, and ultraviolet spectrum; Identify the regional environment image to obtain regional information; segment the regional environment image into a plurality of sub-images, and enhance each sub-image according to the regional information; acquire flow field data of the target area in real time, and obtain a predicted movement trajectory of the object to be identified through fluid dynamics simulation; and extract static water ripple features based on the regional environment image using a target detection algorithm to generate a static water ripple feature map; Performing comparative analysis on each enhanced sub-image and the underwater image to be identified to obtain a preliminary comparison result; Performing a comprehensive analysis based on the predicted movement trajectory, the preset fluid dynamics conditions, and the static water pattern characteristic map to obtain a water pattern category label; using the water pattern category label and the predicted movement trajectory, revising the preliminary comparison result to obtain a target recognition result of the object to be identified; The water ripple category label is obtained by performing a comprehensive analysis based on the predicted movement trajectory, the preset fluid dynamics conditions, and the static water ripple feature map, including: According to the time information of the flow field data of the target area, a water ripple dynamic video within the same time period is obtained, and the water ripple dynamic video is split into a water ripple image sequence; Setting a feature condition based on feature similarity according to the static watermark feature map, and screening a plurality of watermark images from the watermark image sequence according to the feature condition to obtain a screening result; Deducing based on the screening results, the predicted movement trajectory, and preset fluid dynamics conditions to obtain several water pattern category labels; The water ripple image of the screening result includes interference objects; the deduction based on the screening result, the predicted movement trajectory and the preset fluid mechanics conditions is performed to obtain several water ripple category labels, including: Extract the contour points of each interference object in the watermark image of the screening result; Taking the contour points of the interference object as a reference, truncating the water ripple lines in each water ripple image containing the interference object in the screening results, extracting the feature points of the water ripple lines, and connecting the feature points to obtain the water ripple motion trajectory; Based on the water ripple motion trajectory, calling a pre-trained water ripple prediction model to obtain a water ripple prediction trajectory; According to the water ripple predicted trajectory and the predicted movement trajectory, deduction is performed on the basis of preset fluid dynamics conditions to obtain a number of water ripple motion patterns, and the water ripple motion patterns are determined as water ripple category labels.

2. The underwater target recognition method according to claim 1, wherein: The comparative analysis of each enhanced sub-image and the underwater image to be identified to obtain a preliminary comparison result includes: Segmenting the underwater image to be identified based on the enhanced sub-images and the underwater image to be identified, separating the target signal and background noise; Performing feature recognition on the target signal to obtain a signal feature set; Using a dynamic mapping framework, the signal feature set is updated in real time to obtain a target feature description; wherein the dynamic mapping framework is used to describe the association mapping between environmental parameters and signal features; Based on the target feature description, the preliminary comparison result is obtained.

3. The underwater target recognition method according to claim 2, wherein: Before the signal feature set is updated in real time using the dynamic mapping framework, the method further includes: Acquire a sample image, perform feature recognition on the sample image to obtain sample features; and acquire environmental parameters of an area corresponding to the sample image; wherein the environmental parameters include suspended particle information and illumination change information; Screening out significant features from the sample features based on the signal intensity of the sample image; and constructing a dynamic mapping relationship between the significant features and the suspended particle information and the illumination change information using a data table; The dynamic mapping framework is obtained according to the dynamic mapping relationship.

4. The underwater target recognition method according to claim 3, wherein: The significant features include image contrast, and the suspended particle information includes water turbidity; and using a data table to construct a dynamic mapping relationship between the significant features and the suspended particle information and illumination change information includes: Based on the water turbidity and light change information, a linear regression model was constructed; Obtaining a correlation coefficient based on the linear regression model and the image contrast; Using the data table to record the model parameters of the linear regression model, the image contrast and the correlation coefficient, to obtain a recorded result; The recording result is recorded as the dynamic mapping relationship.

5. The underwater target recognition method according to claim 1, wherein: The real-time acquisition of flow field data of the target area and the obtained predicted movement trajectory of the object to be identified through fluid dynamics simulation include: Acquiring flow field data of the target area through an underwater sensor network, and determining an initial trajectory based on the flow field data; wherein the flow field data includes flow velocity information and flow direction information of the flow field; Performing real-time analysis on the flow velocity information and flow direction information to obtain the disturbance frequency and disturbance amplitude of the flow field; The disturbance frequency and disturbance amplitude of the flow field are decomposed by Fourier transform to determine the characteristic frequency and amplitude fluctuation range of the disturbance; Determining trajectory adjustment parameters according to the disturbance characteristic frequency and amplitude fluctuation range; A fluid mechanics simulation is performed based on the Navier-Stokes equations, and a fluid model is constructed according to the flow field data. Based on the fluid model, the trajectory adjustment parameters are used to predict the next moment of the initial trajectory to obtain the predicted movement trajectory of the object to be identified.

6. The underwater target recognition method according to claim 1, wherein: The method of modifying the preliminary comparison result by using the watermark category label and the predicted movement trajectory to obtain the target recognition result of the object to be recognized includes: Obtaining a plurality of distinguishing pixel features of each enhanced sub-image and the underwater image to be identified from the preliminary comparison result; each distinguishing pixel feature includes a plurality of pixel points; Determining a reference trajectory corresponding to the water ripple motion pattern according to the water ripple category label; analyzing the predicted motion trajectory to obtain a deviation value between the predicted motion trajectory and the reference trajectory; Filtering out trajectory points whose deviation values ​​are greater than a preset deviation threshold from the predicted movement trajectory; According to the coincidence of the trajectory points and the distinguishing pixel features, the preliminary comparison result is corrected to obtain the target recognition result.

7. An underwater target recognition system, characterized in that: It includes image acquisition module, regional information processing module, comparison module and target recognition module; among them, The image acquisition module is used to acquire in real time the underwater image to be identified and the regional environment image of the target area; wherein the regional environment image is collected by a multispectral camera and a laser radar, and the bands used by the multispectral camera include visible light, infrared spectrum and ultraviolet spectrum; The regional information processing module is used to identify the regional environment image and obtain regional information; segment the regional environment image into a plurality of sub-images and perform enhancement processing on each sub-image according to the regional information; obtain flow field data of the target area in real time and obtain the predicted movement trajectory of the object to be identified through fluid dynamics simulation; and use the target detection algorithm to extract static water ripple features based on the regional environment image to generate a static water ripple feature map; The comparison module is used to perform comparison analysis based on each enhanced sub-image and the underwater image to be identified to obtain a preliminary comparison result; The target recognition module is configured to perform a comprehensive analysis based on the predicted movement trajectory, the preset fluid dynamics conditions, and the static water pattern characteristic map to obtain a water pattern category label; and to modify the preliminary comparison result using the water pattern category label and the predicted movement trajectory to obtain a target recognition result for the object to be identified; The target recognition module performs a comprehensive analysis based on the predicted movement trajectory, the preset fluid dynamics conditions, and the static water ripple feature map to obtain a water ripple category label, including: The target recognition module obtains a water ripple dynamic video within the same time period according to the time information of the flow field data of the target area, and splits the water ripple dynamic video into a water ripple image sequence; Setting a feature condition based on feature similarity according to the static watermark feature map, and screening a plurality of watermark images from the watermark image sequence according to the feature condition to obtain a screening result; Deducing based on the screening results, the predicted movement trajectory, and preset fluid dynamics conditions to obtain several water pattern category labels; The water ripple image of the screening result includes interference objects; the target recognition module deduces based on the screening result, the predicted movement trajectory and the preset fluid dynamics conditions to obtain several water ripple category labels, including: The target recognition module extracts the contour points of each interference object in the watermark image of the screening result; Taking the contour points of the interference object as a reference, truncating the water ripple lines in each water ripple image containing the interference object in the screening results, extracting the feature points of the water ripple lines, and connecting the feature points to obtain the water ripple motion trajectory; Based on the water ripple motion trajectory, calling a pre-trained water ripple prediction model to obtain a water ripple prediction trajectory; According to the water ripple predicted trajectory and the predicted movement trajectory, deduction is performed on the basis of preset fluid dynamics conditions to obtain a number of water ripple motion patterns, and the water ripple motion patterns are determined as water ripple category labels.

8. An underwater target recognition system according to claim 7, characterized in that: The target recognition module performs comparative analysis on each enhanced sub-image and the underwater image to be recognized to obtain preliminary comparison results, including: The target recognition module segments the underwater image to be identified based on the enhanced sub-images and the underwater image to be identified, and separates the target signal and background noise; Performing feature recognition on the target signal to obtain a signal feature set; Using a dynamic mapping framework, the signal feature set is updated in real time to obtain a target feature description; wherein the dynamic mapping framework is used to describe the association mapping between environmental parameters and signal features; Based on the target feature description, the preliminary comparison result is obtained.

Images